/src/speex/libspeex/nb_celp.c
Line | Count | Source (jump to first uncovered line) |
1 | | /* Copyright (C) 2002-2006 Jean-Marc Valin |
2 | | File: nb_celp.c |
3 | | |
4 | | Redistribution and use in source and binary forms, with or without |
5 | | modification, are permitted provided that the following conditions |
6 | | are met: |
7 | | |
8 | | - Redistributions of source code must retain the above copyright |
9 | | notice, this list of conditions and the following disclaimer. |
10 | | |
11 | | - Redistributions in binary form must reproduce the above copyright |
12 | | notice, this list of conditions and the following disclaimer in the |
13 | | documentation and/or other materials provided with the distribution. |
14 | | |
15 | | - Neither the name of the Xiph.org Foundation nor the names of its |
16 | | contributors may be used to endorse or promote products derived from |
17 | | this software without specific prior written permission. |
18 | | |
19 | | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
20 | | ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
21 | | LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
22 | | A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR |
23 | | CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
24 | | EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
25 | | PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
26 | | PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF |
27 | | LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING |
28 | | NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
29 | | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
30 | | */ |
31 | | |
32 | | #ifdef HAVE_CONFIG_H |
33 | | #include "config.h" |
34 | | #endif |
35 | | |
36 | | #include <math.h> |
37 | | #include "nb_celp.h" |
38 | | #include "lpc.h" |
39 | | #include "lsp.h" |
40 | | #include "ltp.h" |
41 | | #include "quant_lsp.h" |
42 | | #include "cb_search.h" |
43 | | #include "filters.h" |
44 | | #include "stack_alloc.h" |
45 | | #include "vq.h" |
46 | | #include "vbr.h" |
47 | | #include "arch.h" |
48 | | #include "math_approx.h" |
49 | | #include "os_support.h" |
50 | | |
51 | | #ifdef VORBIS_PSYCHO |
52 | | #include "vorbis_psy.h" |
53 | | #endif |
54 | | |
55 | | #ifndef NULL |
56 | | #define NULL 0 |
57 | | #endif |
58 | | |
59 | 2.29M | #define SUBMODE(x) st->submodes[st->submodeID]->x |
60 | | |
61 | | /* Default size for the encoder and decoder stack (can be changed at compile time). |
62 | | This does not apply when using variable-size arrays or alloca. */ |
63 | | #ifndef NB_ENC_STACK |
64 | | #define NB_ENC_STACK (8000*sizeof(spx_sig_t)) |
65 | | #endif |
66 | | |
67 | | #ifndef NB_DEC_STACK |
68 | | #define NB_DEC_STACK (4000*sizeof(spx_sig_t)) |
69 | | #endif |
70 | | |
71 | | |
72 | | #ifdef FIXED_POINT |
73 | | static const spx_word32_t ol_gain_table[32]={18900, 25150, 33468, 44536, 59265, 78865, 104946, 139653, 185838, 247297, 329081, 437913, 582736, 775454, 1031906, 1373169, 1827293, 2431601, 3235761, 4305867, 5729870, 7624808, 10146425, 13501971, 17967238, 23909222, 31816294, 42338330, 56340132, 74972501, 99766822, 132760927}; |
74 | | static const spx_word16_t exc_gain_quant_scal3_bound[7]={1841, 3883, 6051, 8062, 10444, 13580, 18560}; |
75 | | static const spx_word16_t exc_gain_quant_scal3[8]={1002, 2680, 5086, 7016, 9108, 11781, 15380, 21740}; |
76 | | static const spx_word16_t exc_gain_quant_scal1_bound[1]={14385}; |
77 | | static const spx_word16_t exc_gain_quant_scal1[2]={11546, 17224}; |
78 | | |
79 | 219k | #define LSP_MARGIN 16 |
80 | 19.6k | #define LSP_DELTA1 6553 |
81 | | #define LSP_DELTA2 1638 |
82 | | |
83 | | #else |
84 | | |
85 | | static const float exc_gain_quant_scal3_bound[7]={0.112338f, 0.236980f, 0.369316f, 0.492054f, 0.637471f, 0.828874f, 1.132784f}; |
86 | | static const float exc_gain_quant_scal3[8]={0.061130f, 0.163546f, 0.310413f, 0.428220f, 0.555887f, 0.719055f, 0.938694f, 1.326874f}; |
87 | | static const float exc_gain_quant_scal1_bound[1]={0.87798f}; |
88 | | static const float exc_gain_quant_scal1[2]={0.70469f, 1.05127f}; |
89 | | |
90 | 110k | #define LSP_MARGIN .002f |
91 | 8.59k | #define LSP_DELTA1 .2f |
92 | | #define LSP_DELTA2 .05f |
93 | | |
94 | | #endif |
95 | | |
96 | | extern const spx_word16_t lag_window[]; |
97 | | extern const spx_word16_t lpc_window[]; |
98 | | |
99 | | #ifndef DISABLE_ENCODER |
100 | | void *nb_encoder_init(const SpeexMode *m) |
101 | 7.91k | { |
102 | 7.91k | EncState *st; |
103 | 7.91k | const SpeexNBMode *mode; |
104 | 7.91k | int i; |
105 | | |
106 | 7.91k | mode=(const SpeexNBMode *)m->mode; |
107 | 7.91k | st = (EncState*)speex_alloc(sizeof(EncState)); |
108 | 7.91k | if (!st) |
109 | 0 | return NULL; |
110 | 7.91k | #if defined(VAR_ARRAYS) || defined (USE_ALLOCA) |
111 | 7.91k | st->stack = NULL; |
112 | | #else |
113 | | st->stack = (char*)speex_alloc_scratch(NB_ENC_STACK); |
114 | | #endif |
115 | | |
116 | 7.91k | st->mode=m; |
117 | | |
118 | 7.91k | st->gamma1=mode->gamma1; |
119 | 7.91k | st->gamma2=mode->gamma2; |
120 | 7.91k | st->lpc_floor = mode->lpc_floor; |
121 | | |
122 | 7.91k | st->submodes=mode->submodes; |
123 | 7.91k | st->submodeID=st->submodeSelect=mode->defaultSubmode; |
124 | 7.91k | st->bounded_pitch = 1; |
125 | | |
126 | 7.91k | st->encode_submode = 1; |
127 | | |
128 | | #ifdef VORBIS_PSYCHO |
129 | | st->psy = vorbis_psy_init(8000, 256); |
130 | | st->curve = (float*)speex_alloc(128*sizeof(float)); |
131 | | st->old_curve = (float*)speex_alloc(128*sizeof(float)); |
132 | | st->psy_window = (float*)speex_alloc(256*sizeof(float)); |
133 | | #endif |
134 | | |
135 | 7.91k | st->cumul_gain = 1024; |
136 | | |
137 | 7.91k | st->window= lpc_window; |
138 | | |
139 | | /* Create the window for autocorrelation (lag-windowing) */ |
140 | 7.91k | st->lagWindow = lag_window; |
141 | | |
142 | 7.91k | st->first = 1; |
143 | 87.0k | for (i=0;i<NB_ORDER;i++) |
144 | 79.1k | st->old_lsp[i]= DIV32(MULT16_16(QCONST16(3.1415927f, LSP_SHIFT), i+1), NB_ORDER+1); |
145 | | |
146 | 7.91k | st->innov_rms_save = NULL; |
147 | | |
148 | 7.91k | #ifndef DISABLE_VBR |
149 | 7.91k | vbr_init(&st->vbr); |
150 | 7.91k | st->vbr_quality = 8; |
151 | 7.91k | st->vbr_enabled = 0; |
152 | 7.91k | st->vbr_max = 0; |
153 | 7.91k | st->vad_enabled = 0; |
154 | 7.91k | st->dtx_enabled = 0; |
155 | 7.91k | st->dtx_count=0; |
156 | 7.91k | st->abr_enabled = 0; |
157 | 7.91k | st->abr_drift = 0; |
158 | 7.91k | st->abr_drift2 = 0; |
159 | 7.91k | #endif /* #ifndef DISABLE_VBR */ |
160 | | |
161 | 7.91k | st->plc_tuning = 2; |
162 | 7.91k | st->complexity=2; |
163 | 7.91k | st->sampling_rate=8000; |
164 | 7.91k | st->isWideband = 0; |
165 | 7.91k | st->highpass_enabled = 1; |
166 | | |
167 | | #ifdef ENABLE_VALGRIND |
168 | | VALGRIND_MAKE_MEM_DEFINED(st, NB_ENC_STACK); |
169 | | #endif |
170 | 7.91k | return st; |
171 | 7.91k | } Line | Count | Source | 101 | 3.95k | { | 102 | 3.95k | EncState *st; | 103 | 3.95k | const SpeexNBMode *mode; | 104 | 3.95k | int i; | 105 | | | 106 | 3.95k | mode=(const SpeexNBMode *)m->mode; | 107 | 3.95k | st = (EncState*)speex_alloc(sizeof(EncState)); | 108 | 3.95k | if (!st) | 109 | 0 | return NULL; | 110 | 3.95k | #if defined(VAR_ARRAYS) || defined (USE_ALLOCA) | 111 | 3.95k | st->stack = NULL; | 112 | | #else | 113 | | st->stack = (char*)speex_alloc_scratch(NB_ENC_STACK); | 114 | | #endif | 115 | | | 116 | 3.95k | st->mode=m; | 117 | | | 118 | 3.95k | st->gamma1=mode->gamma1; | 119 | 3.95k | st->gamma2=mode->gamma2; | 120 | 3.95k | st->lpc_floor = mode->lpc_floor; | 121 | | | 122 | 3.95k | st->submodes=mode->submodes; | 123 | 3.95k | st->submodeID=st->submodeSelect=mode->defaultSubmode; | 124 | 3.95k | st->bounded_pitch = 1; | 125 | | | 126 | 3.95k | st->encode_submode = 1; | 127 | | | 128 | | #ifdef VORBIS_PSYCHO | 129 | | st->psy = vorbis_psy_init(8000, 256); | 130 | | st->curve = (float*)speex_alloc(128*sizeof(float)); | 131 | | st->old_curve = (float*)speex_alloc(128*sizeof(float)); | 132 | | st->psy_window = (float*)speex_alloc(256*sizeof(float)); | 133 | | #endif | 134 | | | 135 | 3.95k | st->cumul_gain = 1024; | 136 | | | 137 | 3.95k | st->window= lpc_window; | 138 | | | 139 | | /* Create the window for autocorrelation (lag-windowing) */ | 140 | 3.95k | st->lagWindow = lag_window; | 141 | | | 142 | 3.95k | st->first = 1; | 143 | 43.5k | for (i=0;i<NB_ORDER;i++) | 144 | 39.5k | st->old_lsp[i]= DIV32(MULT16_16(QCONST16(3.1415927f, LSP_SHIFT), i+1), NB_ORDER+1); | 145 | | | 146 | 3.95k | st->innov_rms_save = NULL; | 147 | | | 148 | 3.95k | #ifndef DISABLE_VBR | 149 | 3.95k | vbr_init(&st->vbr); | 150 | 3.95k | st->vbr_quality = 8; | 151 | 3.95k | st->vbr_enabled = 0; | 152 | 3.95k | st->vbr_max = 0; | 153 | 3.95k | st->vad_enabled = 0; | 154 | 3.95k | st->dtx_enabled = 0; | 155 | 3.95k | st->dtx_count=0; | 156 | 3.95k | st->abr_enabled = 0; | 157 | 3.95k | st->abr_drift = 0; | 158 | 3.95k | st->abr_drift2 = 0; | 159 | 3.95k | #endif /* #ifndef DISABLE_VBR */ | 160 | | | 161 | 3.95k | st->plc_tuning = 2; | 162 | 3.95k | st->complexity=2; | 163 | 3.95k | st->sampling_rate=8000; | 164 | 3.95k | st->isWideband = 0; | 165 | 3.95k | st->highpass_enabled = 1; | 166 | | | 167 | | #ifdef ENABLE_VALGRIND | 168 | | VALGRIND_MAKE_MEM_DEFINED(st, NB_ENC_STACK); | 169 | | #endif | 170 | 3.95k | return st; | 171 | 3.95k | } |
Line | Count | Source | 101 | 3.95k | { | 102 | 3.95k | EncState *st; | 103 | 3.95k | const SpeexNBMode *mode; | 104 | 3.95k | int i; | 105 | | | 106 | 3.95k | mode=(const SpeexNBMode *)m->mode; | 107 | 3.95k | st = (EncState*)speex_alloc(sizeof(EncState)); | 108 | 3.95k | if (!st) | 109 | 0 | return NULL; | 110 | 3.95k | #if defined(VAR_ARRAYS) || defined (USE_ALLOCA) | 111 | 3.95k | st->stack = NULL; | 112 | | #else | 113 | | st->stack = (char*)speex_alloc_scratch(NB_ENC_STACK); | 114 | | #endif | 115 | | | 116 | 3.95k | st->mode=m; | 117 | | | 118 | 3.95k | st->gamma1=mode->gamma1; | 119 | 3.95k | st->gamma2=mode->gamma2; | 120 | 3.95k | st->lpc_floor = mode->lpc_floor; | 121 | | | 122 | 3.95k | st->submodes=mode->submodes; | 123 | 3.95k | st->submodeID=st->submodeSelect=mode->defaultSubmode; | 124 | 3.95k | st->bounded_pitch = 1; | 125 | | | 126 | 3.95k | st->encode_submode = 1; | 127 | | | 128 | | #ifdef VORBIS_PSYCHO | 129 | | st->psy = vorbis_psy_init(8000, 256); | 130 | | st->curve = (float*)speex_alloc(128*sizeof(float)); | 131 | | st->old_curve = (float*)speex_alloc(128*sizeof(float)); | 132 | | st->psy_window = (float*)speex_alloc(256*sizeof(float)); | 133 | | #endif | 134 | | | 135 | 3.95k | st->cumul_gain = 1024; | 136 | | | 137 | 3.95k | st->window= lpc_window; | 138 | | | 139 | | /* Create the window for autocorrelation (lag-windowing) */ | 140 | 3.95k | st->lagWindow = lag_window; | 141 | | | 142 | 3.95k | st->first = 1; | 143 | 43.5k | for (i=0;i<NB_ORDER;i++) | 144 | 39.5k | st->old_lsp[i]= DIV32(MULT16_16(QCONST16(3.1415927f, LSP_SHIFT), i+1), NB_ORDER+1); | 145 | | | 146 | 3.95k | st->innov_rms_save = NULL; | 147 | | | 148 | 3.95k | #ifndef DISABLE_VBR | 149 | 3.95k | vbr_init(&st->vbr); | 150 | 3.95k | st->vbr_quality = 8; | 151 | 3.95k | st->vbr_enabled = 0; | 152 | 3.95k | st->vbr_max = 0; | 153 | 3.95k | st->vad_enabled = 0; | 154 | 3.95k | st->dtx_enabled = 0; | 155 | 3.95k | st->dtx_count=0; | 156 | 3.95k | st->abr_enabled = 0; | 157 | 3.95k | st->abr_drift = 0; | 158 | 3.95k | st->abr_drift2 = 0; | 159 | 3.95k | #endif /* #ifndef DISABLE_VBR */ | 160 | | | 161 | 3.95k | st->plc_tuning = 2; | 162 | 3.95k | st->complexity=2; | 163 | 3.95k | st->sampling_rate=8000; | 164 | 3.95k | st->isWideband = 0; | 165 | 3.95k | st->highpass_enabled = 1; | 166 | | | 167 | | #ifdef ENABLE_VALGRIND | 168 | | VALGRIND_MAKE_MEM_DEFINED(st, NB_ENC_STACK); | 169 | | #endif | 170 | 3.95k | return st; | 171 | 3.95k | } |
|
172 | | |
173 | | void nb_encoder_destroy(void *state) |
174 | 3.95k | { |
175 | 3.95k | EncState *st=(EncState *)state; |
176 | | /* Free all allocated memory */ |
177 | | #if !(defined(VAR_ARRAYS) || defined (USE_ALLOCA)) |
178 | | speex_free_scratch(st->stack); |
179 | | #endif |
180 | | |
181 | 3.95k | #ifndef DISABLE_VBR |
182 | 3.95k | vbr_destroy(&st->vbr); |
183 | 3.95k | #endif /* #ifndef DISABLE_VBR */ |
184 | | |
185 | | #ifdef VORBIS_PSYCHO |
186 | | vorbis_psy_destroy(st->psy); |
187 | | speex_free (st->curve); |
188 | | speex_free (st->old_curve); |
189 | | speex_free (st->psy_window); |
190 | | #endif |
191 | | |
192 | | /*Free state memory... should be last*/ |
193 | 3.95k | speex_free(st); |
194 | 3.95k | } |
195 | | |
196 | | |
197 | | int nb_encoder_ctl(void *state, int request, void *ptr) |
198 | 620k | { |
199 | 620k | EncState *st; |
200 | 620k | st=(EncState*)state; |
201 | 620k | switch(request) |
202 | 620k | { |
203 | 15.7k | case SPEEX_GET_FRAME_SIZE: |
204 | 15.7k | (*(spx_int32_t*)ptr) = NB_FRAME_SIZE; |
205 | 15.7k | break; |
206 | 0 | case SPEEX_SET_LOW_MODE: |
207 | 82.6k | case SPEEX_SET_MODE: |
208 | 82.6k | st->submodeSelect = st->submodeID = (*(spx_int32_t*)ptr); |
209 | 82.6k | break; |
210 | 21.0k | case SPEEX_GET_LOW_MODE: |
211 | 21.0k | case SPEEX_GET_MODE: |
212 | 21.0k | (*(spx_int32_t*)ptr) = st->submodeID; |
213 | 21.0k | break; |
214 | 0 | #ifndef DISABLE_VBR |
215 | 4.85k | case SPEEX_SET_VBR: |
216 | 4.85k | st->vbr_enabled = (*(spx_int32_t*)ptr); |
217 | 4.85k | break; |
218 | 0 | case SPEEX_GET_VBR: |
219 | 0 | (*(spx_int32_t*)ptr) = st->vbr_enabled; |
220 | 0 | break; |
221 | 3.57k | case SPEEX_SET_VAD: |
222 | 3.57k | st->vad_enabled = (*(spx_int32_t*)ptr); |
223 | 3.57k | break; |
224 | 0 | case SPEEX_GET_VAD: |
225 | 0 | (*(spx_int32_t*)ptr) = st->vad_enabled; |
226 | 0 | break; |
227 | 3.42k | case SPEEX_SET_DTX: |
228 | 3.42k | st->dtx_enabled = (*(spx_int32_t*)ptr); |
229 | 3.42k | break; |
230 | 0 | case SPEEX_GET_DTX: |
231 | 0 | (*(spx_int32_t*)ptr) = st->dtx_enabled; |
232 | 0 | break; |
233 | 1.08k | case SPEEX_SET_ABR: |
234 | 1.08k | st->abr_enabled = (*(spx_int32_t*)ptr); |
235 | 1.08k | st->vbr_enabled = st->abr_enabled!=0; |
236 | 1.08k | if (st->vbr_enabled) |
237 | 1.08k | { |
238 | 1.08k | spx_int32_t i=10; |
239 | 1.08k | spx_int32_t rate, target; |
240 | 1.08k | float vbr_qual; |
241 | 1.08k | target = (*(spx_int32_t*)ptr); |
242 | 13.0k | while (i>=0) |
243 | 11.9k | { |
244 | 11.9k | speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i); |
245 | 11.9k | speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate); |
246 | 11.9k | if (rate <= target) |
247 | 0 | break; |
248 | 11.9k | i--; |
249 | 11.9k | } |
250 | 1.08k | vbr_qual=i; |
251 | 1.08k | if (vbr_qual<0) |
252 | 1.08k | vbr_qual=0; |
253 | 1.08k | speex_encoder_ctl(st, SPEEX_SET_VBR_QUALITY, &vbr_qual); |
254 | 1.08k | st->abr_count=0; |
255 | 1.08k | st->abr_drift=0; |
256 | 1.08k | st->abr_drift2=0; |
257 | 1.08k | } |
258 | | |
259 | 1.08k | break; |
260 | 0 | case SPEEX_GET_ABR: |
261 | 0 | (*(spx_int32_t*)ptr) = st->abr_enabled; |
262 | 0 | break; |
263 | 0 | #endif /* #ifndef DISABLE_VBR */ |
264 | 0 | #if !defined(DISABLE_VBR) && !defined(DISABLE_FLOAT_API) |
265 | 5.93k | case SPEEX_SET_VBR_QUALITY: |
266 | 5.93k | st->vbr_quality = (*(float*)ptr); |
267 | 5.93k | if (st->vbr_quality < 0) |
268 | 0 | st->vbr_quality = 0; |
269 | 5.93k | else if (st->vbr_quality > 10) |
270 | 0 | st->vbr_quality = 10; |
271 | 5.93k | break; |
272 | 0 | case SPEEX_GET_VBR_QUALITY: |
273 | 0 | (*(float*)ptr) = st->vbr_quality; |
274 | 0 | break; |
275 | 0 | #endif /* !defined(DISABLE_VBR) && !defined(DISABLE_FLOAT_API) */ |
276 | 144k | case SPEEX_SET_QUALITY: |
277 | 144k | { |
278 | 144k | int quality = (*(spx_int32_t*)ptr); |
279 | 144k | if (quality < 0) |
280 | 0 | quality = 0; |
281 | 144k | if (quality > 10) |
282 | 0 | quality = 10; |
283 | 144k | st->submodeSelect = st->submodeID = ((const SpeexNBMode*)(st->mode->mode))->quality_map[quality]; |
284 | 144k | } |
285 | 144k | break; |
286 | 7.91k | case SPEEX_SET_COMPLEXITY: |
287 | 7.91k | st->complexity = (*(spx_int32_t*)ptr); |
288 | 7.91k | if (st->complexity<0) |
289 | 0 | st->complexity=0; |
290 | 7.91k | break; |
291 | 0 | case SPEEX_GET_COMPLEXITY: |
292 | 0 | (*(spx_int32_t*)ptr) = st->complexity; |
293 | 0 | break; |
294 | 13.5k | case SPEEX_SET_BITRATE: |
295 | 13.5k | { |
296 | 13.5k | spx_int32_t i=10; |
297 | 13.5k | spx_int32_t rate, target; |
298 | 13.5k | target = (*(spx_int32_t*)ptr); |
299 | 133k | while (i>=0) |
300 | 126k | { |
301 | 126k | speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i); |
302 | 126k | speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate); |
303 | 126k | if (rate <= target) |
304 | 6.65k | break; |
305 | 119k | i--; |
306 | 119k | } |
307 | 13.5k | } |
308 | 13.5k | break; |
309 | 220k | case SPEEX_GET_BITRATE: |
310 | 220k | if (st->submodes[st->submodeID]) |
311 | 215k | (*(spx_int32_t*)ptr) = st->sampling_rate*SUBMODE(bits_per_frame)/NB_FRAME_SIZE; |
312 | 5.14k | else |
313 | 5.14k | (*(spx_int32_t*)ptr) = st->sampling_rate*(NB_SUBMODE_BITS+1)/NB_FRAME_SIZE; |
314 | 220k | break; |
315 | 7.91k | case SPEEX_SET_SAMPLING_RATE: |
316 | 7.91k | st->sampling_rate = (*(spx_int32_t*)ptr); |
317 | 7.91k | break; |
318 | 3.88k | case SPEEX_GET_SAMPLING_RATE: |
319 | 3.88k | (*(spx_int32_t*)ptr)=st->sampling_rate; |
320 | 3.88k | break; |
321 | 0 | case SPEEX_RESET_STATE: |
322 | 0 | { |
323 | 0 | int i; |
324 | 0 | st->bounded_pitch = 1; |
325 | 0 | st->first = 1; |
326 | 0 | for (i=0;i<NB_ORDER;i++) |
327 | 0 | st->old_lsp[i]= DIV32(MULT16_16(QCONST16(3.1415927f, LSP_SHIFT), i+1), NB_ORDER+1); |
328 | 0 | for (i=0;i<NB_ORDER;i++) |
329 | 0 | st->mem_sw[i]=st->mem_sw_whole[i]=st->mem_sp[i]=st->mem_exc[i]=0; |
330 | 0 | for (i=0;i<NB_FRAME_SIZE+NB_PITCH_END+1;i++) |
331 | 0 | st->excBuf[i]=st->swBuf[i]=0; |
332 | 0 | for (i=0;i<NB_WINDOW_SIZE-NB_FRAME_SIZE;i++) |
333 | 0 | st->winBuf[i]=0; |
334 | 0 | } |
335 | 0 | break; |
336 | 0 | case SPEEX_SET_SUBMODE_ENCODING: |
337 | 0 | st->encode_submode = (*(spx_int32_t*)ptr); |
338 | 0 | break; |
339 | 0 | case SPEEX_GET_SUBMODE_ENCODING: |
340 | 0 | (*(spx_int32_t*)ptr) = st->encode_submode; |
341 | 0 | break; |
342 | 7.91k | case SPEEX_GET_LOOKAHEAD: |
343 | 7.91k | (*(spx_int32_t*)ptr)=(NB_WINDOW_SIZE-NB_FRAME_SIZE); |
344 | 7.91k | break; |
345 | 0 | case SPEEX_SET_PLC_TUNING: |
346 | 0 | st->plc_tuning = (*(spx_int32_t*)ptr); |
347 | 0 | if (st->plc_tuning>100) |
348 | 0 | st->plc_tuning=100; |
349 | 0 | break; |
350 | 0 | case SPEEX_GET_PLC_TUNING: |
351 | 0 | (*(spx_int32_t*)ptr)=(st->plc_tuning); |
352 | 0 | break; |
353 | 0 | #ifndef DISABLE_VBR |
354 | 3.55k | case SPEEX_SET_VBR_MAX_BITRATE: |
355 | 3.55k | st->vbr_max = (*(spx_int32_t*)ptr); |
356 | 3.55k | break; |
357 | 0 | case SPEEX_GET_VBR_MAX_BITRATE: |
358 | 0 | (*(spx_int32_t*)ptr) = st->vbr_max; |
359 | 0 | break; |
360 | 0 | #endif /* #ifndef DISABLE_VBR */ |
361 | 7.91k | case SPEEX_SET_HIGHPASS: |
362 | 7.91k | st->highpass_enabled = (*(spx_int32_t*)ptr); |
363 | 7.91k | break; |
364 | 0 | case SPEEX_GET_HIGHPASS: |
365 | 0 | (*(spx_int32_t*)ptr) = st->highpass_enabled; |
366 | 0 | break; |
367 | | |
368 | | /* This is all internal stuff past this point */ |
369 | 16.2k | case SPEEX_GET_PI_GAIN: |
370 | 16.2k | { |
371 | 16.2k | int i; |
372 | 16.2k | spx_word32_t *g = (spx_word32_t*)ptr; |
373 | 81.0k | for (i=0;i<NB_NB_SUBFRAMES;i++) |
374 | 64.8k | g[i]=st->pi_gain[i]; |
375 | 16.2k | } |
376 | 16.2k | break; |
377 | 16.2k | case SPEEX_GET_EXC: |
378 | 16.2k | { |
379 | 16.2k | int i; |
380 | 81.0k | for (i=0;i<NB_NB_SUBFRAMES;i++) |
381 | 64.8k | ((spx_word16_t*)ptr)[i] = compute_rms16(st->exc+i*NB_SUBFRAME_SIZE, NB_SUBFRAME_SIZE); |
382 | 16.2k | } |
383 | 16.2k | break; |
384 | 0 | #ifndef DISABLE_VBR |
385 | 11.0k | case SPEEX_GET_RELATIVE_QUALITY: |
386 | 11.0k | (*(float*)ptr)=st->relative_quality; |
387 | 11.0k | break; |
388 | 0 | #endif /* #ifndef DISABLE_VBR */ |
389 | 16.2k | case SPEEX_SET_INNOVATION_SAVE: |
390 | 16.2k | st->innov_rms_save = (spx_word16_t*)ptr; |
391 | 16.2k | break; |
392 | 5.48k | case SPEEX_SET_WIDEBAND: |
393 | 5.48k | st->isWideband = *((spx_int32_t*)ptr); |
394 | 5.48k | break; |
395 | 0 | case SPEEX_GET_STACK: |
396 | 0 | *((char**)ptr) = st->stack; |
397 | 0 | break; |
398 | 0 | default: |
399 | 0 | speex_warning_int("Unknown nb_ctl request: ", request); |
400 | 0 | return -1; |
401 | 620k | } |
402 | 620k | return 0; |
403 | 620k | } Line | Count | Source | 198 | 310k | { | 199 | 310k | EncState *st; | 200 | 310k | st=(EncState*)state; | 201 | 310k | switch(request) | 202 | 310k | { | 203 | 7.85k | case SPEEX_GET_FRAME_SIZE: | 204 | 7.85k | (*(spx_int32_t*)ptr) = NB_FRAME_SIZE; | 205 | 7.85k | break; | 206 | 0 | case SPEEX_SET_LOW_MODE: | 207 | 41.3k | case SPEEX_SET_MODE: | 208 | 41.3k | st->submodeSelect = st->submodeID = (*(spx_int32_t*)ptr); | 209 | 41.3k | break; | 210 | 10.5k | case SPEEX_GET_LOW_MODE: | 211 | 10.5k | case SPEEX_GET_MODE: | 212 | 10.5k | (*(spx_int32_t*)ptr) = st->submodeID; | 213 | 10.5k | break; | 214 | 0 | #ifndef DISABLE_VBR | 215 | 2.42k | case SPEEX_SET_VBR: | 216 | 2.42k | st->vbr_enabled = (*(spx_int32_t*)ptr); | 217 | 2.42k | break; | 218 | 0 | case SPEEX_GET_VBR: | 219 | 0 | (*(spx_int32_t*)ptr) = st->vbr_enabled; | 220 | 0 | break; | 221 | 1.78k | case SPEEX_SET_VAD: | 222 | 1.78k | st->vad_enabled = (*(spx_int32_t*)ptr); | 223 | 1.78k | break; | 224 | 0 | case SPEEX_GET_VAD: | 225 | 0 | (*(spx_int32_t*)ptr) = st->vad_enabled; | 226 | 0 | break; | 227 | 1.71k | case SPEEX_SET_DTX: | 228 | 1.71k | st->dtx_enabled = (*(spx_int32_t*)ptr); | 229 | 1.71k | break; | 230 | 0 | case SPEEX_GET_DTX: | 231 | 0 | (*(spx_int32_t*)ptr) = st->dtx_enabled; | 232 | 0 | break; | 233 | 544 | case SPEEX_SET_ABR: | 234 | 544 | st->abr_enabled = (*(spx_int32_t*)ptr); | 235 | 544 | st->vbr_enabled = st->abr_enabled!=0; | 236 | 544 | if (st->vbr_enabled) | 237 | 544 | { | 238 | 544 | spx_int32_t i=10; | 239 | 544 | spx_int32_t rate, target; | 240 | 544 | float vbr_qual; | 241 | 544 | target = (*(spx_int32_t*)ptr); | 242 | 6.52k | while (i>=0) | 243 | 5.98k | { | 244 | 5.98k | speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i); | 245 | 5.98k | speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate); | 246 | 5.98k | if (rate <= target) | 247 | 0 | break; | 248 | 5.98k | i--; | 249 | 5.98k | } | 250 | 544 | vbr_qual=i; | 251 | 544 | if (vbr_qual<0) | 252 | 544 | vbr_qual=0; | 253 | 544 | speex_encoder_ctl(st, SPEEX_SET_VBR_QUALITY, &vbr_qual); | 254 | 544 | st->abr_count=0; | 255 | 544 | st->abr_drift=0; | 256 | 544 | st->abr_drift2=0; | 257 | 544 | } | 258 | | | 259 | 544 | break; | 260 | 0 | case SPEEX_GET_ABR: | 261 | 0 | (*(spx_int32_t*)ptr) = st->abr_enabled; | 262 | 0 | break; | 263 | 0 | #endif /* #ifndef DISABLE_VBR */ | 264 | 0 | #if !defined(DISABLE_VBR) && !defined(DISABLE_FLOAT_API) | 265 | 2.96k | case SPEEX_SET_VBR_QUALITY: | 266 | 2.96k | st->vbr_quality = (*(float*)ptr); | 267 | 2.96k | if (st->vbr_quality < 0) | 268 | 0 | st->vbr_quality = 0; | 269 | 2.96k | else if (st->vbr_quality > 10) | 270 | 0 | st->vbr_quality = 10; | 271 | 2.96k | break; | 272 | 0 | case SPEEX_GET_VBR_QUALITY: | 273 | 0 | (*(float*)ptr) = st->vbr_quality; | 274 | 0 | break; | 275 | 0 | #endif /* !defined(DISABLE_VBR) && !defined(DISABLE_FLOAT_API) */ | 276 | 72.0k | case SPEEX_SET_QUALITY: | 277 | 72.0k | { | 278 | 72.0k | int quality = (*(spx_int32_t*)ptr); | 279 | 72.0k | if (quality < 0) | 280 | 0 | quality = 0; | 281 | 72.0k | if (quality > 10) | 282 | 0 | quality = 10; | 283 | 72.0k | st->submodeSelect = st->submodeID = ((const SpeexNBMode*)(st->mode->mode))->quality_map[quality]; | 284 | 72.0k | } | 285 | 72.0k | break; | 286 | 3.95k | case SPEEX_SET_COMPLEXITY: | 287 | 3.95k | st->complexity = (*(spx_int32_t*)ptr); | 288 | 3.95k | if (st->complexity<0) | 289 | 0 | st->complexity=0; | 290 | 3.95k | break; | 291 | 0 | case SPEEX_GET_COMPLEXITY: | 292 | 0 | (*(spx_int32_t*)ptr) = st->complexity; | 293 | 0 | break; | 294 | 6.77k | case SPEEX_SET_BITRATE: | 295 | 6.77k | { | 296 | 6.77k | spx_int32_t i=10; | 297 | 6.77k | spx_int32_t rate, target; | 298 | 6.77k | target = (*(spx_int32_t*)ptr); | 299 | 66.5k | while (i>=0) | 300 | 63.1k | { | 301 | 63.1k | speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i); | 302 | 63.1k | speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate); | 303 | 63.1k | if (rate <= target) | 304 | 3.32k | break; | 305 | 59.8k | i--; | 306 | 59.8k | } | 307 | 6.77k | } | 308 | 6.77k | break; | 309 | 110k | case SPEEX_GET_BITRATE: | 310 | 110k | if (st->submodes[st->submodeID]) | 311 | 107k | (*(spx_int32_t*)ptr) = st->sampling_rate*SUBMODE(bits_per_frame)/NB_FRAME_SIZE; | 312 | 2.57k | else | 313 | 2.57k | (*(spx_int32_t*)ptr) = st->sampling_rate*(NB_SUBMODE_BITS+1)/NB_FRAME_SIZE; | 314 | 110k | break; | 315 | 3.95k | case SPEEX_SET_SAMPLING_RATE: | 316 | 3.95k | st->sampling_rate = (*(spx_int32_t*)ptr); | 317 | 3.95k | break; | 318 | 1.94k | case SPEEX_GET_SAMPLING_RATE: | 319 | 1.94k | (*(spx_int32_t*)ptr)=st->sampling_rate; | 320 | 1.94k | break; | 321 | 0 | case SPEEX_RESET_STATE: | 322 | 0 | { | 323 | 0 | int i; | 324 | 0 | st->bounded_pitch = 1; | 325 | 0 | st->first = 1; | 326 | 0 | for (i=0;i<NB_ORDER;i++) | 327 | 0 | st->old_lsp[i]= DIV32(MULT16_16(QCONST16(3.1415927f, LSP_SHIFT), i+1), NB_ORDER+1); | 328 | 0 | for (i=0;i<NB_ORDER;i++) | 329 | 0 | st->mem_sw[i]=st->mem_sw_whole[i]=st->mem_sp[i]=st->mem_exc[i]=0; | 330 | 0 | for (i=0;i<NB_FRAME_SIZE+NB_PITCH_END+1;i++) | 331 | 0 | st->excBuf[i]=st->swBuf[i]=0; | 332 | 0 | for (i=0;i<NB_WINDOW_SIZE-NB_FRAME_SIZE;i++) | 333 | 0 | st->winBuf[i]=0; | 334 | 0 | } | 335 | 0 | break; | 336 | 0 | case SPEEX_SET_SUBMODE_ENCODING: | 337 | 0 | st->encode_submode = (*(spx_int32_t*)ptr); | 338 | 0 | break; | 339 | 0 | case SPEEX_GET_SUBMODE_ENCODING: | 340 | 0 | (*(spx_int32_t*)ptr) = st->encode_submode; | 341 | 0 | break; | 342 | 3.95k | case SPEEX_GET_LOOKAHEAD: | 343 | 3.95k | (*(spx_int32_t*)ptr)=(NB_WINDOW_SIZE-NB_FRAME_SIZE); | 344 | 3.95k | break; | 345 | 0 | case SPEEX_SET_PLC_TUNING: | 346 | 0 | st->plc_tuning = (*(spx_int32_t*)ptr); | 347 | 0 | if (st->plc_tuning>100) | 348 | 0 | st->plc_tuning=100; | 349 | 0 | break; | 350 | 0 | case SPEEX_GET_PLC_TUNING: | 351 | 0 | (*(spx_int32_t*)ptr)=(st->plc_tuning); | 352 | 0 | break; | 353 | 0 | #ifndef DISABLE_VBR | 354 | 1.77k | case SPEEX_SET_VBR_MAX_BITRATE: | 355 | 1.77k | st->vbr_max = (*(spx_int32_t*)ptr); | 356 | 1.77k | break; | 357 | 0 | case SPEEX_GET_VBR_MAX_BITRATE: | 358 | 0 | (*(spx_int32_t*)ptr) = st->vbr_max; | 359 | 0 | break; | 360 | 0 | #endif /* #ifndef DISABLE_VBR */ | 361 | 3.95k | case SPEEX_SET_HIGHPASS: | 362 | 3.95k | st->highpass_enabled = (*(spx_int32_t*)ptr); | 363 | 3.95k | break; | 364 | 0 | case SPEEX_GET_HIGHPASS: | 365 | 0 | (*(spx_int32_t*)ptr) = st->highpass_enabled; | 366 | 0 | break; | 367 | | | 368 | | /* This is all internal stuff past this point */ | 369 | 8.10k | case SPEEX_GET_PI_GAIN: | 370 | 8.10k | { | 371 | 8.10k | int i; | 372 | 8.10k | spx_word32_t *g = (spx_word32_t*)ptr; | 373 | 40.5k | for (i=0;i<NB_NB_SUBFRAMES;i++) | 374 | 32.4k | g[i]=st->pi_gain[i]; | 375 | 8.10k | } | 376 | 8.10k | break; | 377 | 8.10k | case SPEEX_GET_EXC: | 378 | 8.10k | { | 379 | 8.10k | int i; | 380 | 40.5k | for (i=0;i<NB_NB_SUBFRAMES;i++) | 381 | 32.4k | ((spx_word16_t*)ptr)[i] = compute_rms16(st->exc+i*NB_SUBFRAME_SIZE, NB_SUBFRAME_SIZE); | 382 | 8.10k | } | 383 | 8.10k | break; | 384 | 0 | #ifndef DISABLE_VBR | 385 | 5.53k | case SPEEX_GET_RELATIVE_QUALITY: | 386 | 5.53k | (*(float*)ptr)=st->relative_quality; | 387 | 5.53k | break; | 388 | 0 | #endif /* #ifndef DISABLE_VBR */ | 389 | 8.10k | case SPEEX_SET_INNOVATION_SAVE: | 390 | 8.10k | st->innov_rms_save = (spx_word16_t*)ptr; | 391 | 8.10k | break; | 392 | 2.74k | case SPEEX_SET_WIDEBAND: | 393 | 2.74k | st->isWideband = *((spx_int32_t*)ptr); | 394 | 2.74k | break; | 395 | 0 | case SPEEX_GET_STACK: | 396 | 0 | *((char**)ptr) = st->stack; | 397 | 0 | break; | 398 | 0 | default: | 399 | 0 | speex_warning_int("Unknown nb_ctl request: ", request); | 400 | 0 | return -1; | 401 | 310k | } | 402 | 310k | return 0; | 403 | 310k | } |
Line | Count | Source | 198 | 310k | { | 199 | 310k | EncState *st; | 200 | 310k | st=(EncState*)state; | 201 | 310k | switch(request) | 202 | 310k | { | 203 | 7.85k | case SPEEX_GET_FRAME_SIZE: | 204 | 7.85k | (*(spx_int32_t*)ptr) = NB_FRAME_SIZE; | 205 | 7.85k | break; | 206 | 0 | case SPEEX_SET_LOW_MODE: | 207 | 41.3k | case SPEEX_SET_MODE: | 208 | 41.3k | st->submodeSelect = st->submodeID = (*(spx_int32_t*)ptr); | 209 | 41.3k | break; | 210 | 10.5k | case SPEEX_GET_LOW_MODE: | 211 | 10.5k | case SPEEX_GET_MODE: | 212 | 10.5k | (*(spx_int32_t*)ptr) = st->submodeID; | 213 | 10.5k | break; | 214 | 0 | #ifndef DISABLE_VBR | 215 | 2.42k | case SPEEX_SET_VBR: | 216 | 2.42k | st->vbr_enabled = (*(spx_int32_t*)ptr); | 217 | 2.42k | break; | 218 | 0 | case SPEEX_GET_VBR: | 219 | 0 | (*(spx_int32_t*)ptr) = st->vbr_enabled; | 220 | 0 | break; | 221 | 1.78k | case SPEEX_SET_VAD: | 222 | 1.78k | st->vad_enabled = (*(spx_int32_t*)ptr); | 223 | 1.78k | break; | 224 | 0 | case SPEEX_GET_VAD: | 225 | 0 | (*(spx_int32_t*)ptr) = st->vad_enabled; | 226 | 0 | break; | 227 | 1.71k | case SPEEX_SET_DTX: | 228 | 1.71k | st->dtx_enabled = (*(spx_int32_t*)ptr); | 229 | 1.71k | break; | 230 | 0 | case SPEEX_GET_DTX: | 231 | 0 | (*(spx_int32_t*)ptr) = st->dtx_enabled; | 232 | 0 | break; | 233 | 544 | case SPEEX_SET_ABR: | 234 | 544 | st->abr_enabled = (*(spx_int32_t*)ptr); | 235 | 544 | st->vbr_enabled = st->abr_enabled!=0; | 236 | 544 | if (st->vbr_enabled) | 237 | 544 | { | 238 | 544 | spx_int32_t i=10; | 239 | 544 | spx_int32_t rate, target; | 240 | 544 | float vbr_qual; | 241 | 544 | target = (*(spx_int32_t*)ptr); | 242 | 6.52k | while (i>=0) | 243 | 5.98k | { | 244 | 5.98k | speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i); | 245 | 5.98k | speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate); | 246 | 5.98k | if (rate <= target) | 247 | 0 | break; | 248 | 5.98k | i--; | 249 | 5.98k | } | 250 | 544 | vbr_qual=i; | 251 | 544 | if (vbr_qual<0) | 252 | 544 | vbr_qual=0; | 253 | 544 | speex_encoder_ctl(st, SPEEX_SET_VBR_QUALITY, &vbr_qual); | 254 | 544 | st->abr_count=0; | 255 | 544 | st->abr_drift=0; | 256 | 544 | st->abr_drift2=0; | 257 | 544 | } | 258 | | | 259 | 544 | break; | 260 | 0 | case SPEEX_GET_ABR: | 261 | 0 | (*(spx_int32_t*)ptr) = st->abr_enabled; | 262 | 0 | break; | 263 | 0 | #endif /* #ifndef DISABLE_VBR */ | 264 | 0 | #if !defined(DISABLE_VBR) && !defined(DISABLE_FLOAT_API) | 265 | 2.96k | case SPEEX_SET_VBR_QUALITY: | 266 | 2.96k | st->vbr_quality = (*(float*)ptr); | 267 | 2.96k | if (st->vbr_quality < 0) | 268 | 0 | st->vbr_quality = 0; | 269 | 2.96k | else if (st->vbr_quality > 10) | 270 | 0 | st->vbr_quality = 10; | 271 | 2.96k | break; | 272 | 0 | case SPEEX_GET_VBR_QUALITY: | 273 | 0 | (*(float*)ptr) = st->vbr_quality; | 274 | 0 | break; | 275 | 0 | #endif /* !defined(DISABLE_VBR) && !defined(DISABLE_FLOAT_API) */ | 276 | 72.0k | case SPEEX_SET_QUALITY: | 277 | 72.0k | { | 278 | 72.0k | int quality = (*(spx_int32_t*)ptr); | 279 | 72.0k | if (quality < 0) | 280 | 0 | quality = 0; | 281 | 72.0k | if (quality > 10) | 282 | 0 | quality = 10; | 283 | 72.0k | st->submodeSelect = st->submodeID = ((const SpeexNBMode*)(st->mode->mode))->quality_map[quality]; | 284 | 72.0k | } | 285 | 72.0k | break; | 286 | 3.95k | case SPEEX_SET_COMPLEXITY: | 287 | 3.95k | st->complexity = (*(spx_int32_t*)ptr); | 288 | 3.95k | if (st->complexity<0) | 289 | 0 | st->complexity=0; | 290 | 3.95k | break; | 291 | 0 | case SPEEX_GET_COMPLEXITY: | 292 | 0 | (*(spx_int32_t*)ptr) = st->complexity; | 293 | 0 | break; | 294 | 6.77k | case SPEEX_SET_BITRATE: | 295 | 6.77k | { | 296 | 6.77k | spx_int32_t i=10; | 297 | 6.77k | spx_int32_t rate, target; | 298 | 6.77k | target = (*(spx_int32_t*)ptr); | 299 | 66.5k | while (i>=0) | 300 | 63.1k | { | 301 | 63.1k | speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i); | 302 | 63.1k | speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate); | 303 | 63.1k | if (rate <= target) | 304 | 3.32k | break; | 305 | 59.8k | i--; | 306 | 59.8k | } | 307 | 6.77k | } | 308 | 6.77k | break; | 309 | 110k | case SPEEX_GET_BITRATE: | 310 | 110k | if (st->submodes[st->submodeID]) | 311 | 107k | (*(spx_int32_t*)ptr) = st->sampling_rate*SUBMODE(bits_per_frame)/NB_FRAME_SIZE; | 312 | 2.57k | else | 313 | 2.57k | (*(spx_int32_t*)ptr) = st->sampling_rate*(NB_SUBMODE_BITS+1)/NB_FRAME_SIZE; | 314 | 110k | break; | 315 | 3.95k | case SPEEX_SET_SAMPLING_RATE: | 316 | 3.95k | st->sampling_rate = (*(spx_int32_t*)ptr); | 317 | 3.95k | break; | 318 | 1.94k | case SPEEX_GET_SAMPLING_RATE: | 319 | 1.94k | (*(spx_int32_t*)ptr)=st->sampling_rate; | 320 | 1.94k | break; | 321 | 0 | case SPEEX_RESET_STATE: | 322 | 0 | { | 323 | 0 | int i; | 324 | 0 | st->bounded_pitch = 1; | 325 | 0 | st->first = 1; | 326 | 0 | for (i=0;i<NB_ORDER;i++) | 327 | 0 | st->old_lsp[i]= DIV32(MULT16_16(QCONST16(3.1415927f, LSP_SHIFT), i+1), NB_ORDER+1); | 328 | 0 | for (i=0;i<NB_ORDER;i++) | 329 | 0 | st->mem_sw[i]=st->mem_sw_whole[i]=st->mem_sp[i]=st->mem_exc[i]=0; | 330 | 0 | for (i=0;i<NB_FRAME_SIZE+NB_PITCH_END+1;i++) | 331 | 0 | st->excBuf[i]=st->swBuf[i]=0; | 332 | 0 | for (i=0;i<NB_WINDOW_SIZE-NB_FRAME_SIZE;i++) | 333 | 0 | st->winBuf[i]=0; | 334 | 0 | } | 335 | 0 | break; | 336 | 0 | case SPEEX_SET_SUBMODE_ENCODING: | 337 | 0 | st->encode_submode = (*(spx_int32_t*)ptr); | 338 | 0 | break; | 339 | 0 | case SPEEX_GET_SUBMODE_ENCODING: | 340 | 0 | (*(spx_int32_t*)ptr) = st->encode_submode; | 341 | 0 | break; | 342 | 3.95k | case SPEEX_GET_LOOKAHEAD: | 343 | 3.95k | (*(spx_int32_t*)ptr)=(NB_WINDOW_SIZE-NB_FRAME_SIZE); | 344 | 3.95k | break; | 345 | 0 | case SPEEX_SET_PLC_TUNING: | 346 | 0 | st->plc_tuning = (*(spx_int32_t*)ptr); | 347 | 0 | if (st->plc_tuning>100) | 348 | 0 | st->plc_tuning=100; | 349 | 0 | break; | 350 | 0 | case SPEEX_GET_PLC_TUNING: | 351 | 0 | (*(spx_int32_t*)ptr)=(st->plc_tuning); | 352 | 0 | break; | 353 | 0 | #ifndef DISABLE_VBR | 354 | 1.77k | case SPEEX_SET_VBR_MAX_BITRATE: | 355 | 1.77k | st->vbr_max = (*(spx_int32_t*)ptr); | 356 | 1.77k | break; | 357 | 0 | case SPEEX_GET_VBR_MAX_BITRATE: | 358 | 0 | (*(spx_int32_t*)ptr) = st->vbr_max; | 359 | 0 | break; | 360 | 0 | #endif /* #ifndef DISABLE_VBR */ | 361 | 3.95k | case SPEEX_SET_HIGHPASS: | 362 | 3.95k | st->highpass_enabled = (*(spx_int32_t*)ptr); | 363 | 3.95k | break; | 364 | 0 | case SPEEX_GET_HIGHPASS: | 365 | 0 | (*(spx_int32_t*)ptr) = st->highpass_enabled; | 366 | 0 | break; | 367 | | | 368 | | /* This is all internal stuff past this point */ | 369 | 8.10k | case SPEEX_GET_PI_GAIN: | 370 | 8.10k | { | 371 | 8.10k | int i; | 372 | 8.10k | spx_word32_t *g = (spx_word32_t*)ptr; | 373 | 40.5k | for (i=0;i<NB_NB_SUBFRAMES;i++) | 374 | 32.4k | g[i]=st->pi_gain[i]; | 375 | 8.10k | } | 376 | 8.10k | break; | 377 | 8.10k | case SPEEX_GET_EXC: | 378 | 8.10k | { | 379 | 8.10k | int i; | 380 | 40.5k | for (i=0;i<NB_NB_SUBFRAMES;i++) | 381 | 32.4k | ((spx_word16_t*)ptr)[i] = compute_rms16(st->exc+i*NB_SUBFRAME_SIZE, NB_SUBFRAME_SIZE); | 382 | 8.10k | } | 383 | 8.10k | break; | 384 | 0 | #ifndef DISABLE_VBR | 385 | 5.53k | case SPEEX_GET_RELATIVE_QUALITY: | 386 | 5.53k | (*(float*)ptr)=st->relative_quality; | 387 | 5.53k | break; | 388 | 0 | #endif /* #ifndef DISABLE_VBR */ | 389 | 8.10k | case SPEEX_SET_INNOVATION_SAVE: | 390 | 8.10k | st->innov_rms_save = (spx_word16_t*)ptr; | 391 | 8.10k | break; | 392 | 2.74k | case SPEEX_SET_WIDEBAND: | 393 | 2.74k | st->isWideband = *((spx_int32_t*)ptr); | 394 | 2.74k | break; | 395 | 0 | case SPEEX_GET_STACK: | 396 | 0 | *((char**)ptr) = st->stack; | 397 | 0 | break; | 398 | 0 | default: | 399 | 0 | speex_warning_int("Unknown nb_ctl request: ", request); | 400 | 0 | return -1; | 401 | 310k | } | 402 | 310k | return 0; | 403 | 310k | } |
|
404 | | |
405 | | |
406 | | int nb_encode(void *state, void *vin, SpeexBits *bits) |
407 | 28.2k | { |
408 | 28.2k | EncState *st; |
409 | 28.2k | int i, sub, roots; |
410 | 28.2k | int ol_pitch; |
411 | 28.2k | spx_word16_t ol_pitch_coef; |
412 | 28.2k | spx_word32_t ol_gain; |
413 | 28.2k | VARDECL(spx_word16_t *target); |
414 | 28.2k | VARDECL(spx_sig_t *innov); |
415 | 28.2k | VARDECL(spx_word32_t *exc32); |
416 | 28.2k | VARDECL(spx_mem_t *mem); |
417 | 28.2k | VARDECL(spx_coef_t *bw_lpc1); |
418 | 28.2k | VARDECL(spx_coef_t *bw_lpc2); |
419 | 28.2k | VARDECL(spx_coef_t *lpc); |
420 | 28.2k | VARDECL(spx_lsp_t *lsp); |
421 | 28.2k | VARDECL(spx_lsp_t *qlsp); |
422 | 28.2k | VARDECL(spx_lsp_t *interp_lsp); |
423 | 28.2k | VARDECL(spx_lsp_t *interp_qlsp); |
424 | 28.2k | VARDECL(spx_coef_t *interp_lpc); |
425 | 28.2k | VARDECL(spx_coef_t *interp_qlpc); |
426 | 28.2k | char *stack; |
427 | 28.2k | VARDECL(spx_word16_t *syn_resp); |
428 | | |
429 | 28.2k | spx_word32_t ener=0; |
430 | 28.2k | spx_word16_t fine_gain; |
431 | 28.2k | spx_word16_t *in = (spx_word16_t*)vin; |
432 | | |
433 | 28.2k | st=(EncState *)state; |
434 | 28.2k | stack=st->stack; |
435 | | |
436 | 28.2k | ALLOC(lpc, NB_ORDER, spx_coef_t); |
437 | 28.2k | ALLOC(bw_lpc1, NB_ORDER, spx_coef_t); |
438 | 28.2k | ALLOC(bw_lpc2, NB_ORDER, spx_coef_t); |
439 | 28.2k | ALLOC(lsp, NB_ORDER, spx_lsp_t); |
440 | 28.2k | ALLOC(qlsp, NB_ORDER, spx_lsp_t); |
441 | 28.2k | ALLOC(interp_lsp, NB_ORDER, spx_lsp_t); |
442 | 28.2k | ALLOC(interp_qlsp, NB_ORDER, spx_lsp_t); |
443 | 28.2k | ALLOC(interp_lpc, NB_ORDER, spx_coef_t); |
444 | 28.2k | ALLOC(interp_qlpc, NB_ORDER, spx_coef_t); |
445 | | |
446 | 28.2k | st->exc = st->excBuf + NB_PITCH_END + 2; |
447 | 28.2k | st->sw = st->swBuf + NB_PITCH_END + 2; |
448 | | /* Move signals 1 frame towards the past */ |
449 | 28.2k | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, NB_PITCH_END+2); |
450 | 28.2k | SPEEX_MOVE(st->swBuf, st->swBuf+NB_FRAME_SIZE, NB_PITCH_END+2); |
451 | | |
452 | 28.2k | if (st->highpass_enabled) |
453 | 11.8k | highpass(in, in, NB_FRAME_SIZE, (st->isWideband?HIGHPASS_WIDEBAND:HIGHPASS_NARROWBAND)|HIGHPASS_INPUT, st->mem_hp); |
454 | | |
455 | 28.2k | { |
456 | 28.2k | VARDECL(spx_word16_t *w_sig); |
457 | 28.2k | VARDECL(spx_word16_t *autocorr); |
458 | 28.2k | ALLOC(w_sig, NB_WINDOW_SIZE, spx_word16_t); |
459 | 28.2k | ALLOC(autocorr, NB_ORDER+1, spx_word16_t); |
460 | | /* Window for analysis */ |
461 | 1.15M | for (i=0;i<NB_WINDOW_SIZE-NB_FRAME_SIZE;i++) |
462 | 1.13M | w_sig[i] = MULT16_16_Q15(st->winBuf[i],st->window[i]); |
463 | 4.55M | for (;i<NB_WINDOW_SIZE;i++) |
464 | 4.52M | w_sig[i] = MULT16_16_Q15(in[i-NB_WINDOW_SIZE+NB_FRAME_SIZE],st->window[i]); |
465 | | /* Compute auto-correlation */ |
466 | 28.2k | _spx_autocorr(w_sig, autocorr, NB_ORDER+1, NB_WINDOW_SIZE); |
467 | 28.2k | autocorr[0] = ADD16(autocorr[0],MULT16_16_Q15(autocorr[0],st->lpc_floor)); /* Noise floor in auto-correlation domain */ |
468 | | |
469 | | /* Lag windowing: equivalent to filtering in the power-spectrum domain */ |
470 | 339k | for (i=0;i<NB_ORDER+1;i++) |
471 | 311k | autocorr[i] = MULT16_16_Q15(autocorr[i],st->lagWindow[i]); |
472 | 28.2k | autocorr[0] = ADD16(autocorr[0],1); |
473 | | |
474 | | /* Levinson-Durbin */ |
475 | 28.2k | _spx_lpc(lpc, autocorr, NB_ORDER); |
476 | | /* LPC to LSPs (x-domain) transform */ |
477 | 28.2k | roots=lpc_to_lsp (lpc, NB_ORDER, lsp, 10, LSP_DELTA1, stack); |
478 | | /* Check if we found all the roots */ |
479 | 28.2k | if (roots!=NB_ORDER) |
480 | 56 | { |
481 | | /*If we can't find all LSP's, do some damage control and use previous filter*/ |
482 | 616 | for (i=0;i<NB_ORDER;i++) |
483 | 560 | { |
484 | 560 | lsp[i]=st->old_lsp[i]; |
485 | 560 | } |
486 | 56 | } |
487 | 28.2k | } |
488 | | |
489 | | |
490 | | |
491 | | |
492 | | /* Whole frame analysis (open-loop estimation of pitch and excitation gain) */ |
493 | 28.2k | { |
494 | 28.2k | int diff = NB_WINDOW_SIZE-NB_FRAME_SIZE; |
495 | 28.2k | if (st->first) |
496 | 63.3k | for (i=0;i<NB_ORDER;i++) |
497 | 57.6k | interp_lsp[i] = lsp[i]; |
498 | 22.5k | else |
499 | 22.5k | lsp_interpolate(st->old_lsp, lsp, interp_lsp, NB_ORDER, NB_NB_SUBFRAMES, NB_NB_SUBFRAMES<<1, LSP_MARGIN); |
500 | | |
501 | | /* Compute interpolated LPCs (unquantized) for whole frame*/ |
502 | 28.2k | lsp_to_lpc(interp_lsp, interp_lpc, NB_ORDER,stack); |
503 | | |
504 | | |
505 | | /*Open-loop pitch*/ |
506 | 28.2k | if (!st->submodes[st->submodeID] || (st->complexity>2 && SUBMODE(have_subframe_gain)<3) || SUBMODE(forced_pitch_gain) || SUBMODE(lbr_pitch) != -1 |
507 | 28.2k | #ifndef DISABLE_VBR |
508 | 28.2k | || st->vbr_enabled || st->vad_enabled |
509 | 28.2k | #endif |
510 | 28.2k | ) |
511 | 26.1k | { |
512 | 26.1k | int nol_pitch[6]; |
513 | 26.1k | spx_word16_t nol_pitch_coef[6]; |
514 | | |
515 | 26.1k | bw_lpc(QCONST16(0.9,15), interp_lpc, bw_lpc1, NB_ORDER); |
516 | 26.1k | bw_lpc(QCONST16(0.55,15), interp_lpc, bw_lpc2, NB_ORDER); |
517 | | |
518 | 26.1k | SPEEX_COPY(st->sw, st->winBuf, diff); |
519 | 26.1k | SPEEX_COPY(st->sw+diff, in, NB_FRAME_SIZE-diff); |
520 | 26.1k | filter10(st->sw, bw_lpc1, bw_lpc2, st->sw, NB_FRAME_SIZE, st->mem_sw_whole, stack); |
521 | | |
522 | 26.1k | open_loop_nbest_pitch(st->sw, NB_PITCH_START, NB_PITCH_END, NB_FRAME_SIZE, |
523 | 26.1k | nol_pitch, nol_pitch_coef, 6, stack); |
524 | 26.1k | ol_pitch=nol_pitch[0]; |
525 | 26.1k | ol_pitch_coef = nol_pitch_coef[0]; |
526 | | /*Try to remove pitch multiples*/ |
527 | 156k | for (i=1;i<6;i++) |
528 | 130k | { |
529 | | #ifdef FIXED_POINT |
530 | 90.8k | if ((nol_pitch_coef[i]>MULT16_16_Q15(nol_pitch_coef[0],27853)) && |
531 | | #else |
532 | 39.9k | if ((nol_pitch_coef[i]>.85*nol_pitch_coef[0]) && |
533 | | #endif |
534 | 39.9k | (ABS(2*nol_pitch[i]-ol_pitch)<=2 || ABS(3*nol_pitch[i]-ol_pitch)<=3 || |
535 | 29.7k | ABS(4*nol_pitch[i]-ol_pitch)<=4 || ABS(5*nol_pitch[i]-ol_pitch)<=5)) |
536 | 1.04k | { |
537 | | /*ol_pitch_coef=nol_pitch_coef[i];*/ |
538 | 1.04k | ol_pitch = nol_pitch[i]; |
539 | 1.04k | } |
540 | 130k | } |
541 | | /*if (ol_pitch>50) |
542 | | ol_pitch/=2;*/ |
543 | | /*ol_pitch_coef = sqrt(ol_pitch_coef);*/ |
544 | | |
545 | 26.1k | } else { |
546 | 2.11k | ol_pitch=0; |
547 | 2.11k | ol_pitch_coef=0; |
548 | 2.11k | } |
549 | | |
550 | | /*Compute "real" excitation*/ |
551 | | /*SPEEX_COPY(st->exc, st->winBuf, diff); |
552 | | SPEEX_COPY(st->exc+diff, in, NB_FRAME_SIZE-diff);*/ |
553 | 28.2k | fir_mem16(st->winBuf, interp_lpc, st->exc, diff, NB_ORDER, st->mem_exc, stack); |
554 | 28.2k | fir_mem16(in, interp_lpc, st->exc+diff, NB_FRAME_SIZE-diff, NB_ORDER, st->mem_exc, stack); |
555 | | |
556 | | /* Compute open-loop excitation gain */ |
557 | 28.2k | { |
558 | 28.2k | spx_word16_t g = compute_rms16(st->exc, NB_FRAME_SIZE); |
559 | 28.2k | if (st->submodeID!=1 && ol_pitch>0) |
560 | 12.7k | ol_gain = MULT16_16(g, MULT16_16_Q14(QCONST16(1.1,14), |
561 | 28.2k | spx_sqrt(QCONST32(1.,28)-MULT16_32_Q15(QCONST16(.8,15),SHL32(MULT16_16(ol_pitch_coef,ol_pitch_coef),16))))); |
562 | 15.5k | else |
563 | 15.5k | ol_gain = SHL32(EXTEND32(g),SIG_SHIFT); |
564 | 28.2k | } |
565 | 28.2k | } |
566 | | |
567 | | #ifdef VORBIS_PSYCHO |
568 | | SPEEX_MOVE(st->psy_window, st->psy_window+NB_FRAME_SIZE, 256-NB_FRAME_SIZE); |
569 | | SPEEX_COPY(&st->psy_window[256-NB_FRAME_SIZE], in, NB_FRAME_SIZE); |
570 | | compute_curve(st->psy, st->psy_window, st->curve); |
571 | | /*print_vec(st->curve, 128, "curve");*/ |
572 | | if (st->first) |
573 | | SPEEX_COPY(st->old_curve, st->curve, 128); |
574 | | #endif |
575 | | |
576 | | /*VBR stuff*/ |
577 | 28.2k | #ifndef DISABLE_VBR |
578 | 28.2k | if (st->vbr_enabled||st->vad_enabled) |
579 | 21.5k | { |
580 | 21.5k | float lsp_dist=0; |
581 | 236k | for (i=0;i<NB_ORDER;i++) |
582 | 215k | lsp_dist += (st->old_lsp[i] - lsp[i])*(st->old_lsp[i] - lsp[i]); |
583 | 21.5k | lsp_dist /= LSP_SCALING*LSP_SCALING; |
584 | | |
585 | 21.5k | if (st->abr_enabled) |
586 | 5.44k | { |
587 | 5.44k | float qual_change=0; |
588 | 5.44k | if (st->abr_drift2 * st->abr_drift > 0) |
589 | 4.89k | { |
590 | | /* Only adapt if long-term and short-term drift are the same sign */ |
591 | 4.89k | qual_change = -.00001*st->abr_drift/(1+st->abr_count); |
592 | 4.89k | if (qual_change>.05) |
593 | 0 | qual_change=.05; |
594 | 4.89k | if (qual_change<-.05) |
595 | 1.19k | qual_change=-.05; |
596 | 4.89k | } |
597 | 5.44k | st->vbr_quality += qual_change; |
598 | 5.44k | if (st->vbr_quality>10) |
599 | 0 | st->vbr_quality=10; |
600 | 5.44k | if (st->vbr_quality<0) |
601 | 4.89k | st->vbr_quality=0; |
602 | 5.44k | } |
603 | | |
604 | 21.5k | st->relative_quality = vbr_analysis(&st->vbr, in, NB_FRAME_SIZE, ol_pitch, GAIN_SCALING_1*ol_pitch_coef); |
605 | | /*if (delta_qual<0)*/ |
606 | | /* delta_qual*=.1*(3+st->vbr_quality);*/ |
607 | 21.5k | if (st->vbr_enabled) |
608 | 16.9k | { |
609 | 16.9k | spx_int32_t mode; |
610 | 16.9k | int choice=0; |
611 | 16.9k | float min_diff=100; |
612 | 16.9k | mode = 8; |
613 | 152k | while (mode) |
614 | 135k | { |
615 | 135k | int v1; |
616 | 135k | float thresh; |
617 | 135k | v1=(int)floor(st->vbr_quality); |
618 | 135k | if (v1==10) |
619 | 9.61k | thresh = vbr_nb_thresh[mode][v1]; |
620 | 125k | else |
621 | 125k | thresh = (st->vbr_quality-v1)*vbr_nb_thresh[mode][v1+1] + (1+v1-st->vbr_quality)*vbr_nb_thresh[mode][v1]; |
622 | 135k | if (st->relative_quality > thresh && |
623 | 135k | st->relative_quality-thresh<min_diff) |
624 | 17.5k | { |
625 | 17.5k | choice = mode; |
626 | 17.5k | min_diff = st->relative_quality-thresh; |
627 | 17.5k | } |
628 | 135k | mode--; |
629 | 135k | } |
630 | 16.9k | mode=choice; |
631 | 16.9k | if (mode==0) |
632 | 4.85k | { |
633 | 4.85k | if (st->dtx_count==0 || lsp_dist>.05 || !st->dtx_enabled || st->dtx_count>20) |
634 | 2.88k | { |
635 | 2.88k | mode=1; |
636 | 2.88k | st->dtx_count=1; |
637 | 2.88k | } else { |
638 | 1.96k | mode=0; |
639 | 1.96k | st->dtx_count++; |
640 | 1.96k | } |
641 | 12.0k | } else { |
642 | 12.0k | st->dtx_count=0; |
643 | 12.0k | } |
644 | | |
645 | 16.9k | speex_encoder_ctl(state, SPEEX_SET_MODE, &mode); |
646 | 16.9k | if (st->vbr_max>0) |
647 | 12.5k | { |
648 | 12.5k | spx_int32_t rate; |
649 | 12.5k | speex_encoder_ctl(state, SPEEX_GET_BITRATE, &rate); |
650 | 12.5k | if (rate > st->vbr_max) |
651 | 4.76k | { |
652 | 4.76k | rate = st->vbr_max; |
653 | 4.76k | speex_encoder_ctl(state, SPEEX_SET_BITRATE, &rate); |
654 | 4.76k | } |
655 | 12.5k | } |
656 | | |
657 | 16.9k | if (st->abr_enabled) |
658 | 5.44k | { |
659 | 5.44k | spx_int32_t bitrate; |
660 | 5.44k | speex_encoder_ctl(state, SPEEX_GET_BITRATE, &bitrate); |
661 | 5.44k | st->abr_drift+=(bitrate-st->abr_enabled); |
662 | 5.44k | st->abr_drift2 = .95*st->abr_drift2 + .05*(bitrate-st->abr_enabled); |
663 | 5.44k | st->abr_count += 1.0; |
664 | 5.44k | } |
665 | | |
666 | 16.9k | } else { |
667 | | /*VAD only case*/ |
668 | 4.60k | int mode; |
669 | 4.60k | if (st->relative_quality<2) |
670 | 1.47k | { |
671 | 1.47k | if (st->dtx_count==0 || lsp_dist>.05 || !st->dtx_enabled || st->dtx_count>20) |
672 | 906 | { |
673 | 906 | st->dtx_count=1; |
674 | 906 | mode=1; |
675 | 906 | } else { |
676 | 564 | mode=0; |
677 | 564 | st->dtx_count++; |
678 | 564 | } |
679 | 3.13k | } else { |
680 | 3.13k | st->dtx_count = 0; |
681 | 3.13k | mode=st->submodeSelect; |
682 | 3.13k | } |
683 | | /*speex_encoder_ctl(state, SPEEX_SET_MODE, &mode);*/ |
684 | 4.60k | st->submodeID=mode; |
685 | 4.60k | } |
686 | 21.5k | } else { |
687 | 6.74k | st->relative_quality = -1; |
688 | 6.74k | } |
689 | 28.2k | #endif /* #ifndef DISABLE_VBR */ |
690 | | |
691 | 28.2k | if (st->encode_submode) |
692 | 28.2k | { |
693 | | /* First, transmit a zero for narrowband */ |
694 | 28.2k | speex_bits_pack(bits, 0, 1); |
695 | | |
696 | | /* Transmit the sub-mode we use for this frame */ |
697 | 28.2k | speex_bits_pack(bits, st->submodeID, NB_SUBMODE_BITS); |
698 | | |
699 | 28.2k | } |
700 | | |
701 | | /* If null mode (no transmission), just set a couple things to zero*/ |
702 | 28.2k | if (st->submodes[st->submodeID] == NULL) |
703 | 2.38k | { |
704 | 384k | for (i=0;i<NB_FRAME_SIZE;i++) |
705 | 381k | st->exc[i]=st->sw[i]=VERY_SMALL; |
706 | | |
707 | 26.2k | for (i=0;i<NB_ORDER;i++) |
708 | 23.8k | st->mem_sw[i]=0; |
709 | 2.38k | st->first=1; |
710 | 2.38k | st->bounded_pitch = 1; |
711 | | |
712 | 2.38k | SPEEX_COPY(st->winBuf, in+2*NB_FRAME_SIZE-NB_WINDOW_SIZE, NB_WINDOW_SIZE-NB_FRAME_SIZE); |
713 | | |
714 | | /* Clear memory (no need to really compute it) */ |
715 | 26.2k | for (i=0;i<NB_ORDER;i++) |
716 | 23.8k | st->mem_sp[i] = 0; |
717 | 2.38k | return 0; |
718 | | |
719 | 2.38k | } |
720 | | |
721 | | /* LSP Quantization */ |
722 | 25.8k | if (st->first) |
723 | 4.02k | { |
724 | 44.2k | for (i=0;i<NB_ORDER;i++) |
725 | 40.2k | st->old_lsp[i] = lsp[i]; |
726 | 4.02k | } |
727 | | |
728 | | |
729 | | /*Quantize LSPs*/ |
730 | 25.8k | #if 1 /*0 for unquantized*/ |
731 | 25.8k | SUBMODE(lsp_quant)(lsp, qlsp, NB_ORDER, bits); |
732 | | #else |
733 | | for (i=0;i<NB_ORDER;i++) |
734 | | qlsp[i]=lsp[i]; |
735 | | #endif |
736 | | |
737 | | /*If we use low bit-rate pitch mode, transmit open-loop pitch*/ |
738 | 25.8k | if (SUBMODE(lbr_pitch)!=-1) |
739 | 17.4k | { |
740 | 17.4k | speex_bits_pack(bits, ol_pitch-NB_PITCH_START, 7); |
741 | 17.4k | } |
742 | | |
743 | 25.8k | if (SUBMODE(forced_pitch_gain)) |
744 | 15.9k | { |
745 | 15.9k | int quant; |
746 | | /* This just damps the pitch a bit, because it tends to be too aggressive when forced */ |
747 | 15.9k | ol_pitch_coef = MULT16_16_Q15(QCONST16(.9,15), ol_pitch_coef); |
748 | | #ifdef FIXED_POINT |
749 | 11.2k | quant = PSHR16(MULT16_16_16(15, ol_pitch_coef),GAIN_SHIFT); |
750 | | #else |
751 | 4.77k | quant = (int)floor(.5+15*ol_pitch_coef*GAIN_SCALING_1); |
752 | | #endif |
753 | 15.9k | if (quant>15) |
754 | 131 | quant=15; |
755 | 15.9k | if (quant<0) |
756 | 0 | quant=0; |
757 | 15.9k | speex_bits_pack(bits, quant, 4); |
758 | 15.9k | ol_pitch_coef=MULT16_16_P15(QCONST16(0.066667,15),SHL16(quant,GAIN_SHIFT)); |
759 | 15.9k | } |
760 | | |
761 | | |
762 | | /*Quantize and transmit open-loop excitation gain*/ |
763 | | #ifdef FIXED_POINT |
764 | | { |
765 | | int qe = scal_quant32(ol_gain, ol_gain_table, 32); |
766 | | /*ol_gain = exp(qe/3.5)*SIG_SCALING;*/ |
767 | 18.4k | ol_gain = MULT16_32_Q15(28406,ol_gain_table[qe]); |
768 | | speex_bits_pack(bits, qe, 5); |
769 | | } |
770 | | #else |
771 | | { |
772 | 7.39k | int qe = (int)(floor(.5+3.5*log(ol_gain*1.0/SIG_SCALING))); |
773 | 7.39k | if (qe<0) |
774 | 4.67k | qe=0; |
775 | 7.39k | if (qe>31) |
776 | 422 | qe=31; |
777 | 7.39k | ol_gain = exp(qe/3.5)*SIG_SCALING; |
778 | | speex_bits_pack(bits, qe, 5); |
779 | | } |
780 | | #endif |
781 | | |
782 | | |
783 | | |
784 | | /* Special case for first frame */ |
785 | 25.8k | if (st->first) |
786 | 4.02k | { |
787 | 44.2k | for (i=0;i<NB_ORDER;i++) |
788 | 40.2k | st->old_qlsp[i] = qlsp[i]; |
789 | 4.02k | } |
790 | | |
791 | | /* Target signal */ |
792 | 25.8k | ALLOC(target, NB_SUBFRAME_SIZE, spx_word16_t); |
793 | 25.8k | ALLOC(innov, NB_SUBFRAME_SIZE, spx_sig_t); |
794 | 25.8k | ALLOC(exc32, NB_SUBFRAME_SIZE, spx_word32_t); |
795 | 25.8k | ALLOC(syn_resp, NB_SUBFRAME_SIZE, spx_word16_t); |
796 | 25.8k | ALLOC(mem, NB_ORDER, spx_mem_t); |
797 | | |
798 | | /* Loop on sub-frames */ |
799 | 129k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) |
800 | 103k | { |
801 | 103k | int offset; |
802 | 103k | spx_word16_t *sw; |
803 | 103k | spx_word16_t *exc, *inBuf; |
804 | 103k | int pitch; |
805 | 103k | int response_bound = NB_SUBFRAME_SIZE; |
806 | | |
807 | | /* Offset relative to start of frame */ |
808 | 103k | offset = NB_SUBFRAME_SIZE*sub; |
809 | | /* Excitation */ |
810 | 103k | exc=st->exc+offset; |
811 | | /* Weighted signal */ |
812 | 103k | sw=st->sw+offset; |
813 | | |
814 | | /* LSP interpolation (quantized and unquantized) */ |
815 | 103k | lsp_interpolate(st->old_lsp, lsp, interp_lsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); |
816 | 103k | lsp_interpolate(st->old_qlsp, qlsp, interp_qlsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); |
817 | | |
818 | | /* Compute interpolated LPCs (quantized and unquantized) */ |
819 | 103k | lsp_to_lpc(interp_lsp, interp_lpc, NB_ORDER,stack); |
820 | | |
821 | 103k | lsp_to_lpc(interp_qlsp, interp_qlpc, NB_ORDER, stack); |
822 | | |
823 | | /* Compute analysis filter gain at w=pi (for use in SB-CELP) */ |
824 | 103k | { |
825 | 103k | spx_word32_t pi_g=LPC_SCALING; |
826 | 621k | for (i=0;i<NB_ORDER;i+=2) |
827 | 517k | { |
828 | | /*pi_g += -st->interp_qlpc[i] + st->interp_qlpc[i+1];*/ |
829 | 517k | pi_g = ADD32(pi_g, SUB32(EXTEND32(interp_qlpc[i+1]),EXTEND32(interp_qlpc[i]))); |
830 | 517k | } |
831 | 103k | st->pi_gain[sub] = pi_g; |
832 | 103k | } |
833 | | |
834 | | #ifdef VORBIS_PSYCHO |
835 | | { |
836 | | float curr_curve[128]; |
837 | | float fact = ((float)sub+1.0f)/NB_NB_SUBFRAMES; |
838 | | for (i=0;i<128;i++) |
839 | | curr_curve[i] = (1.0f-fact)*st->old_curve[i] + fact*st->curve[i]; |
840 | | curve_to_lpc(st->psy, curr_curve, bw_lpc1, bw_lpc2, 10); |
841 | | } |
842 | | #else |
843 | | /* Compute bandwidth-expanded (unquantized) LPCs for perceptual weighting */ |
844 | 103k | bw_lpc(st->gamma1, interp_lpc, bw_lpc1, NB_ORDER); |
845 | 103k | bw_lpc(st->gamma2, interp_lpc, bw_lpc2, NB_ORDER); |
846 | | /*print_vec(st->bw_lpc1, 10, "bw_lpc");*/ |
847 | 103k | #endif |
848 | | |
849 | | /*FIXME: This will break if we change the window size */ |
850 | 103k | speex_assert(NB_WINDOW_SIZE-NB_FRAME_SIZE == NB_SUBFRAME_SIZE); |
851 | 103k | if (sub==0) |
852 | 25.8k | inBuf = st->winBuf; |
853 | 77.6k | else |
854 | 77.6k | inBuf = &in[((sub-1)*NB_SUBFRAME_SIZE)]; |
855 | 4.24M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
856 | 4.14M | sw[i] = inBuf[i]; |
857 | | |
858 | 103k | if (st->complexity==0) |
859 | 25.1k | response_bound >>= 1; |
860 | 103k | compute_impulse_response(interp_qlpc, bw_lpc1, bw_lpc2, syn_resp, response_bound, NB_ORDER, stack); |
861 | 606k | for (i=response_bound;i<NB_SUBFRAME_SIZE;i++) |
862 | 503k | syn_resp[i]=VERY_SMALL; |
863 | | |
864 | | /* Compute zero response of A(z/g1) / ( A(z/g2) * A(z) ) */ |
865 | 1.13M | for (i=0;i<NB_ORDER;i++) |
866 | 1.03M | mem[i]=SHL32(st->mem_sp[i],1); |
867 | 4.24M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
868 | 4.14M | exc[i] = VERY_SMALL; |
869 | | #ifdef SHORTCUTS2 |
870 | | iir_mem16(exc, interp_qlpc, exc, response_bound, NB_ORDER, mem, stack); |
871 | | for (i=0;i<NB_ORDER;i++) |
872 | | mem[i]=SHL32(st->mem_sw[i],1); |
873 | | filter10(exc, st->bw_lpc1, st->bw_lpc2, exc, response_bound, mem, stack); |
874 | | SPEEX_MEMSET(&exc[response_bound], 0, NB_SUBFRAME_SIZE-response_bound); |
875 | | #else |
876 | 103k | iir_mem16(exc, interp_qlpc, exc, NB_SUBFRAME_SIZE, NB_ORDER, mem, stack); |
877 | 1.13M | for (i=0;i<NB_ORDER;i++) |
878 | 1.03M | mem[i]=SHL32(st->mem_sw[i],1); |
879 | 103k | filter10(exc, bw_lpc1, bw_lpc2, exc, NB_SUBFRAME_SIZE, mem, stack); |
880 | 103k | #endif |
881 | | |
882 | | /* Compute weighted signal */ |
883 | 1.13M | for (i=0;i<NB_ORDER;i++) |
884 | 1.03M | mem[i]=st->mem_sw[i]; |
885 | 103k | filter10(sw, bw_lpc1, bw_lpc2, sw, NB_SUBFRAME_SIZE, mem, stack); |
886 | | |
887 | 103k | if (st->complexity==0) |
888 | 276k | for (i=0;i<NB_ORDER;i++) |
889 | 251k | st->mem_sw[i]=mem[i]; |
890 | | |
891 | | /* Compute target signal (saturation prevents overflows on clipped input speech) */ |
892 | 4.24M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
893 | 4.14M | target[i]=EXTRACT16(SATURATE(SUB32(sw[i],PSHR32(exc[i],1)),32767)); |
894 | | |
895 | 4.24M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
896 | 4.14M | exc[i] = inBuf[i]; |
897 | 103k | fir_mem16(exc, interp_qlpc, exc, NB_SUBFRAME_SIZE, NB_ORDER, st->mem_exc2, stack); |
898 | | /* If we have a long-term predictor (otherwise, something's wrong) */ |
899 | 103k | speex_assert (SUBMODE(ltp_quant)); |
900 | 103k | { |
901 | 103k | int pit_min, pit_max; |
902 | | /* Long-term prediction */ |
903 | 103k | if (SUBMODE(lbr_pitch) != -1) |
904 | 69.9k | { |
905 | | /* Low bit-rate pitch handling */ |
906 | 69.9k | int margin; |
907 | 69.9k | margin = SUBMODE(lbr_pitch); |
908 | 69.9k | if (margin) |
909 | 0 | { |
910 | 0 | if (ol_pitch < NB_PITCH_START+margin-1) |
911 | 0 | ol_pitch=NB_PITCH_START+margin-1; |
912 | 0 | if (ol_pitch > NB_PITCH_END-margin) |
913 | 0 | ol_pitch=NB_PITCH_END-margin; |
914 | 0 | pit_min = ol_pitch-margin+1; |
915 | 0 | pit_max = ol_pitch+margin; |
916 | 69.9k | } else { |
917 | 69.9k | pit_min=pit_max=ol_pitch; |
918 | 69.9k | } |
919 | 69.9k | } else { |
920 | 33.5k | pit_min = NB_PITCH_START; |
921 | 33.5k | pit_max = NB_PITCH_END; |
922 | 33.5k | } |
923 | | |
924 | | /* Force pitch to use only the current frame if needed */ |
925 | 103k | if (st->bounded_pitch && pit_max>offset) |
926 | 32.1k | pit_max=offset; |
927 | | |
928 | | /* Perform pitch search */ |
929 | 103k | pitch = SUBMODE(ltp_quant)(target, sw, interp_qlpc, bw_lpc1, bw_lpc2, |
930 | 103k | exc32, SUBMODE(ltp_params), pit_min, pit_max, ol_pitch_coef, |
931 | 103k | NB_ORDER, NB_SUBFRAME_SIZE, bits, stack, |
932 | 103k | exc, syn_resp, st->complexity, 0, st->plc_tuning, &st->cumul_gain); |
933 | | |
934 | 103k | st->pitch[sub]=pitch; |
935 | 103k | } |
936 | | /* Quantization of innovation */ |
937 | 103k | SPEEX_MEMSET(innov, 0, NB_SUBFRAME_SIZE); |
938 | | |
939 | | /* FIXME: Make sure this is safe from overflows (so far so good) */ |
940 | 4.24M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
941 | 4.14M | exc[i] = EXTRACT16(SUB32(EXTEND32(exc[i]), PSHR32(exc32[i],SIG_SHIFT-1))); |
942 | | |
943 | 103k | ener = SHL32(EXTEND32(compute_rms16(exc, NB_SUBFRAME_SIZE)),SIG_SHIFT); |
944 | | |
945 | | /*FIXME: Should use DIV32_16 and make sure result fits in 16 bits */ |
946 | | #ifdef FIXED_POINT |
947 | | { |
948 | 73.9k | spx_word32_t f = PDIV32(ener,PSHR32(ol_gain,SIG_SHIFT)); |
949 | 73.9k | if (f<=32767) |
950 | 68.2k | fine_gain = f; |
951 | 5.66k | else |
952 | 5.66k | fine_gain = 32767; |
953 | | } |
954 | | #else |
955 | 29.5k | fine_gain = PDIV32_16(ener,PSHR32(ol_gain,SIG_SHIFT)); |
956 | | #endif |
957 | | /* Calculate gain correction for the sub-frame (if any) */ |
958 | 103k | if (SUBMODE(have_subframe_gain)) |
959 | 33.5k | { |
960 | 33.5k | int qe; |
961 | 33.5k | if (SUBMODE(have_subframe_gain)==3) |
962 | 20.3k | { |
963 | 20.3k | qe = scal_quant(fine_gain, exc_gain_quant_scal3_bound, 8); |
964 | 20.3k | speex_bits_pack(bits, qe, 3); |
965 | 20.3k | ener=MULT16_32_Q14(exc_gain_quant_scal3[qe],ol_gain); |
966 | 20.3k | } else { |
967 | 13.2k | qe = scal_quant(fine_gain, exc_gain_quant_scal1_bound, 2); |
968 | 13.2k | speex_bits_pack(bits, qe, 1); |
969 | 13.2k | ener=MULT16_32_Q14(exc_gain_quant_scal1[qe],ol_gain); |
970 | 13.2k | } |
971 | 69.9k | } else { |
972 | 69.9k | ener=ol_gain; |
973 | 69.9k | } |
974 | | |
975 | | /*printf ("%f %f\n", ener, ol_gain);*/ |
976 | | |
977 | | /* Normalize innovation */ |
978 | 103k | signal_div(target, target, ener, NB_SUBFRAME_SIZE); |
979 | | |
980 | | /* Quantize innovation */ |
981 | 103k | speex_assert (SUBMODE(innovation_quant)); |
982 | 103k | { |
983 | | /* Codebook search */ |
984 | 103k | SUBMODE(innovation_quant)(target, interp_qlpc, bw_lpc1, bw_lpc2, |
985 | 103k | SUBMODE(innovation_params), NB_ORDER, NB_SUBFRAME_SIZE, |
986 | 103k | innov, syn_resp, bits, stack, st->complexity, SUBMODE(double_codebook)); |
987 | | |
988 | | /* De-normalize innovation and update excitation */ |
989 | 103k | signal_mul(innov, innov, ener, NB_SUBFRAME_SIZE); |
990 | | |
991 | | /* In some (rare) modes, we do a second search (more bits) to reduce noise even more */ |
992 | 103k | if (SUBMODE(double_codebook)) { |
993 | 3.28k | char *tmp_stack=stack; |
994 | 3.28k | VARDECL(spx_sig_t *innov2); |
995 | 3.28k | ALLOC(innov2, NB_SUBFRAME_SIZE, spx_sig_t); |
996 | 3.28k | SPEEX_MEMSET(innov2, 0, NB_SUBFRAME_SIZE); |
997 | 134k | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
998 | 131k | target[i]=MULT16_16_P13(QCONST16(2.2f,13), target[i]); |
999 | 3.28k | SUBMODE(innovation_quant)(target, interp_qlpc, bw_lpc1, bw_lpc2, |
1000 | 3.28k | SUBMODE(innovation_params), NB_ORDER, NB_SUBFRAME_SIZE, |
1001 | 3.28k | innov2, syn_resp, bits, stack, st->complexity, 0); |
1002 | 3.28k | signal_mul(innov2, innov2, MULT16_32_Q15(QCONST16(0.454545f,15),ener), NB_SUBFRAME_SIZE); |
1003 | 134k | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1004 | 131k | innov[i] = ADD32(innov[i],innov2[i]); |
1005 | 3.28k | stack = tmp_stack; |
1006 | 3.28k | } |
1007 | 4.24M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1008 | 4.14M | exc[i] = EXTRACT16(SATURATE32(PSHR32(ADD32(SHL32(exc32[i],1),innov[i]),SIG_SHIFT),32767)); |
1009 | 103k | if (st->innov_rms_save) |
1010 | 30.5k | st->innov_rms_save[sub] = compute_rms(innov, NB_SUBFRAME_SIZE); |
1011 | 103k | } |
1012 | | |
1013 | | /* Final signal synthesis from excitation */ |
1014 | 103k | iir_mem16(exc, interp_qlpc, sw, NB_SUBFRAME_SIZE, NB_ORDER, st->mem_sp, stack); |
1015 | | |
1016 | | /* Compute weighted signal again, from synthesized speech (not sure it's the right thing) */ |
1017 | 103k | if (st->complexity!=0) |
1018 | 78.3k | filter10(sw, bw_lpc1, bw_lpc2, sw, NB_SUBFRAME_SIZE, st->mem_sw, stack); |
1019 | | |
1020 | 103k | } |
1021 | | |
1022 | | /* Store the LSPs for interpolation in the next frame */ |
1023 | 25.8k | if (st->submodeID>=1) |
1024 | 25.8k | { |
1025 | 284k | for (i=0;i<NB_ORDER;i++) |
1026 | 258k | st->old_lsp[i] = lsp[i]; |
1027 | 284k | for (i=0;i<NB_ORDER;i++) |
1028 | 258k | st->old_qlsp[i] = qlsp[i]; |
1029 | 25.8k | } |
1030 | | |
1031 | | #ifdef VORBIS_PSYCHO |
1032 | | if (st->submodeID>=1) |
1033 | | SPEEX_COPY(st->old_curve, st->curve, 128); |
1034 | | #endif |
1035 | | |
1036 | 25.8k | if (st->submodeID==1) |
1037 | 13.2k | { |
1038 | 13.2k | #ifndef DISABLE_VBR |
1039 | 13.2k | if (st->dtx_count) |
1040 | 3.93k | speex_bits_pack(bits, 15, 4); |
1041 | 9.31k | else |
1042 | 9.31k | #endif |
1043 | 9.31k | speex_bits_pack(bits, 0, 4); |
1044 | 13.2k | } |
1045 | | |
1046 | | /* The next frame will not be the first (Duh!) */ |
1047 | 25.8k | st->first = 0; |
1048 | 25.8k | SPEEX_COPY(st->winBuf, in+2*NB_FRAME_SIZE-NB_WINDOW_SIZE, NB_WINDOW_SIZE-NB_FRAME_SIZE); |
1049 | | |
1050 | 25.8k | if (SUBMODE(innovation_quant) == noise_codebook_quant || st->submodeID==0) |
1051 | 13.2k | st->bounded_pitch = 1; |
1052 | 12.6k | else |
1053 | 12.6k | st->bounded_pitch = 0; |
1054 | | |
1055 | 25.8k | return 1; |
1056 | 28.2k | } Line | Count | Source | 407 | 8.59k | { | 408 | 8.59k | EncState *st; | 409 | 8.59k | int i, sub, roots; | 410 | 8.59k | int ol_pitch; | 411 | 8.59k | spx_word16_t ol_pitch_coef; | 412 | 8.59k | spx_word32_t ol_gain; | 413 | 8.59k | VARDECL(spx_word16_t *target); | 414 | 8.59k | VARDECL(spx_sig_t *innov); | 415 | 8.59k | VARDECL(spx_word32_t *exc32); | 416 | 8.59k | VARDECL(spx_mem_t *mem); | 417 | 8.59k | VARDECL(spx_coef_t *bw_lpc1); | 418 | 8.59k | VARDECL(spx_coef_t *bw_lpc2); | 419 | 8.59k | VARDECL(spx_coef_t *lpc); | 420 | 8.59k | VARDECL(spx_lsp_t *lsp); | 421 | 8.59k | VARDECL(spx_lsp_t *qlsp); | 422 | 8.59k | VARDECL(spx_lsp_t *interp_lsp); | 423 | 8.59k | VARDECL(spx_lsp_t *interp_qlsp); | 424 | 8.59k | VARDECL(spx_coef_t *interp_lpc); | 425 | 8.59k | VARDECL(spx_coef_t *interp_qlpc); | 426 | 8.59k | char *stack; | 427 | 8.59k | VARDECL(spx_word16_t *syn_resp); | 428 | | | 429 | 8.59k | spx_word32_t ener=0; | 430 | 8.59k | spx_word16_t fine_gain; | 431 | 8.59k | spx_word16_t *in = (spx_word16_t*)vin; | 432 | | | 433 | 8.59k | st=(EncState *)state; | 434 | 8.59k | stack=st->stack; | 435 | | | 436 | 8.59k | ALLOC(lpc, NB_ORDER, spx_coef_t); | 437 | 8.59k | ALLOC(bw_lpc1, NB_ORDER, spx_coef_t); | 438 | 8.59k | ALLOC(bw_lpc2, NB_ORDER, spx_coef_t); | 439 | 8.59k | ALLOC(lsp, NB_ORDER, spx_lsp_t); | 440 | 8.59k | ALLOC(qlsp, NB_ORDER, spx_lsp_t); | 441 | 8.59k | ALLOC(interp_lsp, NB_ORDER, spx_lsp_t); | 442 | 8.59k | ALLOC(interp_qlsp, NB_ORDER, spx_lsp_t); | 443 | 8.59k | ALLOC(interp_lpc, NB_ORDER, spx_coef_t); | 444 | 8.59k | ALLOC(interp_qlpc, NB_ORDER, spx_coef_t); | 445 | | | 446 | 8.59k | st->exc = st->excBuf + NB_PITCH_END + 2; | 447 | 8.59k | st->sw = st->swBuf + NB_PITCH_END + 2; | 448 | | /* Move signals 1 frame towards the past */ | 449 | 8.59k | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, NB_PITCH_END+2); | 450 | 8.59k | SPEEX_MOVE(st->swBuf, st->swBuf+NB_FRAME_SIZE, NB_PITCH_END+2); | 451 | | | 452 | 8.59k | if (st->highpass_enabled) | 453 | 4.82k | highpass(in, in, NB_FRAME_SIZE, (st->isWideband?HIGHPASS_WIDEBAND:HIGHPASS_NARROWBAND)|HIGHPASS_INPUT, st->mem_hp); | 454 | | | 455 | 8.59k | { | 456 | 8.59k | VARDECL(spx_word16_t *w_sig); | 457 | 8.59k | VARDECL(spx_word16_t *autocorr); | 458 | 8.59k | ALLOC(w_sig, NB_WINDOW_SIZE, spx_word16_t); | 459 | 8.59k | ALLOC(autocorr, NB_ORDER+1, spx_word16_t); | 460 | | /* Window for analysis */ | 461 | 352k | for (i=0;i<NB_WINDOW_SIZE-NB_FRAME_SIZE;i++) | 462 | 343k | w_sig[i] = MULT16_16_Q15(st->winBuf[i],st->window[i]); | 463 | 1.38M | for (;i<NB_WINDOW_SIZE;i++) | 464 | 1.37M | w_sig[i] = MULT16_16_Q15(in[i-NB_WINDOW_SIZE+NB_FRAME_SIZE],st->window[i]); | 465 | | /* Compute auto-correlation */ | 466 | 8.59k | _spx_autocorr(w_sig, autocorr, NB_ORDER+1, NB_WINDOW_SIZE); | 467 | 8.59k | autocorr[0] = ADD16(autocorr[0],MULT16_16_Q15(autocorr[0],st->lpc_floor)); /* Noise floor in auto-correlation domain */ | 468 | | | 469 | | /* Lag windowing: equivalent to filtering in the power-spectrum domain */ | 470 | 103k | for (i=0;i<NB_ORDER+1;i++) | 471 | 94.5k | autocorr[i] = MULT16_16_Q15(autocorr[i],st->lagWindow[i]); | 472 | 8.59k | autocorr[0] = ADD16(autocorr[0],1); | 473 | | | 474 | | /* Levinson-Durbin */ | 475 | 8.59k | _spx_lpc(lpc, autocorr, NB_ORDER); | 476 | | /* LPC to LSPs (x-domain) transform */ | 477 | 8.59k | roots=lpc_to_lsp (lpc, NB_ORDER, lsp, 10, LSP_DELTA1, stack); | 478 | | /* Check if we found all the roots */ | 479 | 8.59k | if (roots!=NB_ORDER) | 480 | 4 | { | 481 | | /*If we can't find all LSP's, do some damage control and use previous filter*/ | 482 | 44 | for (i=0;i<NB_ORDER;i++) | 483 | 40 | { | 484 | 40 | lsp[i]=st->old_lsp[i]; | 485 | 40 | } | 486 | 4 | } | 487 | 8.59k | } | 488 | | | 489 | | | 490 | | | 491 | | | 492 | | /* Whole frame analysis (open-loop estimation of pitch and excitation gain) */ | 493 | 8.59k | { | 494 | 8.59k | int diff = NB_WINDOW_SIZE-NB_FRAME_SIZE; | 495 | 8.59k | if (st->first) | 496 | 23.8k | for (i=0;i<NB_ORDER;i++) | 497 | 21.6k | interp_lsp[i] = lsp[i]; | 498 | 6.42k | else | 499 | 6.42k | lsp_interpolate(st->old_lsp, lsp, interp_lsp, NB_ORDER, NB_NB_SUBFRAMES, NB_NB_SUBFRAMES<<1, LSP_MARGIN); | 500 | | | 501 | | /* Compute interpolated LPCs (unquantized) for whole frame*/ | 502 | 8.59k | lsp_to_lpc(interp_lsp, interp_lpc, NB_ORDER,stack); | 503 | | | 504 | | | 505 | | /*Open-loop pitch*/ | 506 | 8.59k | if (!st->submodes[st->submodeID] || (st->complexity>2 && SUBMODE(have_subframe_gain)<3) || SUBMODE(forced_pitch_gain) || SUBMODE(lbr_pitch) != -1 | 507 | 8.59k | #ifndef DISABLE_VBR | 508 | 8.59k | || st->vbr_enabled || st->vad_enabled | 509 | 8.59k | #endif | 510 | 8.59k | ) | 511 | 7.98k | { | 512 | 7.98k | int nol_pitch[6]; | 513 | 7.98k | spx_word16_t nol_pitch_coef[6]; | 514 | | | 515 | 7.98k | bw_lpc(QCONST16(0.9,15), interp_lpc, bw_lpc1, NB_ORDER); | 516 | 7.98k | bw_lpc(QCONST16(0.55,15), interp_lpc, bw_lpc2, NB_ORDER); | 517 | | | 518 | 7.98k | SPEEX_COPY(st->sw, st->winBuf, diff); | 519 | 7.98k | SPEEX_COPY(st->sw+diff, in, NB_FRAME_SIZE-diff); | 520 | 7.98k | filter10(st->sw, bw_lpc1, bw_lpc2, st->sw, NB_FRAME_SIZE, st->mem_sw_whole, stack); | 521 | | | 522 | 7.98k | open_loop_nbest_pitch(st->sw, NB_PITCH_START, NB_PITCH_END, NB_FRAME_SIZE, | 523 | 7.98k | nol_pitch, nol_pitch_coef, 6, stack); | 524 | 7.98k | ol_pitch=nol_pitch[0]; | 525 | 7.98k | ol_pitch_coef = nol_pitch_coef[0]; | 526 | | /*Try to remove pitch multiples*/ | 527 | 47.9k | for (i=1;i<6;i++) | 528 | 39.9k | { | 529 | | #ifdef FIXED_POINT | 530 | | if ((nol_pitch_coef[i]>MULT16_16_Q15(nol_pitch_coef[0],27853)) && | 531 | | #else | 532 | 39.9k | if ((nol_pitch_coef[i]>.85*nol_pitch_coef[0]) && | 533 | 39.9k | #endif | 534 | 39.9k | (ABS(2*nol_pitch[i]-ol_pitch)<=2 || ABS(3*nol_pitch[i]-ol_pitch)<=3 || | 535 | 11.1k | ABS(4*nol_pitch[i]-ol_pitch)<=4 || ABS(5*nol_pitch[i]-ol_pitch)<=5)) | 536 | 245 | { | 537 | | /*ol_pitch_coef=nol_pitch_coef[i];*/ | 538 | 245 | ol_pitch = nol_pitch[i]; | 539 | 245 | } | 540 | 39.9k | } | 541 | | /*if (ol_pitch>50) | 542 | | ol_pitch/=2;*/ | 543 | | /*ol_pitch_coef = sqrt(ol_pitch_coef);*/ | 544 | | | 545 | 7.98k | } else { | 546 | 611 | ol_pitch=0; | 547 | 611 | ol_pitch_coef=0; | 548 | 611 | } | 549 | | | 550 | | /*Compute "real" excitation*/ | 551 | | /*SPEEX_COPY(st->exc, st->winBuf, diff); | 552 | | SPEEX_COPY(st->exc+diff, in, NB_FRAME_SIZE-diff);*/ | 553 | 8.59k | fir_mem16(st->winBuf, interp_lpc, st->exc, diff, NB_ORDER, st->mem_exc, stack); | 554 | 8.59k | fir_mem16(in, interp_lpc, st->exc+diff, NB_FRAME_SIZE-diff, NB_ORDER, st->mem_exc, stack); | 555 | | | 556 | | /* Compute open-loop excitation gain */ | 557 | 8.59k | { | 558 | 8.59k | spx_word16_t g = compute_rms16(st->exc, NB_FRAME_SIZE); | 559 | 8.59k | if (st->submodeID!=1 && ol_pitch>0) | 560 | 4.09k | ol_gain = MULT16_16(g, MULT16_16_Q14(QCONST16(1.1,14), | 561 | 8.59k | spx_sqrt(QCONST32(1.,28)-MULT16_32_Q15(QCONST16(.8,15),SHL32(MULT16_16(ol_pitch_coef,ol_pitch_coef),16))))); | 562 | 4.50k | else | 563 | 4.50k | ol_gain = SHL32(EXTEND32(g),SIG_SHIFT); | 564 | 8.59k | } | 565 | 8.59k | } | 566 | | | 567 | | #ifdef VORBIS_PSYCHO | 568 | | SPEEX_MOVE(st->psy_window, st->psy_window+NB_FRAME_SIZE, 256-NB_FRAME_SIZE); | 569 | | SPEEX_COPY(&st->psy_window[256-NB_FRAME_SIZE], in, NB_FRAME_SIZE); | 570 | | compute_curve(st->psy, st->psy_window, st->curve); | 571 | | /*print_vec(st->curve, 128, "curve");*/ | 572 | | if (st->first) | 573 | | SPEEX_COPY(st->old_curve, st->curve, 128); | 574 | | #endif | 575 | | | 576 | | /*VBR stuff*/ | 577 | 8.59k | #ifndef DISABLE_VBR | 578 | 8.59k | if (st->vbr_enabled||st->vad_enabled) | 579 | 6.94k | { | 580 | 6.94k | float lsp_dist=0; | 581 | 76.3k | for (i=0;i<NB_ORDER;i++) | 582 | 69.4k | lsp_dist += (st->old_lsp[i] - lsp[i])*(st->old_lsp[i] - lsp[i]); | 583 | 6.94k | lsp_dist /= LSP_SCALING*LSP_SCALING; | 584 | | | 585 | 6.94k | if (st->abr_enabled) | 586 | 1.87k | { | 587 | 1.87k | float qual_change=0; | 588 | 1.87k | if (st->abr_drift2 * st->abr_drift > 0) | 589 | 1.68k | { | 590 | | /* Only adapt if long-term and short-term drift are the same sign */ | 591 | 1.68k | qual_change = -.00001*st->abr_drift/(1+st->abr_count); | 592 | 1.68k | if (qual_change>.05) | 593 | 0 | qual_change=.05; | 594 | 1.68k | if (qual_change<-.05) | 595 | 332 | qual_change=-.05; | 596 | 1.68k | } | 597 | 1.87k | st->vbr_quality += qual_change; | 598 | 1.87k | if (st->vbr_quality>10) | 599 | 0 | st->vbr_quality=10; | 600 | 1.87k | if (st->vbr_quality<0) | 601 | 1.68k | st->vbr_quality=0; | 602 | 1.87k | } | 603 | | | 604 | 6.94k | st->relative_quality = vbr_analysis(&st->vbr, in, NB_FRAME_SIZE, ol_pitch, GAIN_SCALING_1*ol_pitch_coef); | 605 | | /*if (delta_qual<0)*/ | 606 | | /* delta_qual*=.1*(3+st->vbr_quality);*/ | 607 | 6.94k | if (st->vbr_enabled) | 608 | 5.51k | { | 609 | 5.51k | spx_int32_t mode; | 610 | 5.51k | int choice=0; | 611 | 5.51k | float min_diff=100; | 612 | 5.51k | mode = 8; | 613 | 49.6k | while (mode) | 614 | 44.0k | { | 615 | 44.0k | int v1; | 616 | 44.0k | float thresh; | 617 | 44.0k | v1=(int)floor(st->vbr_quality); | 618 | 44.0k | if (v1==10) | 619 | 4.90k | thresh = vbr_nb_thresh[mode][v1]; | 620 | 39.1k | else | 621 | 39.1k | thresh = (st->vbr_quality-v1)*vbr_nb_thresh[mode][v1+1] + (1+v1-st->vbr_quality)*vbr_nb_thresh[mode][v1]; | 622 | 44.0k | if (st->relative_quality > thresh && | 623 | 44.0k | st->relative_quality-thresh<min_diff) | 624 | 4.71k | { | 625 | 4.71k | choice = mode; | 626 | 4.71k | min_diff = st->relative_quality-thresh; | 627 | 4.71k | } | 628 | 44.0k | mode--; | 629 | 44.0k | } | 630 | 5.51k | mode=choice; | 631 | 5.51k | if (mode==0) | 632 | 2.13k | { | 633 | 2.13k | if (st->dtx_count==0 || lsp_dist>.05 || !st->dtx_enabled || st->dtx_count>20) | 634 | 1.18k | { | 635 | 1.18k | mode=1; | 636 | 1.18k | st->dtx_count=1; | 637 | 1.18k | } else { | 638 | 945 | mode=0; | 639 | 945 | st->dtx_count++; | 640 | 945 | } | 641 | 3.38k | } else { | 642 | 3.38k | st->dtx_count=0; | 643 | 3.38k | } | 644 | | | 645 | 5.51k | speex_encoder_ctl(state, SPEEX_SET_MODE, &mode); | 646 | 5.51k | if (st->vbr_max>0) | 647 | 4.14k | { | 648 | 4.14k | spx_int32_t rate; | 649 | 4.14k | speex_encoder_ctl(state, SPEEX_GET_BITRATE, &rate); | 650 | 4.14k | if (rate > st->vbr_max) | 651 | 1.23k | { | 652 | 1.23k | rate = st->vbr_max; | 653 | 1.23k | speex_encoder_ctl(state, SPEEX_SET_BITRATE, &rate); | 654 | 1.23k | } | 655 | 4.14k | } | 656 | | | 657 | 5.51k | if (st->abr_enabled) | 658 | 1.87k | { | 659 | 1.87k | spx_int32_t bitrate; | 660 | 1.87k | speex_encoder_ctl(state, SPEEX_GET_BITRATE, &bitrate); | 661 | 1.87k | st->abr_drift+=(bitrate-st->abr_enabled); | 662 | 1.87k | st->abr_drift2 = .95*st->abr_drift2 + .05*(bitrate-st->abr_enabled); | 663 | 1.87k | st->abr_count += 1.0; | 664 | 1.87k | } | 665 | | | 666 | 5.51k | } else { | 667 | | /*VAD only case*/ | 668 | 1.43k | int mode; | 669 | 1.43k | if (st->relative_quality<2) | 670 | 732 | { | 671 | 732 | if (st->dtx_count==0 || lsp_dist>.05 || !st->dtx_enabled || st->dtx_count>20) | 672 | 415 | { | 673 | 415 | st->dtx_count=1; | 674 | 415 | mode=1; | 675 | 415 | } else { | 676 | 317 | mode=0; | 677 | 317 | st->dtx_count++; | 678 | 317 | } | 679 | 732 | } else { | 680 | 698 | st->dtx_count = 0; | 681 | 698 | mode=st->submodeSelect; | 682 | 698 | } | 683 | | /*speex_encoder_ctl(state, SPEEX_SET_MODE, &mode);*/ | 684 | 1.43k | st->submodeID=mode; | 685 | 1.43k | } | 686 | 6.94k | } else { | 687 | 1.65k | st->relative_quality = -1; | 688 | 1.65k | } | 689 | 8.59k | #endif /* #ifndef DISABLE_VBR */ | 690 | | | 691 | 8.59k | if (st->encode_submode) | 692 | 8.59k | { | 693 | | /* First, transmit a zero for narrowband */ | 694 | 8.59k | speex_bits_pack(bits, 0, 1); | 695 | | | 696 | | /* Transmit the sub-mode we use for this frame */ | 697 | 8.59k | speex_bits_pack(bits, st->submodeID, NB_SUBMODE_BITS); | 698 | | | 699 | 8.59k | } | 700 | | | 701 | | /* If null mode (no transmission), just set a couple things to zero*/ | 702 | 8.59k | if (st->submodes[st->submodeID] == NULL) | 703 | 1.19k | { | 704 | 192k | for (i=0;i<NB_FRAME_SIZE;i++) | 705 | 191k | st->exc[i]=st->sw[i]=VERY_SMALL; | 706 | | | 707 | 13.1k | for (i=0;i<NB_ORDER;i++) | 708 | 11.9k | st->mem_sw[i]=0; | 709 | 1.19k | st->first=1; | 710 | 1.19k | st->bounded_pitch = 1; | 711 | | | 712 | 1.19k | SPEEX_COPY(st->winBuf, in+2*NB_FRAME_SIZE-NB_WINDOW_SIZE, NB_WINDOW_SIZE-NB_FRAME_SIZE); | 713 | | | 714 | | /* Clear memory (no need to really compute it) */ | 715 | 13.1k | for (i=0;i<NB_ORDER;i++) | 716 | 11.9k | st->mem_sp[i] = 0; | 717 | 1.19k | return 0; | 718 | | | 719 | 1.19k | } | 720 | | | 721 | | /* LSP Quantization */ | 722 | 7.39k | if (st->first) | 723 | 1.27k | { | 724 | 14.0k | for (i=0;i<NB_ORDER;i++) | 725 | 12.7k | st->old_lsp[i] = lsp[i]; | 726 | 1.27k | } | 727 | | | 728 | | | 729 | | /*Quantize LSPs*/ | 730 | 7.39k | #if 1 /*0 for unquantized*/ | 731 | 7.39k | SUBMODE(lsp_quant)(lsp, qlsp, NB_ORDER, bits); | 732 | | #else | 733 | | for (i=0;i<NB_ORDER;i++) | 734 | | qlsp[i]=lsp[i]; | 735 | | #endif | 736 | | | 737 | | /*If we use low bit-rate pitch mode, transmit open-loop pitch*/ | 738 | 7.39k | if (SUBMODE(lbr_pitch)!=-1) | 739 | 5.18k | { | 740 | 5.18k | speex_bits_pack(bits, ol_pitch-NB_PITCH_START, 7); | 741 | 5.18k | } | 742 | | | 743 | 7.39k | if (SUBMODE(forced_pitch_gain)) | 744 | 4.77k | { | 745 | 4.77k | int quant; | 746 | | /* This just damps the pitch a bit, because it tends to be too aggressive when forced */ | 747 | 4.77k | ol_pitch_coef = MULT16_16_Q15(QCONST16(.9,15), ol_pitch_coef); | 748 | | #ifdef FIXED_POINT | 749 | | quant = PSHR16(MULT16_16_16(15, ol_pitch_coef),GAIN_SHIFT); | 750 | | #else | 751 | 4.77k | quant = (int)floor(.5+15*ol_pitch_coef*GAIN_SCALING_1); | 752 | 4.77k | #endif | 753 | 4.77k | if (quant>15) | 754 | 39 | quant=15; | 755 | 4.77k | if (quant<0) | 756 | 0 | quant=0; | 757 | 4.77k | speex_bits_pack(bits, quant, 4); | 758 | 4.77k | ol_pitch_coef=MULT16_16_P15(QCONST16(0.066667,15),SHL16(quant,GAIN_SHIFT)); | 759 | 4.77k | } | 760 | | | 761 | | | 762 | | /*Quantize and transmit open-loop excitation gain*/ | 763 | | #ifdef FIXED_POINT | 764 | | { | 765 | | int qe = scal_quant32(ol_gain, ol_gain_table, 32); | 766 | | /*ol_gain = exp(qe/3.5)*SIG_SCALING;*/ | 767 | | ol_gain = MULT16_32_Q15(28406,ol_gain_table[qe]); | 768 | | speex_bits_pack(bits, qe, 5); | 769 | | } | 770 | | #else | 771 | 7.39k | { | 772 | 7.39k | int qe = (int)(floor(.5+3.5*log(ol_gain*1.0/SIG_SCALING))); | 773 | 7.39k | if (qe<0) | 774 | 4.67k | qe=0; | 775 | 7.39k | if (qe>31) | 776 | 422 | qe=31; | 777 | 7.39k | ol_gain = exp(qe/3.5)*SIG_SCALING; | 778 | 7.39k | speex_bits_pack(bits, qe, 5); | 779 | 7.39k | } | 780 | 7.39k | #endif | 781 | | | 782 | | | 783 | | | 784 | | /* Special case for first frame */ | 785 | 7.39k | if (st->first) | 786 | 1.27k | { | 787 | 14.0k | for (i=0;i<NB_ORDER;i++) | 788 | 12.7k | st->old_qlsp[i] = qlsp[i]; | 789 | 1.27k | } | 790 | | | 791 | | /* Target signal */ | 792 | 7.39k | ALLOC(target, NB_SUBFRAME_SIZE, spx_word16_t); | 793 | 7.39k | ALLOC(innov, NB_SUBFRAME_SIZE, spx_sig_t); | 794 | 7.39k | ALLOC(exc32, NB_SUBFRAME_SIZE, spx_word32_t); | 795 | 7.39k | ALLOC(syn_resp, NB_SUBFRAME_SIZE, spx_word16_t); | 796 | 7.39k | ALLOC(mem, NB_ORDER, spx_mem_t); | 797 | | | 798 | | /* Loop on sub-frames */ | 799 | 36.9k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) | 800 | 29.5k | { | 801 | 29.5k | int offset; | 802 | 29.5k | spx_word16_t *sw; | 803 | 29.5k | spx_word16_t *exc, *inBuf; | 804 | 29.5k | int pitch; | 805 | 29.5k | int response_bound = NB_SUBFRAME_SIZE; | 806 | | | 807 | | /* Offset relative to start of frame */ | 808 | 29.5k | offset = NB_SUBFRAME_SIZE*sub; | 809 | | /* Excitation */ | 810 | 29.5k | exc=st->exc+offset; | 811 | | /* Weighted signal */ | 812 | 29.5k | sw=st->sw+offset; | 813 | | | 814 | | /* LSP interpolation (quantized and unquantized) */ | 815 | 29.5k | lsp_interpolate(st->old_lsp, lsp, interp_lsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); | 816 | 29.5k | lsp_interpolate(st->old_qlsp, qlsp, interp_qlsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); | 817 | | | 818 | | /* Compute interpolated LPCs (quantized and unquantized) */ | 819 | 29.5k | lsp_to_lpc(interp_lsp, interp_lpc, NB_ORDER,stack); | 820 | | | 821 | 29.5k | lsp_to_lpc(interp_qlsp, interp_qlpc, NB_ORDER, stack); | 822 | | | 823 | | /* Compute analysis filter gain at w=pi (for use in SB-CELP) */ | 824 | 29.5k | { | 825 | 29.5k | spx_word32_t pi_g=LPC_SCALING; | 826 | 177k | for (i=0;i<NB_ORDER;i+=2) | 827 | 147k | { | 828 | | /*pi_g += -st->interp_qlpc[i] + st->interp_qlpc[i+1];*/ | 829 | 147k | pi_g = ADD32(pi_g, SUB32(EXTEND32(interp_qlpc[i+1]),EXTEND32(interp_qlpc[i]))); | 830 | 147k | } | 831 | 29.5k | st->pi_gain[sub] = pi_g; | 832 | 29.5k | } | 833 | | | 834 | | #ifdef VORBIS_PSYCHO | 835 | | { | 836 | | float curr_curve[128]; | 837 | | float fact = ((float)sub+1.0f)/NB_NB_SUBFRAMES; | 838 | | for (i=0;i<128;i++) | 839 | | curr_curve[i] = (1.0f-fact)*st->old_curve[i] + fact*st->curve[i]; | 840 | | curve_to_lpc(st->psy, curr_curve, bw_lpc1, bw_lpc2, 10); | 841 | | } | 842 | | #else | 843 | | /* Compute bandwidth-expanded (unquantized) LPCs for perceptual weighting */ | 844 | 29.5k | bw_lpc(st->gamma1, interp_lpc, bw_lpc1, NB_ORDER); | 845 | 29.5k | bw_lpc(st->gamma2, interp_lpc, bw_lpc2, NB_ORDER); | 846 | | /*print_vec(st->bw_lpc1, 10, "bw_lpc");*/ | 847 | 29.5k | #endif | 848 | | | 849 | | /*FIXME: This will break if we change the window size */ | 850 | 29.5k | speex_assert(NB_WINDOW_SIZE-NB_FRAME_SIZE == NB_SUBFRAME_SIZE); | 851 | 29.5k | if (sub==0) | 852 | 7.39k | inBuf = st->winBuf; | 853 | 22.1k | else | 854 | 22.1k | inBuf = &in[((sub-1)*NB_SUBFRAME_SIZE)]; | 855 | 1.21M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 856 | 1.18M | sw[i] = inBuf[i]; | 857 | | | 858 | 29.5k | if (st->complexity==0) | 859 | 6.75k | response_bound >>= 1; | 860 | 29.5k | compute_impulse_response(interp_qlpc, bw_lpc1, bw_lpc2, syn_resp, response_bound, NB_ORDER, stack); | 861 | 164k | for (i=response_bound;i<NB_SUBFRAME_SIZE;i++) | 862 | 135k | syn_resp[i]=VERY_SMALL; | 863 | | | 864 | | /* Compute zero response of A(z/g1) / ( A(z/g2) * A(z) ) */ | 865 | 325k | for (i=0;i<NB_ORDER;i++) | 866 | 295k | mem[i]=SHL32(st->mem_sp[i],1); | 867 | 1.21M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 868 | 1.18M | exc[i] = VERY_SMALL; | 869 | | #ifdef SHORTCUTS2 | 870 | | iir_mem16(exc, interp_qlpc, exc, response_bound, NB_ORDER, mem, stack); | 871 | | for (i=0;i<NB_ORDER;i++) | 872 | | mem[i]=SHL32(st->mem_sw[i],1); | 873 | | filter10(exc, st->bw_lpc1, st->bw_lpc2, exc, response_bound, mem, stack); | 874 | | SPEEX_MEMSET(&exc[response_bound], 0, NB_SUBFRAME_SIZE-response_bound); | 875 | | #else | 876 | 29.5k | iir_mem16(exc, interp_qlpc, exc, NB_SUBFRAME_SIZE, NB_ORDER, mem, stack); | 877 | 325k | for (i=0;i<NB_ORDER;i++) | 878 | 295k | mem[i]=SHL32(st->mem_sw[i],1); | 879 | 29.5k | filter10(exc, bw_lpc1, bw_lpc2, exc, NB_SUBFRAME_SIZE, mem, stack); | 880 | 29.5k | #endif | 881 | | | 882 | | /* Compute weighted signal */ | 883 | 325k | for (i=0;i<NB_ORDER;i++) | 884 | 295k | mem[i]=st->mem_sw[i]; | 885 | 29.5k | filter10(sw, bw_lpc1, bw_lpc2, sw, NB_SUBFRAME_SIZE, mem, stack); | 886 | | | 887 | 29.5k | if (st->complexity==0) | 888 | 74.3k | for (i=0;i<NB_ORDER;i++) | 889 | 67.5k | st->mem_sw[i]=mem[i]; | 890 | | | 891 | | /* Compute target signal (saturation prevents overflows on clipped input speech) */ | 892 | 1.21M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 893 | 1.18M | target[i]=EXTRACT16(SATURATE(SUB32(sw[i],PSHR32(exc[i],1)),32767)); | 894 | | | 895 | 1.21M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 896 | 1.18M | exc[i] = inBuf[i]; | 897 | 29.5k | fir_mem16(exc, interp_qlpc, exc, NB_SUBFRAME_SIZE, NB_ORDER, st->mem_exc2, stack); | 898 | | /* If we have a long-term predictor (otherwise, something's wrong) */ | 899 | 29.5k | speex_assert (SUBMODE(ltp_quant)); | 900 | 29.5k | { | 901 | 29.5k | int pit_min, pit_max; | 902 | | /* Long-term prediction */ | 903 | 29.5k | if (SUBMODE(lbr_pitch) != -1) | 904 | 20.7k | { | 905 | | /* Low bit-rate pitch handling */ | 906 | 20.7k | int margin; | 907 | 20.7k | margin = SUBMODE(lbr_pitch); | 908 | 20.7k | if (margin) | 909 | 0 | { | 910 | 0 | if (ol_pitch < NB_PITCH_START+margin-1) | 911 | 0 | ol_pitch=NB_PITCH_START+margin-1; | 912 | 0 | if (ol_pitch > NB_PITCH_END-margin) | 913 | 0 | ol_pitch=NB_PITCH_END-margin; | 914 | 0 | pit_min = ol_pitch-margin+1; | 915 | 0 | pit_max = ol_pitch+margin; | 916 | 20.7k | } else { | 917 | 20.7k | pit_min=pit_max=ol_pitch; | 918 | 20.7k | } | 919 | 20.7k | } else { | 920 | 8.87k | pit_min = NB_PITCH_START; | 921 | 8.87k | pit_max = NB_PITCH_END; | 922 | 8.87k | } | 923 | | | 924 | | /* Force pitch to use only the current frame if needed */ | 925 | 29.5k | if (st->bounded_pitch && pit_max>offset) | 926 | 8.92k | pit_max=offset; | 927 | | | 928 | | /* Perform pitch search */ | 929 | 29.5k | pitch = SUBMODE(ltp_quant)(target, sw, interp_qlpc, bw_lpc1, bw_lpc2, | 930 | 29.5k | exc32, SUBMODE(ltp_params), pit_min, pit_max, ol_pitch_coef, | 931 | 29.5k | NB_ORDER, NB_SUBFRAME_SIZE, bits, stack, | 932 | 29.5k | exc, syn_resp, st->complexity, 0, st->plc_tuning, &st->cumul_gain); | 933 | | | 934 | 29.5k | st->pitch[sub]=pitch; | 935 | 29.5k | } | 936 | | /* Quantization of innovation */ | 937 | 29.5k | SPEEX_MEMSET(innov, 0, NB_SUBFRAME_SIZE); | 938 | | | 939 | | /* FIXME: Make sure this is safe from overflows (so far so good) */ | 940 | 1.21M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 941 | 1.18M | exc[i] = EXTRACT16(SUB32(EXTEND32(exc[i]), PSHR32(exc32[i],SIG_SHIFT-1))); | 942 | | | 943 | 29.5k | ener = SHL32(EXTEND32(compute_rms16(exc, NB_SUBFRAME_SIZE)),SIG_SHIFT); | 944 | | | 945 | | /*FIXME: Should use DIV32_16 and make sure result fits in 16 bits */ | 946 | | #ifdef FIXED_POINT | 947 | | { | 948 | | spx_word32_t f = PDIV32(ener,PSHR32(ol_gain,SIG_SHIFT)); | 949 | | if (f<=32767) | 950 | | fine_gain = f; | 951 | | else | 952 | | fine_gain = 32767; | 953 | | } | 954 | | #else | 955 | 29.5k | fine_gain = PDIV32_16(ener,PSHR32(ol_gain,SIG_SHIFT)); | 956 | 29.5k | #endif | 957 | | /* Calculate gain correction for the sub-frame (if any) */ | 958 | 29.5k | if (SUBMODE(have_subframe_gain)) | 959 | 8.87k | { | 960 | 8.87k | int qe; | 961 | 8.87k | if (SUBMODE(have_subframe_gain)==3) | 962 | 5.43k | { | 963 | 5.43k | qe = scal_quant(fine_gain, exc_gain_quant_scal3_bound, 8); | 964 | 5.43k | speex_bits_pack(bits, qe, 3); | 965 | 5.43k | ener=MULT16_32_Q14(exc_gain_quant_scal3[qe],ol_gain); | 966 | 5.43k | } else { | 967 | 3.44k | qe = scal_quant(fine_gain, exc_gain_quant_scal1_bound, 2); | 968 | 3.44k | speex_bits_pack(bits, qe, 1); | 969 | 3.44k | ener=MULT16_32_Q14(exc_gain_quant_scal1[qe],ol_gain); | 970 | 3.44k | } | 971 | 20.7k | } else { | 972 | 20.7k | ener=ol_gain; | 973 | 20.7k | } | 974 | | | 975 | | /*printf ("%f %f\n", ener, ol_gain);*/ | 976 | | | 977 | | /* Normalize innovation */ | 978 | 29.5k | signal_div(target, target, ener, NB_SUBFRAME_SIZE); | 979 | | | 980 | | /* Quantize innovation */ | 981 | 29.5k | speex_assert (SUBMODE(innovation_quant)); | 982 | 29.5k | { | 983 | | /* Codebook search */ | 984 | 29.5k | SUBMODE(innovation_quant)(target, interp_qlpc, bw_lpc1, bw_lpc2, | 985 | 29.5k | SUBMODE(innovation_params), NB_ORDER, NB_SUBFRAME_SIZE, | 986 | 29.5k | innov, syn_resp, bits, stack, st->complexity, SUBMODE(double_codebook)); | 987 | | | 988 | | /* De-normalize innovation and update excitation */ | 989 | 29.5k | signal_mul(innov, innov, ener, NB_SUBFRAME_SIZE); | 990 | | | 991 | | /* In some (rare) modes, we do a second search (more bits) to reduce noise even more */ | 992 | 29.5k | if (SUBMODE(double_codebook)) { | 993 | 1.18k | char *tmp_stack=stack; | 994 | 1.18k | VARDECL(spx_sig_t *innov2); | 995 | 1.18k | ALLOC(innov2, NB_SUBFRAME_SIZE, spx_sig_t); | 996 | 1.18k | SPEEX_MEMSET(innov2, 0, NB_SUBFRAME_SIZE); | 997 | 48.7k | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 998 | 47.5k | target[i]=MULT16_16_P13(QCONST16(2.2f,13), target[i]); | 999 | 1.18k | SUBMODE(innovation_quant)(target, interp_qlpc, bw_lpc1, bw_lpc2, | 1000 | 1.18k | SUBMODE(innovation_params), NB_ORDER, NB_SUBFRAME_SIZE, | 1001 | 1.18k | innov2, syn_resp, bits, stack, st->complexity, 0); | 1002 | 1.18k | signal_mul(innov2, innov2, MULT16_32_Q15(QCONST16(0.454545f,15),ener), NB_SUBFRAME_SIZE); | 1003 | 48.7k | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1004 | 47.5k | innov[i] = ADD32(innov[i],innov2[i]); | 1005 | 1.18k | stack = tmp_stack; | 1006 | 1.18k | } | 1007 | 1.21M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1008 | 1.18M | exc[i] = EXTRACT16(SATURATE32(PSHR32(ADD32(SHL32(exc32[i],1),innov[i]),SIG_SHIFT),32767)); | 1009 | 29.5k | if (st->innov_rms_save) | 1010 | 10.0k | st->innov_rms_save[sub] = compute_rms(innov, NB_SUBFRAME_SIZE); | 1011 | 29.5k | } | 1012 | | | 1013 | | /* Final signal synthesis from excitation */ | 1014 | 29.5k | iir_mem16(exc, interp_qlpc, sw, NB_SUBFRAME_SIZE, NB_ORDER, st->mem_sp, stack); | 1015 | | | 1016 | | /* Compute weighted signal again, from synthesized speech (not sure it's the right thing) */ | 1017 | 29.5k | if (st->complexity!=0) | 1018 | 22.8k | filter10(sw, bw_lpc1, bw_lpc2, sw, NB_SUBFRAME_SIZE, st->mem_sw, stack); | 1019 | | | 1020 | 29.5k | } | 1021 | | | 1022 | | /* Store the LSPs for interpolation in the next frame */ | 1023 | 7.39k | if (st->submodeID>=1) | 1024 | 7.39k | { | 1025 | 81.3k | for (i=0;i<NB_ORDER;i++) | 1026 | 73.9k | st->old_lsp[i] = lsp[i]; | 1027 | 81.3k | for (i=0;i<NB_ORDER;i++) | 1028 | 73.9k | st->old_qlsp[i] = qlsp[i]; | 1029 | 7.39k | } | 1030 | | | 1031 | | #ifdef VORBIS_PSYCHO | 1032 | | if (st->submodeID>=1) | 1033 | | SPEEX_COPY(st->old_curve, st->curve, 128); | 1034 | | #endif | 1035 | | | 1036 | 7.39k | if (st->submodeID==1) | 1037 | 3.87k | { | 1038 | 3.87k | #ifndef DISABLE_VBR | 1039 | 3.87k | if (st->dtx_count) | 1040 | 1.66k | speex_bits_pack(bits, 15, 4); | 1041 | 2.21k | else | 1042 | 2.21k | #endif | 1043 | 2.21k | speex_bits_pack(bits, 0, 4); | 1044 | 3.87k | } | 1045 | | | 1046 | | /* The next frame will not be the first (Duh!) */ | 1047 | 7.39k | st->first = 0; | 1048 | 7.39k | SPEEX_COPY(st->winBuf, in+2*NB_FRAME_SIZE-NB_WINDOW_SIZE, NB_WINDOW_SIZE-NB_FRAME_SIZE); | 1049 | | | 1050 | 7.39k | if (SUBMODE(innovation_quant) == noise_codebook_quant || st->submodeID==0) | 1051 | 3.87k | st->bounded_pitch = 1; | 1052 | 3.52k | else | 1053 | 3.52k | st->bounded_pitch = 0; | 1054 | | | 1055 | 7.39k | return 1; | 1056 | 8.59k | } |
Line | Count | Source | 407 | 19.6k | { | 408 | 19.6k | EncState *st; | 409 | 19.6k | int i, sub, roots; | 410 | 19.6k | int ol_pitch; | 411 | 19.6k | spx_word16_t ol_pitch_coef; | 412 | 19.6k | spx_word32_t ol_gain; | 413 | 19.6k | VARDECL(spx_word16_t *target); | 414 | 19.6k | VARDECL(spx_sig_t *innov); | 415 | 19.6k | VARDECL(spx_word32_t *exc32); | 416 | 19.6k | VARDECL(spx_mem_t *mem); | 417 | 19.6k | VARDECL(spx_coef_t *bw_lpc1); | 418 | 19.6k | VARDECL(spx_coef_t *bw_lpc2); | 419 | 19.6k | VARDECL(spx_coef_t *lpc); | 420 | 19.6k | VARDECL(spx_lsp_t *lsp); | 421 | 19.6k | VARDECL(spx_lsp_t *qlsp); | 422 | 19.6k | VARDECL(spx_lsp_t *interp_lsp); | 423 | 19.6k | VARDECL(spx_lsp_t *interp_qlsp); | 424 | 19.6k | VARDECL(spx_coef_t *interp_lpc); | 425 | 19.6k | VARDECL(spx_coef_t *interp_qlpc); | 426 | 19.6k | char *stack; | 427 | 19.6k | VARDECL(spx_word16_t *syn_resp); | 428 | | | 429 | 19.6k | spx_word32_t ener=0; | 430 | 19.6k | spx_word16_t fine_gain; | 431 | 19.6k | spx_word16_t *in = (spx_word16_t*)vin; | 432 | | | 433 | 19.6k | st=(EncState *)state; | 434 | 19.6k | stack=st->stack; | 435 | | | 436 | 19.6k | ALLOC(lpc, NB_ORDER, spx_coef_t); | 437 | 19.6k | ALLOC(bw_lpc1, NB_ORDER, spx_coef_t); | 438 | 19.6k | ALLOC(bw_lpc2, NB_ORDER, spx_coef_t); | 439 | 19.6k | ALLOC(lsp, NB_ORDER, spx_lsp_t); | 440 | 19.6k | ALLOC(qlsp, NB_ORDER, spx_lsp_t); | 441 | 19.6k | ALLOC(interp_lsp, NB_ORDER, spx_lsp_t); | 442 | 19.6k | ALLOC(interp_qlsp, NB_ORDER, spx_lsp_t); | 443 | 19.6k | ALLOC(interp_lpc, NB_ORDER, spx_coef_t); | 444 | 19.6k | ALLOC(interp_qlpc, NB_ORDER, spx_coef_t); | 445 | | | 446 | 19.6k | st->exc = st->excBuf + NB_PITCH_END + 2; | 447 | 19.6k | st->sw = st->swBuf + NB_PITCH_END + 2; | 448 | | /* Move signals 1 frame towards the past */ | 449 | 19.6k | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, NB_PITCH_END+2); | 450 | 19.6k | SPEEX_MOVE(st->swBuf, st->swBuf+NB_FRAME_SIZE, NB_PITCH_END+2); | 451 | | | 452 | 19.6k | if (st->highpass_enabled) | 453 | 7.07k | highpass(in, in, NB_FRAME_SIZE, (st->isWideband?HIGHPASS_WIDEBAND:HIGHPASS_NARROWBAND)|HIGHPASS_INPUT, st->mem_hp); | 454 | | | 455 | 19.6k | { | 456 | 19.6k | VARDECL(spx_word16_t *w_sig); | 457 | 19.6k | VARDECL(spx_word16_t *autocorr); | 458 | 19.6k | ALLOC(w_sig, NB_WINDOW_SIZE, spx_word16_t); | 459 | 19.6k | ALLOC(autocorr, NB_ORDER+1, spx_word16_t); | 460 | | /* Window for analysis */ | 461 | 806k | for (i=0;i<NB_WINDOW_SIZE-NB_FRAME_SIZE;i++) | 462 | 787k | w_sig[i] = MULT16_16_Q15(st->winBuf[i],st->window[i]); | 463 | 3.16M | for (;i<NB_WINDOW_SIZE;i++) | 464 | 3.14M | w_sig[i] = MULT16_16_Q15(in[i-NB_WINDOW_SIZE+NB_FRAME_SIZE],st->window[i]); | 465 | | /* Compute auto-correlation */ | 466 | 19.6k | _spx_autocorr(w_sig, autocorr, NB_ORDER+1, NB_WINDOW_SIZE); | 467 | 19.6k | autocorr[0] = ADD16(autocorr[0],MULT16_16_Q15(autocorr[0],st->lpc_floor)); /* Noise floor in auto-correlation domain */ | 468 | | | 469 | | /* Lag windowing: equivalent to filtering in the power-spectrum domain */ | 470 | 236k | for (i=0;i<NB_ORDER+1;i++) | 471 | 216k | autocorr[i] = MULT16_16_Q15(autocorr[i],st->lagWindow[i]); | 472 | 19.6k | autocorr[0] = ADD16(autocorr[0],1); | 473 | | | 474 | | /* Levinson-Durbin */ | 475 | 19.6k | _spx_lpc(lpc, autocorr, NB_ORDER); | 476 | | /* LPC to LSPs (x-domain) transform */ | 477 | 19.6k | roots=lpc_to_lsp (lpc, NB_ORDER, lsp, 10, LSP_DELTA1, stack); | 478 | | /* Check if we found all the roots */ | 479 | 19.6k | if (roots!=NB_ORDER) | 480 | 52 | { | 481 | | /*If we can't find all LSP's, do some damage control and use previous filter*/ | 482 | 572 | for (i=0;i<NB_ORDER;i++) | 483 | 520 | { | 484 | 520 | lsp[i]=st->old_lsp[i]; | 485 | 520 | } | 486 | 52 | } | 487 | 19.6k | } | 488 | | | 489 | | | 490 | | | 491 | | | 492 | | /* Whole frame analysis (open-loop estimation of pitch and excitation gain) */ | 493 | 19.6k | { | 494 | 19.6k | int diff = NB_WINDOW_SIZE-NB_FRAME_SIZE; | 495 | 19.6k | if (st->first) | 496 | 39.5k | for (i=0;i<NB_ORDER;i++) | 497 | 35.9k | interp_lsp[i] = lsp[i]; | 498 | 16.0k | else | 499 | 16.0k | lsp_interpolate(st->old_lsp, lsp, interp_lsp, NB_ORDER, NB_NB_SUBFRAMES, NB_NB_SUBFRAMES<<1, LSP_MARGIN); | 500 | | | 501 | | /* Compute interpolated LPCs (unquantized) for whole frame*/ | 502 | 19.6k | lsp_to_lpc(interp_lsp, interp_lpc, NB_ORDER,stack); | 503 | | | 504 | | | 505 | | /*Open-loop pitch*/ | 506 | 19.6k | if (!st->submodes[st->submodeID] || (st->complexity>2 && SUBMODE(have_subframe_gain)<3) || SUBMODE(forced_pitch_gain) || SUBMODE(lbr_pitch) != -1 | 507 | 19.6k | #ifndef DISABLE_VBR | 508 | 19.6k | || st->vbr_enabled || st->vad_enabled | 509 | 19.6k | #endif | 510 | 19.6k | ) | 511 | 18.1k | { | 512 | 18.1k | int nol_pitch[6]; | 513 | 18.1k | spx_word16_t nol_pitch_coef[6]; | 514 | | | 515 | 18.1k | bw_lpc(QCONST16(0.9,15), interp_lpc, bw_lpc1, NB_ORDER); | 516 | 18.1k | bw_lpc(QCONST16(0.55,15), interp_lpc, bw_lpc2, NB_ORDER); | 517 | | | 518 | 18.1k | SPEEX_COPY(st->sw, st->winBuf, diff); | 519 | 18.1k | SPEEX_COPY(st->sw+diff, in, NB_FRAME_SIZE-diff); | 520 | 18.1k | filter10(st->sw, bw_lpc1, bw_lpc2, st->sw, NB_FRAME_SIZE, st->mem_sw_whole, stack); | 521 | | | 522 | 18.1k | open_loop_nbest_pitch(st->sw, NB_PITCH_START, NB_PITCH_END, NB_FRAME_SIZE, | 523 | 18.1k | nol_pitch, nol_pitch_coef, 6, stack); | 524 | 18.1k | ol_pitch=nol_pitch[0]; | 525 | 18.1k | ol_pitch_coef = nol_pitch_coef[0]; | 526 | | /*Try to remove pitch multiples*/ | 527 | 109k | for (i=1;i<6;i++) | 528 | 90.8k | { | 529 | 90.8k | #ifdef FIXED_POINT | 530 | 90.8k | if ((nol_pitch_coef[i]>MULT16_16_Q15(nol_pitch_coef[0],27853)) && | 531 | | #else | 532 | | if ((nol_pitch_coef[i]>.85*nol_pitch_coef[0]) && | 533 | | #endif | 534 | 90.8k | (ABS(2*nol_pitch[i]-ol_pitch)<=2 || ABS(3*nol_pitch[i]-ol_pitch)<=3 || | 535 | 18.5k | ABS(4*nol_pitch[i]-ol_pitch)<=4 || ABS(5*nol_pitch[i]-ol_pitch)<=5)) | 536 | 796 | { | 537 | | /*ol_pitch_coef=nol_pitch_coef[i];*/ | 538 | 796 | ol_pitch = nol_pitch[i]; | 539 | 796 | } | 540 | 90.8k | } | 541 | | /*if (ol_pitch>50) | 542 | | ol_pitch/=2;*/ | 543 | | /*ol_pitch_coef = sqrt(ol_pitch_coef);*/ | 544 | | | 545 | 18.1k | } else { | 546 | 1.49k | ol_pitch=0; | 547 | 1.49k | ol_pitch_coef=0; | 548 | 1.49k | } | 549 | | | 550 | | /*Compute "real" excitation*/ | 551 | | /*SPEEX_COPY(st->exc, st->winBuf, diff); | 552 | | SPEEX_COPY(st->exc+diff, in, NB_FRAME_SIZE-diff);*/ | 553 | 19.6k | fir_mem16(st->winBuf, interp_lpc, st->exc, diff, NB_ORDER, st->mem_exc, stack); | 554 | 19.6k | fir_mem16(in, interp_lpc, st->exc+diff, NB_FRAME_SIZE-diff, NB_ORDER, st->mem_exc, stack); | 555 | | | 556 | | /* Compute open-loop excitation gain */ | 557 | 19.6k | { | 558 | 19.6k | spx_word16_t g = compute_rms16(st->exc, NB_FRAME_SIZE); | 559 | 19.6k | if (st->submodeID!=1 && ol_pitch>0) | 560 | 8.61k | ol_gain = MULT16_16(g, MULT16_16_Q14(QCONST16(1.1,14), | 561 | 19.6k | spx_sqrt(QCONST32(1.,28)-MULT16_32_Q15(QCONST16(.8,15),SHL32(MULT16_16(ol_pitch_coef,ol_pitch_coef),16))))); | 562 | 11.0k | else | 563 | 11.0k | ol_gain = SHL32(EXTEND32(g),SIG_SHIFT); | 564 | 19.6k | } | 565 | 19.6k | } | 566 | | | 567 | | #ifdef VORBIS_PSYCHO | 568 | | SPEEX_MOVE(st->psy_window, st->psy_window+NB_FRAME_SIZE, 256-NB_FRAME_SIZE); | 569 | | SPEEX_COPY(&st->psy_window[256-NB_FRAME_SIZE], in, NB_FRAME_SIZE); | 570 | | compute_curve(st->psy, st->psy_window, st->curve); | 571 | | /*print_vec(st->curve, 128, "curve");*/ | 572 | | if (st->first) | 573 | | SPEEX_COPY(st->old_curve, st->curve, 128); | 574 | | #endif | 575 | | | 576 | | /*VBR stuff*/ | 577 | 19.6k | #ifndef DISABLE_VBR | 578 | 19.6k | if (st->vbr_enabled||st->vad_enabled) | 579 | 14.5k | { | 580 | 14.5k | float lsp_dist=0; | 581 | 160k | for (i=0;i<NB_ORDER;i++) | 582 | 145k | lsp_dist += (st->old_lsp[i] - lsp[i])*(st->old_lsp[i] - lsp[i]); | 583 | 14.5k | lsp_dist /= LSP_SCALING*LSP_SCALING; | 584 | | | 585 | 14.5k | if (st->abr_enabled) | 586 | 3.57k | { | 587 | 3.57k | float qual_change=0; | 588 | 3.57k | if (st->abr_drift2 * st->abr_drift > 0) | 589 | 3.21k | { | 590 | | /* Only adapt if long-term and short-term drift are the same sign */ | 591 | 3.21k | qual_change = -.00001*st->abr_drift/(1+st->abr_count); | 592 | 3.21k | if (qual_change>.05) | 593 | 0 | qual_change=.05; | 594 | 3.21k | if (qual_change<-.05) | 595 | 862 | qual_change=-.05; | 596 | 3.21k | } | 597 | 3.57k | st->vbr_quality += qual_change; | 598 | 3.57k | if (st->vbr_quality>10) | 599 | 0 | st->vbr_quality=10; | 600 | 3.57k | if (st->vbr_quality<0) | 601 | 3.21k | st->vbr_quality=0; | 602 | 3.57k | } | 603 | | | 604 | 14.5k | st->relative_quality = vbr_analysis(&st->vbr, in, NB_FRAME_SIZE, ol_pitch, GAIN_SCALING_1*ol_pitch_coef); | 605 | | /*if (delta_qual<0)*/ | 606 | | /* delta_qual*=.1*(3+st->vbr_quality);*/ | 607 | 14.5k | if (st->vbr_enabled) | 608 | 11.4k | { | 609 | 11.4k | spx_int32_t mode; | 610 | 11.4k | int choice=0; | 611 | 11.4k | float min_diff=100; | 612 | 11.4k | mode = 8; | 613 | 102k | while (mode) | 614 | 91.2k | { | 615 | 91.2k | int v1; | 616 | 91.2k | float thresh; | 617 | 91.2k | v1=(int)floor(st->vbr_quality); | 618 | 91.2k | if (v1==10) | 619 | 4.71k | thresh = vbr_nb_thresh[mode][v1]; | 620 | 86.5k | else | 621 | 86.5k | thresh = (st->vbr_quality-v1)*vbr_nb_thresh[mode][v1+1] + (1+v1-st->vbr_quality)*vbr_nb_thresh[mode][v1]; | 622 | 91.2k | if (st->relative_quality > thresh && | 623 | 91.2k | st->relative_quality-thresh<min_diff) | 624 | 12.8k | { | 625 | 12.8k | choice = mode; | 626 | 12.8k | min_diff = st->relative_quality-thresh; | 627 | 12.8k | } | 628 | 91.2k | mode--; | 629 | 91.2k | } | 630 | 11.4k | mode=choice; | 631 | 11.4k | if (mode==0) | 632 | 2.72k | { | 633 | 2.72k | if (st->dtx_count==0 || lsp_dist>.05 || !st->dtx_enabled || st->dtx_count>20) | 634 | 1.70k | { | 635 | 1.70k | mode=1; | 636 | 1.70k | st->dtx_count=1; | 637 | 1.70k | } else { | 638 | 1.01k | mode=0; | 639 | 1.01k | st->dtx_count++; | 640 | 1.01k | } | 641 | 8.69k | } else { | 642 | 8.69k | st->dtx_count=0; | 643 | 8.69k | } | 644 | | | 645 | 11.4k | speex_encoder_ctl(state, SPEEX_SET_MODE, &mode); | 646 | 11.4k | if (st->vbr_max>0) | 647 | 8.36k | { | 648 | 8.36k | spx_int32_t rate; | 649 | 8.36k | speex_encoder_ctl(state, SPEEX_GET_BITRATE, &rate); | 650 | 8.36k | if (rate > st->vbr_max) | 651 | 3.52k | { | 652 | 3.52k | rate = st->vbr_max; | 653 | 3.52k | speex_encoder_ctl(state, SPEEX_SET_BITRATE, &rate); | 654 | 3.52k | } | 655 | 8.36k | } | 656 | | | 657 | 11.4k | if (st->abr_enabled) | 658 | 3.57k | { | 659 | 3.57k | spx_int32_t bitrate; | 660 | 3.57k | speex_encoder_ctl(state, SPEEX_GET_BITRATE, &bitrate); | 661 | 3.57k | st->abr_drift+=(bitrate-st->abr_enabled); | 662 | 3.57k | st->abr_drift2 = .95*st->abr_drift2 + .05*(bitrate-st->abr_enabled); | 663 | 3.57k | st->abr_count += 1.0; | 664 | 3.57k | } | 665 | | | 666 | 11.4k | } else { | 667 | | /*VAD only case*/ | 668 | 3.17k | int mode; | 669 | 3.17k | if (st->relative_quality<2) | 670 | 738 | { | 671 | 738 | if (st->dtx_count==0 || lsp_dist>.05 || !st->dtx_enabled || st->dtx_count>20) | 672 | 491 | { | 673 | 491 | st->dtx_count=1; | 674 | 491 | mode=1; | 675 | 491 | } else { | 676 | 247 | mode=0; | 677 | 247 | st->dtx_count++; | 678 | 247 | } | 679 | 2.43k | } else { | 680 | 2.43k | st->dtx_count = 0; | 681 | 2.43k | mode=st->submodeSelect; | 682 | 2.43k | } | 683 | | /*speex_encoder_ctl(state, SPEEX_SET_MODE, &mode);*/ | 684 | 3.17k | st->submodeID=mode; | 685 | 3.17k | } | 686 | 14.5k | } else { | 687 | 5.09k | st->relative_quality = -1; | 688 | 5.09k | } | 689 | 19.6k | #endif /* #ifndef DISABLE_VBR */ | 690 | | | 691 | 19.6k | if (st->encode_submode) | 692 | 19.6k | { | 693 | | /* First, transmit a zero for narrowband */ | 694 | 19.6k | speex_bits_pack(bits, 0, 1); | 695 | | | 696 | | /* Transmit the sub-mode we use for this frame */ | 697 | 19.6k | speex_bits_pack(bits, st->submodeID, NB_SUBMODE_BITS); | 698 | | | 699 | 19.6k | } | 700 | | | 701 | | /* If null mode (no transmission), just set a couple things to zero*/ | 702 | 19.6k | if (st->submodes[st->submodeID] == NULL) | 703 | 1.18k | { | 704 | 191k | for (i=0;i<NB_FRAME_SIZE;i++) | 705 | 190k | st->exc[i]=st->sw[i]=VERY_SMALL; | 706 | | | 707 | 13.0k | for (i=0;i<NB_ORDER;i++) | 708 | 11.8k | st->mem_sw[i]=0; | 709 | 1.18k | st->first=1; | 710 | 1.18k | st->bounded_pitch = 1; | 711 | | | 712 | 1.18k | SPEEX_COPY(st->winBuf, in+2*NB_FRAME_SIZE-NB_WINDOW_SIZE, NB_WINDOW_SIZE-NB_FRAME_SIZE); | 713 | | | 714 | | /* Clear memory (no need to really compute it) */ | 715 | 13.0k | for (i=0;i<NB_ORDER;i++) | 716 | 11.8k | st->mem_sp[i] = 0; | 717 | 1.18k | return 0; | 718 | | | 719 | 1.18k | } | 720 | | | 721 | | /* LSP Quantization */ | 722 | 18.4k | if (st->first) | 723 | 2.74k | { | 724 | 30.2k | for (i=0;i<NB_ORDER;i++) | 725 | 27.4k | st->old_lsp[i] = lsp[i]; | 726 | 2.74k | } | 727 | | | 728 | | | 729 | | /*Quantize LSPs*/ | 730 | 18.4k | #if 1 /*0 for unquantized*/ | 731 | 18.4k | SUBMODE(lsp_quant)(lsp, qlsp, NB_ORDER, bits); | 732 | | #else | 733 | | for (i=0;i<NB_ORDER;i++) | 734 | | qlsp[i]=lsp[i]; | 735 | | #endif | 736 | | | 737 | | /*If we use low bit-rate pitch mode, transmit open-loop pitch*/ | 738 | 18.4k | if (SUBMODE(lbr_pitch)!=-1) | 739 | 12.3k | { | 740 | 12.3k | speex_bits_pack(bits, ol_pitch-NB_PITCH_START, 7); | 741 | 12.3k | } | 742 | | | 743 | 18.4k | if (SUBMODE(forced_pitch_gain)) | 744 | 11.2k | { | 745 | 11.2k | int quant; | 746 | | /* This just damps the pitch a bit, because it tends to be too aggressive when forced */ | 747 | 11.2k | ol_pitch_coef = MULT16_16_Q15(QCONST16(.9,15), ol_pitch_coef); | 748 | 11.2k | #ifdef FIXED_POINT | 749 | 11.2k | quant = PSHR16(MULT16_16_16(15, ol_pitch_coef),GAIN_SHIFT); | 750 | | #else | 751 | | quant = (int)floor(.5+15*ol_pitch_coef*GAIN_SCALING_1); | 752 | | #endif | 753 | 11.2k | if (quant>15) | 754 | 92 | quant=15; | 755 | 11.2k | if (quant<0) | 756 | 0 | quant=0; | 757 | 11.2k | speex_bits_pack(bits, quant, 4); | 758 | 11.2k | ol_pitch_coef=MULT16_16_P15(QCONST16(0.066667,15),SHL16(quant,GAIN_SHIFT)); | 759 | 11.2k | } | 760 | | | 761 | | | 762 | | /*Quantize and transmit open-loop excitation gain*/ | 763 | 18.4k | #ifdef FIXED_POINT | 764 | 18.4k | { | 765 | 18.4k | int qe = scal_quant32(ol_gain, ol_gain_table, 32); | 766 | | /*ol_gain = exp(qe/3.5)*SIG_SCALING;*/ | 767 | 18.4k | ol_gain = MULT16_32_Q15(28406,ol_gain_table[qe]); | 768 | 18.4k | speex_bits_pack(bits, qe, 5); | 769 | 18.4k | } | 770 | | #else | 771 | | { | 772 | | int qe = (int)(floor(.5+3.5*log(ol_gain*1.0/SIG_SCALING))); | 773 | | if (qe<0) | 774 | | qe=0; | 775 | | if (qe>31) | 776 | | qe=31; | 777 | | ol_gain = exp(qe/3.5)*SIG_SCALING; | 778 | | speex_bits_pack(bits, qe, 5); | 779 | | } | 780 | | #endif | 781 | | | 782 | | | 783 | | | 784 | | /* Special case for first frame */ | 785 | 18.4k | if (st->first) | 786 | 2.74k | { | 787 | 30.2k | for (i=0;i<NB_ORDER;i++) | 788 | 27.4k | st->old_qlsp[i] = qlsp[i]; | 789 | 2.74k | } | 790 | | | 791 | | /* Target signal */ | 792 | 18.4k | ALLOC(target, NB_SUBFRAME_SIZE, spx_word16_t); | 793 | 18.4k | ALLOC(innov, NB_SUBFRAME_SIZE, spx_sig_t); | 794 | 18.4k | ALLOC(exc32, NB_SUBFRAME_SIZE, spx_word32_t); | 795 | 18.4k | ALLOC(syn_resp, NB_SUBFRAME_SIZE, spx_word16_t); | 796 | 18.4k | ALLOC(mem, NB_ORDER, spx_mem_t); | 797 | | | 798 | | /* Loop on sub-frames */ | 799 | 92.4k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) | 800 | 73.9k | { | 801 | 73.9k | int offset; | 802 | 73.9k | spx_word16_t *sw; | 803 | 73.9k | spx_word16_t *exc, *inBuf; | 804 | 73.9k | int pitch; | 805 | 73.9k | int response_bound = NB_SUBFRAME_SIZE; | 806 | | | 807 | | /* Offset relative to start of frame */ | 808 | 73.9k | offset = NB_SUBFRAME_SIZE*sub; | 809 | | /* Excitation */ | 810 | 73.9k | exc=st->exc+offset; | 811 | | /* Weighted signal */ | 812 | 73.9k | sw=st->sw+offset; | 813 | | | 814 | | /* LSP interpolation (quantized and unquantized) */ | 815 | 73.9k | lsp_interpolate(st->old_lsp, lsp, interp_lsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); | 816 | 73.9k | lsp_interpolate(st->old_qlsp, qlsp, interp_qlsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); | 817 | | | 818 | | /* Compute interpolated LPCs (quantized and unquantized) */ | 819 | 73.9k | lsp_to_lpc(interp_lsp, interp_lpc, NB_ORDER,stack); | 820 | | | 821 | 73.9k | lsp_to_lpc(interp_qlsp, interp_qlpc, NB_ORDER, stack); | 822 | | | 823 | | /* Compute analysis filter gain at w=pi (for use in SB-CELP) */ | 824 | 73.9k | { | 825 | 73.9k | spx_word32_t pi_g=LPC_SCALING; | 826 | 443k | for (i=0;i<NB_ORDER;i+=2) | 827 | 369k | { | 828 | | /*pi_g += -st->interp_qlpc[i] + st->interp_qlpc[i+1];*/ | 829 | 369k | pi_g = ADD32(pi_g, SUB32(EXTEND32(interp_qlpc[i+1]),EXTEND32(interp_qlpc[i]))); | 830 | 369k | } | 831 | 73.9k | st->pi_gain[sub] = pi_g; | 832 | 73.9k | } | 833 | | | 834 | | #ifdef VORBIS_PSYCHO | 835 | | { | 836 | | float curr_curve[128]; | 837 | | float fact = ((float)sub+1.0f)/NB_NB_SUBFRAMES; | 838 | | for (i=0;i<128;i++) | 839 | | curr_curve[i] = (1.0f-fact)*st->old_curve[i] + fact*st->curve[i]; | 840 | | curve_to_lpc(st->psy, curr_curve, bw_lpc1, bw_lpc2, 10); | 841 | | } | 842 | | #else | 843 | | /* Compute bandwidth-expanded (unquantized) LPCs for perceptual weighting */ | 844 | 73.9k | bw_lpc(st->gamma1, interp_lpc, bw_lpc1, NB_ORDER); | 845 | 73.9k | bw_lpc(st->gamma2, interp_lpc, bw_lpc2, NB_ORDER); | 846 | | /*print_vec(st->bw_lpc1, 10, "bw_lpc");*/ | 847 | 73.9k | #endif | 848 | | | 849 | | /*FIXME: This will break if we change the window size */ | 850 | 73.9k | speex_assert(NB_WINDOW_SIZE-NB_FRAME_SIZE == NB_SUBFRAME_SIZE); | 851 | 73.9k | if (sub==0) | 852 | 18.4k | inBuf = st->winBuf; | 853 | 55.4k | else | 854 | 55.4k | inBuf = &in[((sub-1)*NB_SUBFRAME_SIZE)]; | 855 | 3.03M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 856 | 2.95M | sw[i] = inBuf[i]; | 857 | | | 858 | 73.9k | if (st->complexity==0) | 859 | 18.4k | response_bound >>= 1; | 860 | 73.9k | compute_impulse_response(interp_qlpc, bw_lpc1, bw_lpc2, syn_resp, response_bound, NB_ORDER, stack); | 861 | 442k | for (i=response_bound;i<NB_SUBFRAME_SIZE;i++) | 862 | 368k | syn_resp[i]=VERY_SMALL; | 863 | | | 864 | | /* Compute zero response of A(z/g1) / ( A(z/g2) * A(z) ) */ | 865 | 813k | for (i=0;i<NB_ORDER;i++) | 866 | 739k | mem[i]=SHL32(st->mem_sp[i],1); | 867 | 3.03M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 868 | 2.95M | exc[i] = VERY_SMALL; | 869 | | #ifdef SHORTCUTS2 | 870 | | iir_mem16(exc, interp_qlpc, exc, response_bound, NB_ORDER, mem, stack); | 871 | | for (i=0;i<NB_ORDER;i++) | 872 | | mem[i]=SHL32(st->mem_sw[i],1); | 873 | | filter10(exc, st->bw_lpc1, st->bw_lpc2, exc, response_bound, mem, stack); | 874 | | SPEEX_MEMSET(&exc[response_bound], 0, NB_SUBFRAME_SIZE-response_bound); | 875 | | #else | 876 | 73.9k | iir_mem16(exc, interp_qlpc, exc, NB_SUBFRAME_SIZE, NB_ORDER, mem, stack); | 877 | 813k | for (i=0;i<NB_ORDER;i++) | 878 | 739k | mem[i]=SHL32(st->mem_sw[i],1); | 879 | 73.9k | filter10(exc, bw_lpc1, bw_lpc2, exc, NB_SUBFRAME_SIZE, mem, stack); | 880 | 73.9k | #endif | 881 | | | 882 | | /* Compute weighted signal */ | 883 | 813k | for (i=0;i<NB_ORDER;i++) | 884 | 739k | mem[i]=st->mem_sw[i]; | 885 | 73.9k | filter10(sw, bw_lpc1, bw_lpc2, sw, NB_SUBFRAME_SIZE, mem, stack); | 886 | | | 887 | 73.9k | if (st->complexity==0) | 888 | 202k | for (i=0;i<NB_ORDER;i++) | 889 | 184k | st->mem_sw[i]=mem[i]; | 890 | | | 891 | | /* Compute target signal (saturation prevents overflows on clipped input speech) */ | 892 | 3.03M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 893 | 2.95M | target[i]=EXTRACT16(SATURATE(SUB32(sw[i],PSHR32(exc[i],1)),32767)); | 894 | | | 895 | 3.03M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 896 | 2.95M | exc[i] = inBuf[i]; | 897 | 73.9k | fir_mem16(exc, interp_qlpc, exc, NB_SUBFRAME_SIZE, NB_ORDER, st->mem_exc2, stack); | 898 | | /* If we have a long-term predictor (otherwise, something's wrong) */ | 899 | 73.9k | speex_assert (SUBMODE(ltp_quant)); | 900 | 73.9k | { | 901 | 73.9k | int pit_min, pit_max; | 902 | | /* Long-term prediction */ | 903 | 73.9k | if (SUBMODE(lbr_pitch) != -1) | 904 | 49.2k | { | 905 | | /* Low bit-rate pitch handling */ | 906 | 49.2k | int margin; | 907 | 49.2k | margin = SUBMODE(lbr_pitch); | 908 | 49.2k | if (margin) | 909 | 0 | { | 910 | 0 | if (ol_pitch < NB_PITCH_START+margin-1) | 911 | 0 | ol_pitch=NB_PITCH_START+margin-1; | 912 | 0 | if (ol_pitch > NB_PITCH_END-margin) | 913 | 0 | ol_pitch=NB_PITCH_END-margin; | 914 | 0 | pit_min = ol_pitch-margin+1; | 915 | 0 | pit_max = ol_pitch+margin; | 916 | 49.2k | } else { | 917 | 49.2k | pit_min=pit_max=ol_pitch; | 918 | 49.2k | } | 919 | 49.2k | } else { | 920 | 24.7k | pit_min = NB_PITCH_START; | 921 | 24.7k | pit_max = NB_PITCH_END; | 922 | 24.7k | } | 923 | | | 924 | | /* Force pitch to use only the current frame if needed */ | 925 | 73.9k | if (st->bounded_pitch && pit_max>offset) | 926 | 23.2k | pit_max=offset; | 927 | | | 928 | | /* Perform pitch search */ | 929 | 73.9k | pitch = SUBMODE(ltp_quant)(target, sw, interp_qlpc, bw_lpc1, bw_lpc2, | 930 | 73.9k | exc32, SUBMODE(ltp_params), pit_min, pit_max, ol_pitch_coef, | 931 | 73.9k | NB_ORDER, NB_SUBFRAME_SIZE, bits, stack, | 932 | 73.9k | exc, syn_resp, st->complexity, 0, st->plc_tuning, &st->cumul_gain); | 933 | | | 934 | 73.9k | st->pitch[sub]=pitch; | 935 | 73.9k | } | 936 | | /* Quantization of innovation */ | 937 | 73.9k | SPEEX_MEMSET(innov, 0, NB_SUBFRAME_SIZE); | 938 | | | 939 | | /* FIXME: Make sure this is safe from overflows (so far so good) */ | 940 | 3.03M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 941 | 2.95M | exc[i] = EXTRACT16(SUB32(EXTEND32(exc[i]), PSHR32(exc32[i],SIG_SHIFT-1))); | 942 | | | 943 | 73.9k | ener = SHL32(EXTEND32(compute_rms16(exc, NB_SUBFRAME_SIZE)),SIG_SHIFT); | 944 | | | 945 | | /*FIXME: Should use DIV32_16 and make sure result fits in 16 bits */ | 946 | 73.9k | #ifdef FIXED_POINT | 947 | 73.9k | { | 948 | 73.9k | spx_word32_t f = PDIV32(ener,PSHR32(ol_gain,SIG_SHIFT)); | 949 | 73.9k | if (f<=32767) | 950 | 68.2k | fine_gain = f; | 951 | 5.66k | else | 952 | 5.66k | fine_gain = 32767; | 953 | 73.9k | } | 954 | | #else | 955 | | fine_gain = PDIV32_16(ener,PSHR32(ol_gain,SIG_SHIFT)); | 956 | | #endif | 957 | | /* Calculate gain correction for the sub-frame (if any) */ | 958 | 73.9k | if (SUBMODE(have_subframe_gain)) | 959 | 24.7k | { | 960 | 24.7k | int qe; | 961 | 24.7k | if (SUBMODE(have_subframe_gain)==3) | 962 | 14.9k | { | 963 | 14.9k | qe = scal_quant(fine_gain, exc_gain_quant_scal3_bound, 8); | 964 | 14.9k | speex_bits_pack(bits, qe, 3); | 965 | 14.9k | ener=MULT16_32_Q14(exc_gain_quant_scal3[qe],ol_gain); | 966 | 14.9k | } else { | 967 | 9.77k | qe = scal_quant(fine_gain, exc_gain_quant_scal1_bound, 2); | 968 | 9.77k | speex_bits_pack(bits, qe, 1); | 969 | 9.77k | ener=MULT16_32_Q14(exc_gain_quant_scal1[qe],ol_gain); | 970 | 9.77k | } | 971 | 49.2k | } else { | 972 | 49.2k | ener=ol_gain; | 973 | 49.2k | } | 974 | | | 975 | | /*printf ("%f %f\n", ener, ol_gain);*/ | 976 | | | 977 | | /* Normalize innovation */ | 978 | 73.9k | signal_div(target, target, ener, NB_SUBFRAME_SIZE); | 979 | | | 980 | | /* Quantize innovation */ | 981 | 73.9k | speex_assert (SUBMODE(innovation_quant)); | 982 | 73.9k | { | 983 | | /* Codebook search */ | 984 | 73.9k | SUBMODE(innovation_quant)(target, interp_qlpc, bw_lpc1, bw_lpc2, | 985 | 73.9k | SUBMODE(innovation_params), NB_ORDER, NB_SUBFRAME_SIZE, | 986 | 73.9k | innov, syn_resp, bits, stack, st->complexity, SUBMODE(double_codebook)); | 987 | | | 988 | | /* De-normalize innovation and update excitation */ | 989 | 73.9k | signal_mul(innov, innov, ener, NB_SUBFRAME_SIZE); | 990 | | | 991 | | /* In some (rare) modes, we do a second search (more bits) to reduce noise even more */ | 992 | 73.9k | if (SUBMODE(double_codebook)) { | 993 | 2.09k | char *tmp_stack=stack; | 994 | 2.09k | VARDECL(spx_sig_t *innov2); | 995 | 2.09k | ALLOC(innov2, NB_SUBFRAME_SIZE, spx_sig_t); | 996 | 2.09k | SPEEX_MEMSET(innov2, 0, NB_SUBFRAME_SIZE); | 997 | 85.9k | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 998 | 83.8k | target[i]=MULT16_16_P13(QCONST16(2.2f,13), target[i]); | 999 | 2.09k | SUBMODE(innovation_quant)(target, interp_qlpc, bw_lpc1, bw_lpc2, | 1000 | 2.09k | SUBMODE(innovation_params), NB_ORDER, NB_SUBFRAME_SIZE, | 1001 | 2.09k | innov2, syn_resp, bits, stack, st->complexity, 0); | 1002 | 2.09k | signal_mul(innov2, innov2, MULT16_32_Q15(QCONST16(0.454545f,15),ener), NB_SUBFRAME_SIZE); | 1003 | 85.9k | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1004 | 83.8k | innov[i] = ADD32(innov[i],innov2[i]); | 1005 | 2.09k | stack = tmp_stack; | 1006 | 2.09k | } | 1007 | 3.03M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1008 | 2.95M | exc[i] = EXTRACT16(SATURATE32(PSHR32(ADD32(SHL32(exc32[i],1),innov[i]),SIG_SHIFT),32767)); | 1009 | 73.9k | if (st->innov_rms_save) | 1010 | 20.4k | st->innov_rms_save[sub] = compute_rms(innov, NB_SUBFRAME_SIZE); | 1011 | 73.9k | } | 1012 | | | 1013 | | /* Final signal synthesis from excitation */ | 1014 | 73.9k | iir_mem16(exc, interp_qlpc, sw, NB_SUBFRAME_SIZE, NB_ORDER, st->mem_sp, stack); | 1015 | | | 1016 | | /* Compute weighted signal again, from synthesized speech (not sure it's the right thing) */ | 1017 | 73.9k | if (st->complexity!=0) | 1018 | 55.5k | filter10(sw, bw_lpc1, bw_lpc2, sw, NB_SUBFRAME_SIZE, st->mem_sw, stack); | 1019 | | | 1020 | 73.9k | } | 1021 | | | 1022 | | /* Store the LSPs for interpolation in the next frame */ | 1023 | 18.4k | if (st->submodeID>=1) | 1024 | 18.4k | { | 1025 | 203k | for (i=0;i<NB_ORDER;i++) | 1026 | 184k | st->old_lsp[i] = lsp[i]; | 1027 | 203k | for (i=0;i<NB_ORDER;i++) | 1028 | 184k | st->old_qlsp[i] = qlsp[i]; | 1029 | 18.4k | } | 1030 | | | 1031 | | #ifdef VORBIS_PSYCHO | 1032 | | if (st->submodeID>=1) | 1033 | | SPEEX_COPY(st->old_curve, st->curve, 128); | 1034 | | #endif | 1035 | | | 1036 | 18.4k | if (st->submodeID==1) | 1037 | 9.37k | { | 1038 | 9.37k | #ifndef DISABLE_VBR | 1039 | 9.37k | if (st->dtx_count) | 1040 | 2.26k | speex_bits_pack(bits, 15, 4); | 1041 | 7.10k | else | 1042 | 7.10k | #endif | 1043 | 7.10k | speex_bits_pack(bits, 0, 4); | 1044 | 9.37k | } | 1045 | | | 1046 | | /* The next frame will not be the first (Duh!) */ | 1047 | 18.4k | st->first = 0; | 1048 | 18.4k | SPEEX_COPY(st->winBuf, in+2*NB_FRAME_SIZE-NB_WINDOW_SIZE, NB_WINDOW_SIZE-NB_FRAME_SIZE); | 1049 | | | 1050 | 18.4k | if (SUBMODE(innovation_quant) == noise_codebook_quant || st->submodeID==0) | 1051 | 9.37k | st->bounded_pitch = 1; | 1052 | 9.11k | else | 1053 | 9.11k | st->bounded_pitch = 0; | 1054 | | | 1055 | 18.4k | return 1; | 1056 | 19.6k | } |
|
1057 | | #endif /* DISABLE_ENCODER */ |
1058 | | |
1059 | | |
1060 | | #ifndef DISABLE_DECODER |
1061 | | void *nb_decoder_init(const SpeexMode *m) |
1062 | 4.07k | { |
1063 | 4.07k | DecState *st; |
1064 | 4.07k | const SpeexNBMode *mode; |
1065 | 4.07k | int i; |
1066 | | |
1067 | 4.07k | mode=(const SpeexNBMode*)m->mode; |
1068 | 4.07k | st = (DecState *)speex_alloc(sizeof(DecState)); |
1069 | 4.07k | if (!st) |
1070 | 0 | return NULL; |
1071 | 4.07k | #if defined(VAR_ARRAYS) || defined (USE_ALLOCA) |
1072 | 4.07k | st->stack = NULL; |
1073 | | #else |
1074 | | st->stack = (char*)speex_alloc_scratch(NB_DEC_STACK); |
1075 | | #endif |
1076 | | |
1077 | 4.07k | st->mode=m; |
1078 | | |
1079 | | |
1080 | 4.07k | st->encode_submode = 1; |
1081 | | |
1082 | 4.07k | st->first=1; |
1083 | | /* Codec parameters, should eventually have several "modes"*/ |
1084 | | |
1085 | 4.07k | st->submodes=mode->submodes; |
1086 | 4.07k | st->submodeID=mode->defaultSubmode; |
1087 | | |
1088 | 4.07k | st->lpc_enh_enabled=1; |
1089 | | |
1090 | 4.07k | SPEEX_MEMSET(st->excBuf, 0, NB_FRAME_SIZE + NB_PITCH_END); |
1091 | | |
1092 | 4.07k | st->last_pitch = 40; |
1093 | 4.07k | st->count_lost=0; |
1094 | 4.07k | st->pitch_gain_buf[0] = st->pitch_gain_buf[1] = st->pitch_gain_buf[2] = 0; |
1095 | 4.07k | st->pitch_gain_buf_idx = 0; |
1096 | 4.07k | st->seed = 1000; |
1097 | | |
1098 | 4.07k | st->sampling_rate=8000; |
1099 | 4.07k | st->last_ol_gain = 0; |
1100 | | |
1101 | 4.07k | st->user_callback.func = &speex_default_user_handler; |
1102 | 4.07k | st->user_callback.data = NULL; |
1103 | 69.2k | for (i=0;i<16;i++) |
1104 | 65.2k | st->speex_callbacks[i].func = NULL; |
1105 | | |
1106 | 4.07k | st->voc_m1=st->voc_m2=st->voc_mean=0; |
1107 | 4.07k | st->voc_offset=0; |
1108 | 4.07k | st->dtx_enabled=0; |
1109 | 4.07k | st->isWideband = 0; |
1110 | 4.07k | st->highpass_enabled = 1; |
1111 | | |
1112 | | #ifdef ENABLE_VALGRIND |
1113 | | VALGRIND_MAKE_MEM_DEFINED(st, NB_DEC_STACK); |
1114 | | #endif |
1115 | 4.07k | return st; |
1116 | 4.07k | } |
1117 | | |
1118 | | void nb_decoder_destroy(void *state) |
1119 | 4.07k | { |
1120 | | #if !(defined(VAR_ARRAYS) || defined (USE_ALLOCA)) |
1121 | | DecState *st; |
1122 | | st=(DecState*)state; |
1123 | | |
1124 | | speex_free_scratch(st->stack); |
1125 | | #endif |
1126 | | |
1127 | 4.07k | speex_free(state); |
1128 | 4.07k | } |
1129 | | |
1130 | | int nb_decoder_ctl(void *state, int request, void *ptr) |
1131 | 181k | { |
1132 | 181k | DecState *st; |
1133 | 181k | st=(DecState*)state; |
1134 | 181k | switch(request) |
1135 | 181k | { |
1136 | 0 | case SPEEX_SET_LOW_MODE: |
1137 | 0 | case SPEEX_SET_MODE: |
1138 | 0 | st->submodeID = (*(spx_int32_t*)ptr); |
1139 | 0 | break; |
1140 | 0 | case SPEEX_GET_LOW_MODE: |
1141 | 0 | case SPEEX_GET_MODE: |
1142 | 0 | (*(spx_int32_t*)ptr) = st->submodeID; |
1143 | 0 | break; |
1144 | 4.07k | case SPEEX_SET_ENH: |
1145 | 4.07k | st->lpc_enh_enabled = *((spx_int32_t*)ptr); |
1146 | 4.07k | break; |
1147 | 0 | case SPEEX_GET_ENH: |
1148 | 0 | *((spx_int32_t*)ptr) = st->lpc_enh_enabled; |
1149 | 0 | break; |
1150 | 9.98k | case SPEEX_GET_FRAME_SIZE: |
1151 | 9.98k | (*(spx_int32_t*)ptr) = NB_FRAME_SIZE; |
1152 | 9.98k | break; |
1153 | 0 | case SPEEX_GET_BITRATE: |
1154 | 0 | if (st->submodes[st->submodeID]) |
1155 | 0 | (*(spx_int32_t*)ptr) = st->sampling_rate*SUBMODE(bits_per_frame)/NB_FRAME_SIZE; |
1156 | 0 | else |
1157 | 0 | (*(spx_int32_t*)ptr) = st->sampling_rate*(NB_SUBMODE_BITS+1)/NB_FRAME_SIZE; |
1158 | 0 | break; |
1159 | 4.07k | case SPEEX_SET_SAMPLING_RATE: |
1160 | 4.07k | st->sampling_rate = (*(spx_int32_t*)ptr); |
1161 | 4.07k | break; |
1162 | 2.87k | case SPEEX_GET_SAMPLING_RATE: |
1163 | 2.87k | (*(spx_int32_t*)ptr)=st->sampling_rate; |
1164 | 2.87k | break; |
1165 | 2.09k | case SPEEX_SET_HANDLER: |
1166 | 2.09k | { |
1167 | 2.09k | SpeexCallback *c = (SpeexCallback*)ptr; |
1168 | 2.09k | st->speex_callbacks[c->callback_id].func=c->func; |
1169 | 2.09k | st->speex_callbacks[c->callback_id].data=c->data; |
1170 | 2.09k | st->speex_callbacks[c->callback_id].callback_id=c->callback_id; |
1171 | 2.09k | } |
1172 | 2.09k | break; |
1173 | 0 | case SPEEX_SET_USER_HANDLER: |
1174 | 0 | { |
1175 | 0 | SpeexCallback *c = (SpeexCallback*)ptr; |
1176 | 0 | st->user_callback.func=c->func; |
1177 | 0 | st->user_callback.data=c->data; |
1178 | 0 | st->user_callback.callback_id=c->callback_id; |
1179 | 0 | } |
1180 | 0 | break; |
1181 | 0 | case SPEEX_RESET_STATE: |
1182 | 0 | { |
1183 | 0 | int i; |
1184 | 0 | for (i=0;i<NB_ORDER;i++) |
1185 | 0 | st->mem_sp[i]=0; |
1186 | 0 | for (i=0;i<NB_FRAME_SIZE + NB_PITCH_END + 1;i++) |
1187 | 0 | st->excBuf[i]=0; |
1188 | 0 | } |
1189 | 0 | break; |
1190 | 0 | case SPEEX_SET_SUBMODE_ENCODING: |
1191 | 0 | st->encode_submode = (*(spx_int32_t*)ptr); |
1192 | 0 | break; |
1193 | 0 | case SPEEX_GET_SUBMODE_ENCODING: |
1194 | 0 | (*(spx_int32_t*)ptr) = st->encode_submode; |
1195 | 0 | break; |
1196 | 3.91k | case SPEEX_GET_LOOKAHEAD: |
1197 | 3.91k | (*(spx_int32_t*)ptr)=NB_SUBFRAME_SIZE; |
1198 | 3.91k | break; |
1199 | 0 | case SPEEX_SET_HIGHPASS: |
1200 | 0 | st->highpass_enabled = (*(spx_int32_t*)ptr); |
1201 | 0 | break; |
1202 | 0 | case SPEEX_GET_HIGHPASS: |
1203 | 0 | (*(spx_int32_t*)ptr) = st->highpass_enabled; |
1204 | 0 | break; |
1205 | | /* FIXME: Convert to fixed-point and re-enable even when float API is disabled */ |
1206 | 0 | #ifndef DISABLE_FLOAT_API |
1207 | 0 | case SPEEX_GET_ACTIVITY: |
1208 | 0 | { |
1209 | 0 | float ret; |
1210 | 0 | ret = log(st->level/st->min_level)/log(st->max_level/st->min_level); |
1211 | 0 | if (ret>1) |
1212 | 0 | ret = 1; |
1213 | | /* Done in a strange way to catch NaNs as well */ |
1214 | 0 | if (!(ret > 0)) |
1215 | 0 | ret = 0; |
1216 | | /*printf ("%f %f %f %f\n", st->level, st->min_level, st->max_level, ret);*/ |
1217 | 0 | (*(spx_int32_t*)ptr) = (int)(100*ret); |
1218 | 0 | } |
1219 | 0 | break; |
1220 | 0 | #endif |
1221 | 5.12k | case SPEEX_GET_PI_GAIN: |
1222 | 5.12k | { |
1223 | 5.12k | int i; |
1224 | 5.12k | spx_word32_t *g = (spx_word32_t*)ptr; |
1225 | 25.6k | for (i=0;i<NB_NB_SUBFRAMES;i++) |
1226 | 20.5k | g[i]=st->pi_gain[i]; |
1227 | 5.12k | } |
1228 | 5.12k | break; |
1229 | 5.12k | case SPEEX_GET_EXC: |
1230 | 5.12k | { |
1231 | 5.12k | int i; |
1232 | 25.6k | for (i=0;i<NB_NB_SUBFRAMES;i++) |
1233 | 20.5k | ((spx_word16_t*)ptr)[i] = compute_rms16(st->exc+i*NB_SUBFRAME_SIZE, NB_SUBFRAME_SIZE); |
1234 | 5.12k | } |
1235 | 5.12k | break; |
1236 | 87.6k | case SPEEX_GET_DTX_STATUS: |
1237 | 87.6k | *((spx_int32_t*)ptr) = st->dtx_enabled; |
1238 | 87.6k | break; |
1239 | 51.8k | case SPEEX_SET_INNOVATION_SAVE: |
1240 | 51.8k | st->innov_save = (spx_word16_t*)ptr; |
1241 | 51.8k | break; |
1242 | 4.59k | case SPEEX_SET_WIDEBAND: |
1243 | 4.59k | st->isWideband = *((spx_int32_t*)ptr); |
1244 | 4.59k | break; |
1245 | 0 | case SPEEX_GET_STACK: |
1246 | 0 | *((char**)ptr) = st->stack; |
1247 | 0 | break; |
1248 | 0 | default: |
1249 | 0 | speex_warning_int("Unknown nb_ctl request: ", request); |
1250 | 0 | return -1; |
1251 | 181k | } |
1252 | 181k | return 0; |
1253 | 181k | } |
1254 | | |
1255 | | |
1256 | 0 | #define median3(a, b, c) ((a) < (b) ? ((b) < (c) ? (b) : ((a) < (c) ? (c) : (a))) : ((c) < (b) ? (b) : ((c) < (a) ? (c) : (a)))) |
1257 | | |
1258 | | #ifdef FIXED_POINT |
1259 | | const spx_word16_t attenuation[10] = {32767, 31483, 27923, 22861, 17278, 12055, 7764, 4616, 2533, 1283}; |
1260 | | #else |
1261 | | const spx_word16_t attenuation[10] = {1., 0.961, 0.852, 0.698, 0.527, 0.368, 0.237, 0.141, 0.077, 0.039}; |
1262 | | |
1263 | | #endif |
1264 | | |
1265 | | static void nb_decode_lost(DecState *st, spx_word16_t *out, char *stack) |
1266 | 0 | { |
1267 | 0 | int i; |
1268 | 0 | int pitch_val; |
1269 | 0 | spx_word16_t pitch_gain; |
1270 | 0 | spx_word16_t fact; |
1271 | 0 | spx_word16_t gain_med; |
1272 | 0 | spx_word16_t innov_gain; |
1273 | 0 | spx_word16_t noise_gain; |
1274 | |
|
1275 | 0 | st->exc = st->excBuf + 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 6; |
1276 | |
|
1277 | 0 | if (st->count_lost<10) |
1278 | 0 | fact = attenuation[st->count_lost]; |
1279 | 0 | else |
1280 | 0 | fact = 0; |
1281 | |
|
1282 | 0 | gain_med = median3(st->pitch_gain_buf[0], st->pitch_gain_buf[1], st->pitch_gain_buf[2]); |
1283 | 0 | if (gain_med < st->last_pitch_gain) |
1284 | 0 | st->last_pitch_gain = gain_med; |
1285 | |
|
1286 | | #ifdef FIXED_POINT |
1287 | | pitch_gain = st->last_pitch_gain; |
1288 | 0 | if (pitch_gain>54) |
1289 | 0 | pitch_gain = 54; |
1290 | 0 | pitch_gain = SHL16(pitch_gain, 9); |
1291 | | #else |
1292 | 0 | pitch_gain = GAIN_SCALING_1*st->last_pitch_gain; |
1293 | 0 | if (pitch_gain>.85) |
1294 | 0 | pitch_gain=.85; |
1295 | | #endif |
1296 | 0 | pitch_gain = MULT16_16_Q15(fact,pitch_gain) + VERY_SMALL; |
1297 | | /* FIXME: This was rms of innovation (not exc) */ |
1298 | 0 | innov_gain = compute_rms16(st->exc, NB_FRAME_SIZE); |
1299 | 0 | noise_gain = MULT16_16_Q15(innov_gain, MULT16_16_Q15(fact, SUB16(Q15ONE,MULT16_16_Q15(pitch_gain,pitch_gain)))); |
1300 | | /* Shift all buffers by one frame */ |
1301 | 0 | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 12); |
1302 | | |
1303 | |
|
1304 | 0 | pitch_val = st->last_pitch + SHR32((spx_int32_t)speex_rand(1+st->count_lost, &st->seed),SIG_SHIFT); |
1305 | 0 | if (pitch_val > NB_PITCH_END) |
1306 | 0 | pitch_val = NB_PITCH_END; |
1307 | 0 | if (pitch_val < NB_PITCH_START) |
1308 | 0 | pitch_val = NB_PITCH_START; |
1309 | 0 | for (i=0;i<NB_FRAME_SIZE;i++) |
1310 | 0 | { |
1311 | 0 | st->exc[i]= MULT16_16_Q15(pitch_gain, (st->exc[i-pitch_val]+VERY_SMALL)) + |
1312 | 0 | speex_rand(noise_gain, &st->seed); |
1313 | 0 | } |
1314 | |
|
1315 | 0 | bw_lpc(QCONST16(.98,15), st->interp_qlpc, st->interp_qlpc, NB_ORDER); |
1316 | 0 | iir_mem16(&st->exc[-NB_SUBFRAME_SIZE], st->interp_qlpc, out, NB_FRAME_SIZE, |
1317 | 0 | NB_ORDER, st->mem_sp, stack); |
1318 | 0 | highpass(out, out, NB_FRAME_SIZE, HIGHPASS_NARROWBAND|HIGHPASS_OUTPUT, st->mem_hp); |
1319 | |
|
1320 | 0 | st->first = 0; |
1321 | 0 | st->count_lost++; |
1322 | 0 | st->pitch_gain_buf[st->pitch_gain_buf_idx++] = PSHR16(pitch_gain,9); |
1323 | 0 | if (st->pitch_gain_buf_idx > 2) /* rollover */ |
1324 | 0 | st->pitch_gain_buf_idx = 0; |
1325 | 0 | } Unexecuted instantiation: nb_celp.c:nb_decode_lost Unexecuted instantiation: nb_celp.c:nb_decode_lost |
1326 | | |
1327 | | /* Just so we don't need to carry the complete wideband mode information */ |
1328 | | static const int wb_skip_table[8] = {0, 36, 112, 192, 352, 0, 0, 0}; |
1329 | | |
1330 | | int nb_decode(void *state, SpeexBits *bits, void *vout) |
1331 | 68.6k | { |
1332 | 68.6k | DecState *st; |
1333 | 68.6k | int i, sub; |
1334 | 68.6k | int pitch; |
1335 | 68.6k | spx_word16_t pitch_gain[3]; |
1336 | 68.6k | spx_word32_t ol_gain=0; |
1337 | 68.6k | int ol_pitch=0; |
1338 | 68.6k | spx_word16_t ol_pitch_coef=0; |
1339 | 68.6k | int best_pitch=40; |
1340 | 68.6k | spx_word16_t best_pitch_gain=0; |
1341 | 68.6k | int wideband; |
1342 | 68.6k | int m; |
1343 | 68.6k | char *stack; |
1344 | 68.6k | VARDECL(spx_sig_t *innov); |
1345 | 68.6k | VARDECL(spx_word32_t *exc32); |
1346 | 68.6k | VARDECL(spx_coef_t *ak); |
1347 | 68.6k | VARDECL(spx_lsp_t *qlsp); |
1348 | 68.6k | spx_word16_t pitch_average=0; |
1349 | | |
1350 | 68.6k | spx_word16_t *out = (spx_word16_t*)vout; |
1351 | 68.6k | VARDECL(spx_lsp_t *interp_qlsp); |
1352 | | |
1353 | 68.6k | st=(DecState*)state; |
1354 | 68.6k | stack=st->stack; |
1355 | | |
1356 | 68.6k | st->exc = st->excBuf + 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 6; |
1357 | | |
1358 | | /* Check if we're in DTX mode*/ |
1359 | 68.6k | if (!bits && st->dtx_enabled) |
1360 | 0 | { |
1361 | 0 | st->submodeID=0; |
1362 | 0 | } else |
1363 | 68.6k | { |
1364 | | /* If bits is NULL, consider the packet to be lost (what could we do anyway) */ |
1365 | 68.6k | if (!bits) |
1366 | 0 | { |
1367 | 0 | nb_decode_lost(st, out, stack); |
1368 | 0 | return 0; |
1369 | 0 | } |
1370 | | |
1371 | 68.6k | if (st->encode_submode) |
1372 | 68.6k | { |
1373 | | |
1374 | | /* Search for next narrowband block (handle requests, skip wideband blocks) */ |
1375 | 73.7k | do { |
1376 | 73.7k | if (speex_bits_remaining(bits)<5) |
1377 | 13.5k | return -1; |
1378 | 60.1k | wideband = speex_bits_unpack_unsigned(bits, 1); |
1379 | 60.1k | if (wideband) /* Skip wideband block (for compatibility) */ |
1380 | 3.93k | { |
1381 | 3.93k | int submode; |
1382 | 3.93k | int advance; |
1383 | 3.93k | advance = submode = speex_bits_unpack_unsigned(bits, SB_SUBMODE_BITS); |
1384 | | /*speex_mode_query(&speex_wb_mode, SPEEX_SUBMODE_BITS_PER_FRAME, &advance);*/ |
1385 | 3.93k | advance = wb_skip_table[submode]; |
1386 | 3.93k | if (advance < 0) |
1387 | 0 | { |
1388 | 0 | speex_notify("Invalid mode encountered. The stream is corrupted."); |
1389 | 0 | return -2; |
1390 | 0 | } |
1391 | 3.93k | advance -= (SB_SUBMODE_BITS+1); |
1392 | 3.93k | speex_bits_advance(bits, advance); |
1393 | | |
1394 | 3.93k | if (speex_bits_remaining(bits)<5) |
1395 | 936 | return -1; |
1396 | 2.99k | wideband = speex_bits_unpack_unsigned(bits, 1); |
1397 | 2.99k | if (wideband) |
1398 | 2.34k | { |
1399 | 2.34k | advance = submode = speex_bits_unpack_unsigned(bits, SB_SUBMODE_BITS); |
1400 | | /*speex_mode_query(&speex_wb_mode, SPEEX_SUBMODE_BITS_PER_FRAME, &advance);*/ |
1401 | 2.34k | advance = wb_skip_table[submode]; |
1402 | 2.34k | if (advance < 0) |
1403 | 0 | { |
1404 | 0 | speex_notify("Invalid mode encountered. The stream is corrupted."); |
1405 | 0 | return -2; |
1406 | 0 | } |
1407 | 2.34k | advance -= (SB_SUBMODE_BITS+1); |
1408 | 2.34k | speex_bits_advance(bits, advance); |
1409 | 2.34k | wideband = speex_bits_unpack_unsigned(bits, 1); |
1410 | 2.34k | if (wideband) |
1411 | 1.89k | { |
1412 | 1.89k | speex_notify("More than two wideband layers found. The stream is corrupted."); |
1413 | 1.89k | return -2; |
1414 | 1.89k | } |
1415 | | |
1416 | 2.34k | } |
1417 | 2.99k | } |
1418 | 57.2k | if (speex_bits_remaining(bits)<4) |
1419 | 427 | return -1; |
1420 | | /* FIXME: Check for overflow */ |
1421 | 56.8k | m = speex_bits_unpack_unsigned(bits, 4); |
1422 | 56.8k | if (m==15) /* We found a terminator */ |
1423 | 968 | { |
1424 | 968 | return -1; |
1425 | 55.9k | } else if (m==14) /* Speex in-band request */ |
1426 | 4.36k | { |
1427 | 4.36k | int ret = speex_inband_handler(bits, st->speex_callbacks, state); |
1428 | 4.36k | if (ret) |
1429 | 0 | return ret; |
1430 | 51.5k | } else if (m==13) /* User in-band request */ |
1431 | 687 | { |
1432 | 687 | int ret = st->user_callback.func(bits, state, st->user_callback.data); |
1433 | 687 | if (ret) |
1434 | 0 | return ret; |
1435 | 50.8k | } else if (m>8) /* Invalid mode */ |
1436 | 3.09k | { |
1437 | 3.09k | speex_notify("Invalid mode encountered. The stream is corrupted."); |
1438 | 3.09k | return -2; |
1439 | 3.09k | } |
1440 | | |
1441 | 56.8k | } while (m>8); |
1442 | | |
1443 | | /* Get the sub-mode that was used */ |
1444 | 47.7k | st->submodeID = m; |
1445 | 47.7k | } |
1446 | | |
1447 | 68.6k | } |
1448 | | |
1449 | | /* Shift all buffers by one frame */ |
1450 | 47.7k | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 12); |
1451 | | |
1452 | | /* If null mode (no transmission), just set a couple things to zero*/ |
1453 | 47.7k | if (st->submodes[st->submodeID] == NULL) |
1454 | 22.8k | { |
1455 | 22.8k | VARDECL(spx_coef_t *lpc); |
1456 | 22.8k | ALLOC(lpc, NB_ORDER, spx_coef_t); |
1457 | 22.8k | bw_lpc(QCONST16(0.93f,15), st->interp_qlpc, lpc, NB_ORDER); |
1458 | 22.8k | { |
1459 | 22.8k | spx_word16_t innov_gain=0; |
1460 | | /* FIXME: This was innov, not exc */ |
1461 | 22.8k | innov_gain = compute_rms16(st->exc, NB_FRAME_SIZE); |
1462 | 3.67M | for (i=0;i<NB_FRAME_SIZE;i++) |
1463 | 3.65M | st->exc[i]=speex_rand(innov_gain, &st->seed); |
1464 | 22.8k | } |
1465 | | |
1466 | | |
1467 | 22.8k | st->first=1; |
1468 | | |
1469 | | /* Final signal synthesis from excitation */ |
1470 | 22.8k | iir_mem16(st->exc, lpc, out, NB_FRAME_SIZE, NB_ORDER, st->mem_sp, stack); |
1471 | | |
1472 | | /* Normally this is written to later but since this is returning early, |
1473 | | avoid reading uninitialized memory in caller */ |
1474 | 22.8k | if (st->innov_save) |
1475 | 18.8k | SPEEX_MEMSET(st->innov_save, 0, NB_NB_SUBFRAMES*NB_SUBFRAME_SIZE); |
1476 | | |
1477 | 22.8k | st->count_lost=0; |
1478 | 22.8k | return 0; |
1479 | 22.8k | } |
1480 | | |
1481 | 24.9k | ALLOC(qlsp, NB_ORDER, spx_lsp_t); |
1482 | | |
1483 | | /* Unquantize LSPs */ |
1484 | 24.9k | SUBMODE(lsp_unquant)(qlsp, NB_ORDER, bits); |
1485 | | |
1486 | | /*Damp memory if a frame was lost and the LSP changed too much*/ |
1487 | 24.9k | if (st->count_lost) |
1488 | 0 | { |
1489 | 0 | spx_word16_t fact; |
1490 | 0 | spx_word32_t lsp_dist=0; |
1491 | 0 | for (i=0;i<NB_ORDER;i++) |
1492 | 0 | lsp_dist = ADD32(lsp_dist, EXTEND32(ABS(st->old_qlsp[i] - qlsp[i]))); |
1493 | | #ifdef FIXED_POINT |
1494 | 0 | fact = SHR16(19661,SHR32(lsp_dist,LSP_SHIFT+2)); |
1495 | | #else |
1496 | | fact = .6*exp(-.2*lsp_dist); |
1497 | | #endif |
1498 | 0 | for (i=0;i<NB_ORDER;i++) |
1499 | 0 | st->mem_sp[i] = MULT16_32_Q15(fact,st->mem_sp[i]); |
1500 | 0 | } |
1501 | | |
1502 | | |
1503 | | /* Handle first frame and lost-packet case */ |
1504 | 24.9k | if (st->first || st->count_lost) |
1505 | 4.85k | { |
1506 | 53.4k | for (i=0;i<NB_ORDER;i++) |
1507 | 48.5k | st->old_qlsp[i] = qlsp[i]; |
1508 | 4.85k | } |
1509 | | |
1510 | | /* Get open-loop pitch estimation for low bit-rate pitch coding */ |
1511 | 24.9k | if (SUBMODE(lbr_pitch)!=-1) |
1512 | 10.1k | { |
1513 | 10.1k | ol_pitch = NB_PITCH_START+speex_bits_unpack_unsigned(bits, 7); |
1514 | 10.1k | } |
1515 | | |
1516 | 24.9k | if (SUBMODE(forced_pitch_gain)) |
1517 | 8.36k | { |
1518 | 8.36k | int quant; |
1519 | 8.36k | quant = speex_bits_unpack_unsigned(bits, 4); |
1520 | 8.36k | ol_pitch_coef=MULT16_16_P15(QCONST16(0.066667,15),SHL16(quant,GAIN_SHIFT)); |
1521 | 8.36k | } |
1522 | | |
1523 | | /* Get global excitation gain */ |
1524 | 24.9k | { |
1525 | 24.9k | int qe; |
1526 | 24.9k | qe = speex_bits_unpack_unsigned(bits, 5); |
1527 | | #ifdef FIXED_POINT |
1528 | | /* FIXME: Perhaps we could slightly lower the gain here when the output is going to saturate? */ |
1529 | 13.8k | ol_gain = MULT16_32_Q15(28406,ol_gain_table[qe]); |
1530 | | #else |
1531 | 11.1k | ol_gain = SIG_SCALING*exp(qe/3.5); |
1532 | | #endif |
1533 | 24.9k | } |
1534 | | |
1535 | 24.9k | ALLOC(ak, NB_ORDER, spx_coef_t); |
1536 | 24.9k | ALLOC(innov, NB_SUBFRAME_SIZE, spx_sig_t); |
1537 | 24.9k | ALLOC(exc32, NB_SUBFRAME_SIZE, spx_word32_t); |
1538 | | |
1539 | 24.9k | if (st->submodeID==1) |
1540 | 4.23k | { |
1541 | 4.23k | int extra; |
1542 | 4.23k | extra = speex_bits_unpack_unsigned(bits, 4); |
1543 | | |
1544 | 4.23k | if (extra==15) |
1545 | 806 | st->dtx_enabled=1; |
1546 | 3.42k | else |
1547 | 3.42k | st->dtx_enabled=0; |
1548 | 4.23k | } |
1549 | 24.9k | if (st->submodeID>1) |
1550 | 20.6k | st->dtx_enabled=0; |
1551 | | |
1552 | | /*Loop on subframes */ |
1553 | 124k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) |
1554 | 99.7k | { |
1555 | 99.7k | int offset; |
1556 | 99.7k | spx_word16_t *exc; |
1557 | 99.7k | spx_word16_t *innov_save = NULL; |
1558 | 99.7k | spx_word16_t tmp; |
1559 | | |
1560 | | /* Offset relative to start of frame */ |
1561 | 99.7k | offset = NB_SUBFRAME_SIZE*sub; |
1562 | | /* Excitation */ |
1563 | 99.7k | exc=st->exc+offset; |
1564 | | /* Original signal */ |
1565 | 99.7k | if (st->innov_save) |
1566 | 70.9k | innov_save = st->innov_save+offset; |
1567 | | |
1568 | | |
1569 | | /* Reset excitation */ |
1570 | 99.7k | SPEEX_MEMSET(exc, 0, NB_SUBFRAME_SIZE); |
1571 | | |
1572 | | /*Adaptive codebook contribution*/ |
1573 | 99.7k | speex_assert (SUBMODE(ltp_unquant)); |
1574 | 99.7k | { |
1575 | 99.7k | int pit_min, pit_max; |
1576 | | /* Handle pitch constraints if any */ |
1577 | 99.7k | if (SUBMODE(lbr_pitch) != -1) |
1578 | 40.5k | { |
1579 | 40.5k | int margin; |
1580 | 40.5k | margin = SUBMODE(lbr_pitch); |
1581 | 40.5k | if (margin) |
1582 | 0 | { |
1583 | | /* GT - need optimization? |
1584 | | if (ol_pitch < NB_PITCH_START+margin-1) |
1585 | | ol_pitch=NB_PITCH_START+margin-1; |
1586 | | if (ol_pitch > NB_PITCH_END-margin) |
1587 | | ol_pitch=NB_PITCH_END-margin; |
1588 | | pit_min = ol_pitch-margin+1; |
1589 | | pit_max = ol_pitch+margin; |
1590 | | */ |
1591 | 0 | pit_min = ol_pitch-margin+1; |
1592 | 0 | if (pit_min < NB_PITCH_START) |
1593 | 0 | pit_min = NB_PITCH_START; |
1594 | 0 | pit_max = ol_pitch+margin; |
1595 | 0 | if (pit_max > NB_PITCH_END) |
1596 | 0 | pit_max = NB_PITCH_END; |
1597 | 40.5k | } else { |
1598 | 40.5k | pit_min = pit_max = ol_pitch; |
1599 | 40.5k | } |
1600 | 59.1k | } else { |
1601 | 59.1k | pit_min = NB_PITCH_START; |
1602 | 59.1k | pit_max = NB_PITCH_END; |
1603 | 59.1k | } |
1604 | | |
1605 | | |
1606 | | |
1607 | 99.7k | SUBMODE(ltp_unquant)(exc, exc32, pit_min, pit_max, ol_pitch_coef, SUBMODE(ltp_params), |
1608 | 99.7k | NB_SUBFRAME_SIZE, &pitch, &pitch_gain[0], bits, stack, |
1609 | 99.7k | st->count_lost, offset, st->last_pitch_gain, 0); |
1610 | | |
1611 | | /* Ensuring that things aren't blowing up as would happen if e.g. an encoder is |
1612 | | crafting packets to make us produce NaNs and slow down the decoder (vague DoS threat). |
1613 | | We can probably be even more aggressive and limit to 15000 or so. */ |
1614 | 99.7k | sanitize_values32(exc32, NEG32(QCONST32(32000,SIG_SHIFT-1)), QCONST32(32000,SIG_SHIFT-1), NB_SUBFRAME_SIZE); |
1615 | | |
1616 | 99.7k | tmp = gain_3tap_to_1tap(pitch_gain); |
1617 | | |
1618 | 99.7k | pitch_average += tmp; |
1619 | 99.7k | if ((tmp>best_pitch_gain&&ABS(2*best_pitch-pitch)>=3&&ABS(3*best_pitch-pitch)>=4&&ABS(4*best_pitch-pitch)>=5) |
1620 | 99.7k | || (tmp>MULT16_16_Q15(QCONST16(.6,15),best_pitch_gain)&&(ABS(best_pitch-2*pitch)<3||ABS(best_pitch-3*pitch)<4||ABS(best_pitch-4*pitch)<5)) |
1621 | 99.7k | || (MULT16_16_Q15(QCONST16(.67,15),tmp)>best_pitch_gain&&(ABS(2*best_pitch-pitch)<3||ABS(3*best_pitch-pitch)<4||ABS(4*best_pitch-pitch)<5)) ) |
1622 | 15.3k | { |
1623 | 15.3k | best_pitch = pitch; |
1624 | 15.3k | if (tmp > best_pitch_gain) |
1625 | 14.0k | best_pitch_gain = tmp; |
1626 | 15.3k | } |
1627 | 99.7k | } |
1628 | | |
1629 | | /* Unquantize the innovation */ |
1630 | 99.7k | { |
1631 | 99.7k | int q_energy; |
1632 | 99.7k | spx_word32_t ener; |
1633 | | |
1634 | 99.7k | SPEEX_MEMSET(innov, 0, NB_SUBFRAME_SIZE); |
1635 | | |
1636 | | /* Decode sub-frame gain correction */ |
1637 | 99.7k | if (SUBMODE(have_subframe_gain)==3) |
1638 | 22.5k | { |
1639 | 22.5k | q_energy = speex_bits_unpack_unsigned(bits, 3); |
1640 | 22.5k | ener = MULT16_32_Q14(exc_gain_quant_scal3[q_energy],ol_gain); |
1641 | 77.2k | } else if (SUBMODE(have_subframe_gain)==1) |
1642 | 36.6k | { |
1643 | 36.6k | q_energy = speex_bits_unpack_unsigned(bits, 1); |
1644 | 36.6k | ener = MULT16_32_Q14(exc_gain_quant_scal1[q_energy],ol_gain); |
1645 | 40.5k | } else { |
1646 | 40.5k | ener = ol_gain; |
1647 | 40.5k | } |
1648 | | |
1649 | 99.7k | speex_assert (SUBMODE(innovation_unquant)); |
1650 | 99.7k | { |
1651 | | /*Fixed codebook contribution*/ |
1652 | 99.7k | SUBMODE(innovation_unquant)(innov, SUBMODE(innovation_params), NB_SUBFRAME_SIZE, bits, stack, &st->seed); |
1653 | | /* De-normalize innovation and update excitation */ |
1654 | | |
1655 | 99.7k | signal_mul(innov, innov, ener, NB_SUBFRAME_SIZE); |
1656 | | |
1657 | | /* Decode second codebook (only for some modes) */ |
1658 | 99.7k | if (SUBMODE(double_codebook)) |
1659 | 5.88k | { |
1660 | 5.88k | char *tmp_stack=stack; |
1661 | 5.88k | VARDECL(spx_sig_t *innov2); |
1662 | 5.88k | ALLOC(innov2, NB_SUBFRAME_SIZE, spx_sig_t); |
1663 | 5.88k | SPEEX_MEMSET(innov2, 0, NB_SUBFRAME_SIZE); |
1664 | 5.88k | SUBMODE(innovation_unquant)(innov2, SUBMODE(innovation_params), NB_SUBFRAME_SIZE, bits, stack, &st->seed); |
1665 | 5.88k | signal_mul(innov2, innov2, MULT16_32_Q15(QCONST16(0.454545f,15),ener), NB_SUBFRAME_SIZE); |
1666 | 241k | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1667 | 235k | innov[i] = ADD32(innov[i], innov2[i]); |
1668 | 5.88k | stack = tmp_stack; |
1669 | 5.88k | } |
1670 | 4.08M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1671 | 3.98M | exc[i]=EXTRACT16(SATURATE32(PSHR32(ADD32(SHL32(exc32[i],1),innov[i]),SIG_SHIFT),32767)); |
1672 | | /*print_vec(exc, 40, "innov");*/ |
1673 | 99.7k | if (innov_save) |
1674 | 70.9k | { |
1675 | 2.90M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1676 | 2.83M | innov_save[i] = EXTRACT16(PSHR32(innov[i], SIG_SHIFT)); |
1677 | 70.9k | } |
1678 | 99.7k | } |
1679 | | |
1680 | | /*Vocoder mode*/ |
1681 | 99.7k | if (st->submodeID==1) |
1682 | 16.9k | { |
1683 | 16.9k | spx_word16_t g=ol_pitch_coef; |
1684 | 16.9k | g=MULT16_16_P14(QCONST16(1.5f,14),(g-QCONST16(.2f,6))); |
1685 | 16.9k | if (g<0) |
1686 | 9.49k | g=0; |
1687 | 16.9k | if (g>GAIN_SCALING) |
1688 | 2.22k | g=GAIN_SCALING; |
1689 | | |
1690 | 16.9k | SPEEX_MEMSET(exc, 0, NB_SUBFRAME_SIZE); |
1691 | 40.1k | while (st->voc_offset<NB_SUBFRAME_SIZE) |
1692 | 23.2k | { |
1693 | | /* exc[st->voc_offset]= g*sqrt(2*ol_pitch)*ol_gain; |
1694 | | Not quite sure why we need the factor of two in the sqrt */ |
1695 | 23.2k | if (st->voc_offset>=0) |
1696 | 23.2k | exc[st->voc_offset]=MULT16_16(spx_sqrt(MULT16_16_16(2,ol_pitch)),EXTRACT16(PSHR32(MULT16_16(g,PSHR32(ol_gain,SIG_SHIFT)),6))); |
1697 | 23.2k | st->voc_offset+=ol_pitch; |
1698 | 23.2k | } |
1699 | 16.9k | st->voc_offset -= NB_SUBFRAME_SIZE; |
1700 | | |
1701 | 694k | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1702 | 677k | { |
1703 | 677k | spx_word16_t exci=exc[i]; |
1704 | 677k | exc[i]= ADD16(ADD16(MULT16_16_Q15(QCONST16(.7f,15),exc[i]) , MULT16_16_Q15(QCONST16(.3f,15),st->voc_m1)), |
1705 | 677k | SUB16(MULT16_16_Q15(Q15_ONE-MULT16_16_16(QCONST16(.85f,9),g),EXTRACT16(PSHR32(innov[i],SIG_SHIFT))), |
1706 | 677k | MULT16_16_Q15(MULT16_16_16(QCONST16(.15f,9),g),EXTRACT16(PSHR32(st->voc_m2,SIG_SHIFT))) |
1707 | 677k | )); |
1708 | 677k | st->voc_m1 = exci; |
1709 | 677k | st->voc_m2=innov[i]; |
1710 | 677k | st->voc_mean = EXTRACT16(PSHR32(ADD32(MULT16_16(QCONST16(.8f,15),st->voc_mean), MULT16_16(QCONST16(.2f,15),exc[i])), 15)); |
1711 | 677k | exc[i]-=st->voc_mean; |
1712 | 677k | } |
1713 | 16.9k | } |
1714 | | |
1715 | 99.7k | } |
1716 | 99.7k | } |
1717 | | |
1718 | 24.9k | ALLOC(interp_qlsp, NB_ORDER, spx_lsp_t); |
1719 | | |
1720 | 24.9k | if (st->lpc_enh_enabled && SUBMODE(comb_gain)>0 && !st->count_lost) |
1721 | 20.6k | { |
1722 | 20.6k | multicomb(st->exc-NB_SUBFRAME_SIZE, out, st->interp_qlpc, NB_ORDER, 2*NB_SUBFRAME_SIZE, best_pitch, 40, SUBMODE(comb_gain), stack); |
1723 | 20.6k | multicomb(st->exc+NB_SUBFRAME_SIZE, out+2*NB_SUBFRAME_SIZE, st->interp_qlpc, NB_ORDER, 2*NB_SUBFRAME_SIZE, best_pitch, 40, SUBMODE(comb_gain), stack); |
1724 | 20.6k | } else { |
1725 | 4.23k | SPEEX_COPY(out, &st->exc[-NB_SUBFRAME_SIZE], NB_FRAME_SIZE); |
1726 | 4.23k | } |
1727 | | |
1728 | | /* If the last packet was lost, re-scale the excitation to obtain the same energy as encoded in ol_gain */ |
1729 | 24.9k | if (st->count_lost) |
1730 | 0 | { |
1731 | 0 | spx_word16_t exc_ener; |
1732 | 0 | spx_word32_t gain32; |
1733 | 0 | spx_word16_t gain; |
1734 | 0 | exc_ener = compute_rms16 (st->exc, NB_FRAME_SIZE); |
1735 | 0 | gain32 = PDIV32(ol_gain, ADD16(exc_ener,1)); |
1736 | | #ifdef FIXED_POINT |
1737 | 0 | if (gain32 > 32767) |
1738 | 0 | gain32 = 32767; |
1739 | 0 | gain = EXTRACT16(gain32); |
1740 | | #else |
1741 | 0 | if (gain32 > 2) |
1742 | 0 | gain32=2; |
1743 | | gain = gain32; |
1744 | | #endif |
1745 | 0 | for (i=0;i<NB_FRAME_SIZE;i++) |
1746 | 0 | { |
1747 | 0 | st->exc[i] = MULT16_16_Q14(gain, st->exc[i]); |
1748 | 0 | out[i]=st->exc[i-NB_SUBFRAME_SIZE]; |
1749 | 0 | } |
1750 | 0 | } |
1751 | | |
1752 | | /*Loop on subframes */ |
1753 | 124k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) |
1754 | 99.7k | { |
1755 | 99.7k | int offset; |
1756 | 99.7k | spx_word16_t *sp; |
1757 | | |
1758 | | /* Offset relative to start of frame */ |
1759 | 99.7k | offset = NB_SUBFRAME_SIZE*sub; |
1760 | | /* Original signal */ |
1761 | 99.7k | sp=out+offset; |
1762 | | |
1763 | | /* LSP interpolation (quantized and unquantized) */ |
1764 | 99.7k | lsp_interpolate(st->old_qlsp, qlsp, interp_qlsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); |
1765 | | |
1766 | | /* Compute interpolated LPCs (unquantized) */ |
1767 | 99.7k | lsp_to_lpc(interp_qlsp, ak, NB_ORDER, stack); |
1768 | | |
1769 | | /* Compute analysis filter at w=pi */ |
1770 | 99.7k | { |
1771 | 99.7k | spx_word32_t pi_g=LPC_SCALING; |
1772 | 598k | for (i=0;i<NB_ORDER;i+=2) |
1773 | 498k | { |
1774 | | /*pi_g += -st->interp_qlpc[i] + st->interp_qlpc[i+1];*/ |
1775 | 498k | pi_g = ADD32(pi_g, SUB32(EXTEND32(ak[i+1]),EXTEND32(ak[i]))); |
1776 | 498k | } |
1777 | 99.7k | st->pi_gain[sub] = pi_g; |
1778 | 99.7k | } |
1779 | | |
1780 | 99.7k | iir_mem16(sp, st->interp_qlpc, sp, NB_SUBFRAME_SIZE, NB_ORDER, |
1781 | 99.7k | st->mem_sp, stack); |
1782 | | |
1783 | 1.09M | for (i=0;i<NB_ORDER;i++) |
1784 | 997k | st->interp_qlpc[i] = ak[i]; |
1785 | | |
1786 | 99.7k | } |
1787 | | |
1788 | 24.9k | if (st->highpass_enabled) |
1789 | 24.9k | highpass(out, out, NB_FRAME_SIZE, (st->isWideband?HIGHPASS_WIDEBAND:HIGHPASS_NARROWBAND)|HIGHPASS_OUTPUT, st->mem_hp); |
1790 | | /*for (i=0;i<NB_FRAME_SIZE;i++) |
1791 | | printf ("%d\n", (int)st->frame[i]);*/ |
1792 | | |
1793 | | /* Tracking output level */ |
1794 | 24.9k | st->level = 1+PSHR32(ol_gain,SIG_SHIFT); |
1795 | 24.9k | st->max_level = MAX16(MULT16_16_Q15(QCONST16(.99f,15), st->max_level), st->level); |
1796 | 24.9k | st->min_level = MIN16(ADD16(1,MULT16_16_Q14(QCONST16(1.01f,14), st->min_level)), st->level); |
1797 | 24.9k | if (st->max_level < st->min_level+1) |
1798 | 7.98k | st->max_level = st->min_level+1; |
1799 | | /*printf ("%f %f %f %d\n", og, st->min_level, st->max_level, update);*/ |
1800 | | |
1801 | | /* Store the LSPs for interpolation in the next frame */ |
1802 | 274k | for (i=0;i<NB_ORDER;i++) |
1803 | 249k | st->old_qlsp[i] = qlsp[i]; |
1804 | | |
1805 | | /* The next frame will not be the first (Duh!) */ |
1806 | 24.9k | st->first = 0; |
1807 | 24.9k | st->count_lost=0; |
1808 | 24.9k | st->last_pitch = best_pitch; |
1809 | | #ifdef FIXED_POINT |
1810 | 13.8k | st->last_pitch_gain = PSHR16(pitch_average,2); |
1811 | | #else |
1812 | | st->last_pitch_gain = .25*pitch_average; |
1813 | | #endif |
1814 | 24.9k | st->pitch_gain_buf[st->pitch_gain_buf_idx++] = st->last_pitch_gain; |
1815 | 24.9k | if (st->pitch_gain_buf_idx > 2) /* rollover */ |
1816 | 7.41k | st->pitch_gain_buf_idx = 0; |
1817 | | |
1818 | 24.9k | st->last_ol_gain = ol_gain; |
1819 | | |
1820 | 24.9k | return 0; |
1821 | 47.7k | } Line | Count | Source | 1331 | 31.9k | { | 1332 | 31.9k | DecState *st; | 1333 | 31.9k | int i, sub; | 1334 | 31.9k | int pitch; | 1335 | 31.9k | spx_word16_t pitch_gain[3]; | 1336 | 31.9k | spx_word32_t ol_gain=0; | 1337 | 31.9k | int ol_pitch=0; | 1338 | 31.9k | spx_word16_t ol_pitch_coef=0; | 1339 | 31.9k | int best_pitch=40; | 1340 | 31.9k | spx_word16_t best_pitch_gain=0; | 1341 | 31.9k | int wideband; | 1342 | 31.9k | int m; | 1343 | 31.9k | char *stack; | 1344 | 31.9k | VARDECL(spx_sig_t *innov); | 1345 | 31.9k | VARDECL(spx_word32_t *exc32); | 1346 | 31.9k | VARDECL(spx_coef_t *ak); | 1347 | 31.9k | VARDECL(spx_lsp_t *qlsp); | 1348 | 31.9k | spx_word16_t pitch_average=0; | 1349 | | | 1350 | 31.9k | spx_word16_t *out = (spx_word16_t*)vout; | 1351 | 31.9k | VARDECL(spx_lsp_t *interp_qlsp); | 1352 | | | 1353 | 31.9k | st=(DecState*)state; | 1354 | 31.9k | stack=st->stack; | 1355 | | | 1356 | 31.9k | st->exc = st->excBuf + 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 6; | 1357 | | | 1358 | | /* Check if we're in DTX mode*/ | 1359 | 31.9k | if (!bits && st->dtx_enabled) | 1360 | 0 | { | 1361 | 0 | st->submodeID=0; | 1362 | 0 | } else | 1363 | 31.9k | { | 1364 | | /* If bits is NULL, consider the packet to be lost (what could we do anyway) */ | 1365 | 31.9k | if (!bits) | 1366 | 0 | { | 1367 | 0 | nb_decode_lost(st, out, stack); | 1368 | 0 | return 0; | 1369 | 0 | } | 1370 | | | 1371 | 31.9k | if (st->encode_submode) | 1372 | 31.9k | { | 1373 | | | 1374 | | /* Search for next narrowband block (handle requests, skip wideband blocks) */ | 1375 | 34.4k | do { | 1376 | 34.4k | if (speex_bits_remaining(bits)<5) | 1377 | 7.98k | return -1; | 1378 | 26.5k | wideband = speex_bits_unpack_unsigned(bits, 1); | 1379 | 26.5k | if (wideband) /* Skip wideband block (for compatibility) */ | 1380 | 1.80k | { | 1381 | 1.80k | int submode; | 1382 | 1.80k | int advance; | 1383 | 1.80k | advance = submode = speex_bits_unpack_unsigned(bits, SB_SUBMODE_BITS); | 1384 | | /*speex_mode_query(&speex_wb_mode, SPEEX_SUBMODE_BITS_PER_FRAME, &advance);*/ | 1385 | 1.80k | advance = wb_skip_table[submode]; | 1386 | 1.80k | if (advance < 0) | 1387 | 0 | { | 1388 | 0 | speex_notify("Invalid mode encountered. The stream is corrupted."); | 1389 | 0 | return -2; | 1390 | 0 | } | 1391 | 1.80k | advance -= (SB_SUBMODE_BITS+1); | 1392 | 1.80k | speex_bits_advance(bits, advance); | 1393 | | | 1394 | 1.80k | if (speex_bits_remaining(bits)<5) | 1395 | 438 | return -1; | 1396 | 1.36k | wideband = speex_bits_unpack_unsigned(bits, 1); | 1397 | 1.36k | if (wideband) | 1398 | 1.07k | { | 1399 | 1.07k | advance = submode = speex_bits_unpack_unsigned(bits, SB_SUBMODE_BITS); | 1400 | | /*speex_mode_query(&speex_wb_mode, SPEEX_SUBMODE_BITS_PER_FRAME, &advance);*/ | 1401 | 1.07k | advance = wb_skip_table[submode]; | 1402 | 1.07k | if (advance < 0) | 1403 | 0 | { | 1404 | 0 | speex_notify("Invalid mode encountered. The stream is corrupted."); | 1405 | 0 | return -2; | 1406 | 0 | } | 1407 | 1.07k | advance -= (SB_SUBMODE_BITS+1); | 1408 | 1.07k | speex_bits_advance(bits, advance); | 1409 | 1.07k | wideband = speex_bits_unpack_unsigned(bits, 1); | 1410 | 1.07k | if (wideband) | 1411 | 857 | { | 1412 | 857 | speex_notify("More than two wideband layers found. The stream is corrupted."); | 1413 | 857 | return -2; | 1414 | 857 | } | 1415 | | | 1416 | 1.07k | } | 1417 | 1.36k | } | 1418 | 25.2k | if (speex_bits_remaining(bits)<4) | 1419 | 218 | return -1; | 1420 | | /* FIXME: Check for overflow */ | 1421 | 24.9k | m = speex_bits_unpack_unsigned(bits, 4); | 1422 | 24.9k | if (m==15) /* We found a terminator */ | 1423 | 431 | { | 1424 | 431 | return -1; | 1425 | 24.5k | } else if (m==14) /* Speex in-band request */ | 1426 | 2.23k | { | 1427 | 2.23k | int ret = speex_inband_handler(bits, st->speex_callbacks, state); | 1428 | 2.23k | if (ret) | 1429 | 0 | return ret; | 1430 | 22.3k | } else if (m==13) /* User in-band request */ | 1431 | 353 | { | 1432 | 353 | int ret = st->user_callback.func(bits, state, st->user_callback.data); | 1433 | 353 | if (ret) | 1434 | 0 | return ret; | 1435 | 21.9k | } else if (m>8) /* Invalid mode */ | 1436 | 1.28k | { | 1437 | 1.28k | speex_notify("Invalid mode encountered. The stream is corrupted."); | 1438 | 1.28k | return -2; | 1439 | 1.28k | } | 1440 | | | 1441 | 24.9k | } while (m>8); | 1442 | | | 1443 | | /* Get the sub-mode that was used */ | 1444 | 20.7k | st->submodeID = m; | 1445 | 20.7k | } | 1446 | | | 1447 | 31.9k | } | 1448 | | | 1449 | | /* Shift all buffers by one frame */ | 1450 | 20.7k | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 12); | 1451 | | | 1452 | | /* If null mode (no transmission), just set a couple things to zero*/ | 1453 | 20.7k | if (st->submodes[st->submodeID] == NULL) | 1454 | 9.59k | { | 1455 | 9.59k | VARDECL(spx_coef_t *lpc); | 1456 | 9.59k | ALLOC(lpc, NB_ORDER, spx_coef_t); | 1457 | 9.59k | bw_lpc(QCONST16(0.93f,15), st->interp_qlpc, lpc, NB_ORDER); | 1458 | 9.59k | { | 1459 | 9.59k | spx_word16_t innov_gain=0; | 1460 | | /* FIXME: This was innov, not exc */ | 1461 | 9.59k | innov_gain = compute_rms16(st->exc, NB_FRAME_SIZE); | 1462 | 1.54M | for (i=0;i<NB_FRAME_SIZE;i++) | 1463 | 1.53M | st->exc[i]=speex_rand(innov_gain, &st->seed); | 1464 | 9.59k | } | 1465 | | | 1466 | | | 1467 | 9.59k | st->first=1; | 1468 | | | 1469 | | /* Final signal synthesis from excitation */ | 1470 | 9.59k | iir_mem16(st->exc, lpc, out, NB_FRAME_SIZE, NB_ORDER, st->mem_sp, stack); | 1471 | | | 1472 | | /* Normally this is written to later but since this is returning early, | 1473 | | avoid reading uninitialized memory in caller */ | 1474 | 9.59k | if (st->innov_save) | 1475 | 8.25k | SPEEX_MEMSET(st->innov_save, 0, NB_NB_SUBFRAMES*NB_SUBFRAME_SIZE); | 1476 | | | 1477 | 9.59k | st->count_lost=0; | 1478 | 9.59k | return 0; | 1479 | 9.59k | } | 1480 | | | 1481 | 11.1k | ALLOC(qlsp, NB_ORDER, spx_lsp_t); | 1482 | | | 1483 | | /* Unquantize LSPs */ | 1484 | 11.1k | SUBMODE(lsp_unquant)(qlsp, NB_ORDER, bits); | 1485 | | | 1486 | | /*Damp memory if a frame was lost and the LSP changed too much*/ | 1487 | 11.1k | if (st->count_lost) | 1488 | 0 | { | 1489 | 0 | spx_word16_t fact; | 1490 | 0 | spx_word32_t lsp_dist=0; | 1491 | 0 | for (i=0;i<NB_ORDER;i++) | 1492 | 0 | lsp_dist = ADD32(lsp_dist, EXTEND32(ABS(st->old_qlsp[i] - qlsp[i]))); | 1493 | | #ifdef FIXED_POINT | 1494 | | fact = SHR16(19661,SHR32(lsp_dist,LSP_SHIFT+2)); | 1495 | | #else | 1496 | 0 | fact = .6*exp(-.2*lsp_dist); | 1497 | 0 | #endif | 1498 | 0 | for (i=0;i<NB_ORDER;i++) | 1499 | 0 | st->mem_sp[i] = MULT16_32_Q15(fact,st->mem_sp[i]); | 1500 | 0 | } | 1501 | | | 1502 | | | 1503 | | /* Handle first frame and lost-packet case */ | 1504 | 11.1k | if (st->first || st->count_lost) | 1505 | 1.91k | { | 1506 | 21.0k | for (i=0;i<NB_ORDER;i++) | 1507 | 19.1k | st->old_qlsp[i] = qlsp[i]; | 1508 | 1.91k | } | 1509 | | | 1510 | | /* Get open-loop pitch estimation for low bit-rate pitch coding */ | 1511 | 11.1k | if (SUBMODE(lbr_pitch)!=-1) | 1512 | 4.72k | { | 1513 | 4.72k | ol_pitch = NB_PITCH_START+speex_bits_unpack_unsigned(bits, 7); | 1514 | 4.72k | } | 1515 | | | 1516 | 11.1k | if (SUBMODE(forced_pitch_gain)) | 1517 | 4.04k | { | 1518 | 4.04k | int quant; | 1519 | 4.04k | quant = speex_bits_unpack_unsigned(bits, 4); | 1520 | 4.04k | ol_pitch_coef=MULT16_16_P15(QCONST16(0.066667,15),SHL16(quant,GAIN_SHIFT)); | 1521 | 4.04k | } | 1522 | | | 1523 | | /* Get global excitation gain */ | 1524 | 11.1k | { | 1525 | 11.1k | int qe; | 1526 | 11.1k | qe = speex_bits_unpack_unsigned(bits, 5); | 1527 | | #ifdef FIXED_POINT | 1528 | | /* FIXME: Perhaps we could slightly lower the gain here when the output is going to saturate? */ | 1529 | | ol_gain = MULT16_32_Q15(28406,ol_gain_table[qe]); | 1530 | | #else | 1531 | 11.1k | ol_gain = SIG_SCALING*exp(qe/3.5); | 1532 | 11.1k | #endif | 1533 | 11.1k | } | 1534 | | | 1535 | 11.1k | ALLOC(ak, NB_ORDER, spx_coef_t); | 1536 | 11.1k | ALLOC(innov, NB_SUBFRAME_SIZE, spx_sig_t); | 1537 | 11.1k | ALLOC(exc32, NB_SUBFRAME_SIZE, spx_word32_t); | 1538 | | | 1539 | 11.1k | if (st->submodeID==1) | 1540 | 1.91k | { | 1541 | 1.91k | int extra; | 1542 | 1.91k | extra = speex_bits_unpack_unsigned(bits, 4); | 1543 | | | 1544 | 1.91k | if (extra==15) | 1545 | 290 | st->dtx_enabled=1; | 1546 | 1.62k | else | 1547 | 1.62k | st->dtx_enabled=0; | 1548 | 1.91k | } | 1549 | 11.1k | if (st->submodeID>1) | 1550 | 9.18k | st->dtx_enabled=0; | 1551 | | | 1552 | | /*Loop on subframes */ | 1553 | 55.5k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) | 1554 | 44.4k | { | 1555 | 44.4k | int offset; | 1556 | 44.4k | spx_word16_t *exc; | 1557 | 44.4k | spx_word16_t *innov_save = NULL; | 1558 | 44.4k | spx_word16_t tmp; | 1559 | | | 1560 | | /* Offset relative to start of frame */ | 1561 | 44.4k | offset = NB_SUBFRAME_SIZE*sub; | 1562 | | /* Excitation */ | 1563 | 44.4k | exc=st->exc+offset; | 1564 | | /* Original signal */ | 1565 | 44.4k | if (st->innov_save) | 1566 | 30.7k | innov_save = st->innov_save+offset; | 1567 | | | 1568 | | | 1569 | | /* Reset excitation */ | 1570 | 44.4k | SPEEX_MEMSET(exc, 0, NB_SUBFRAME_SIZE); | 1571 | | | 1572 | | /*Adaptive codebook contribution*/ | 1573 | 44.4k | speex_assert (SUBMODE(ltp_unquant)); | 1574 | 44.4k | { | 1575 | 44.4k | int pit_min, pit_max; | 1576 | | /* Handle pitch constraints if any */ | 1577 | 44.4k | if (SUBMODE(lbr_pitch) != -1) | 1578 | 18.8k | { | 1579 | 18.8k | int margin; | 1580 | 18.8k | margin = SUBMODE(lbr_pitch); | 1581 | 18.8k | if (margin) | 1582 | 0 | { | 1583 | | /* GT - need optimization? | 1584 | | if (ol_pitch < NB_PITCH_START+margin-1) | 1585 | | ol_pitch=NB_PITCH_START+margin-1; | 1586 | | if (ol_pitch > NB_PITCH_END-margin) | 1587 | | ol_pitch=NB_PITCH_END-margin; | 1588 | | pit_min = ol_pitch-margin+1; | 1589 | | pit_max = ol_pitch+margin; | 1590 | | */ | 1591 | 0 | pit_min = ol_pitch-margin+1; | 1592 | 0 | if (pit_min < NB_PITCH_START) | 1593 | 0 | pit_min = NB_PITCH_START; | 1594 | 0 | pit_max = ol_pitch+margin; | 1595 | 0 | if (pit_max > NB_PITCH_END) | 1596 | 0 | pit_max = NB_PITCH_END; | 1597 | 18.8k | } else { | 1598 | 18.8k | pit_min = pit_max = ol_pitch; | 1599 | 18.8k | } | 1600 | 25.5k | } else { | 1601 | 25.5k | pit_min = NB_PITCH_START; | 1602 | 25.5k | pit_max = NB_PITCH_END; | 1603 | 25.5k | } | 1604 | | | 1605 | | | 1606 | | | 1607 | 44.4k | SUBMODE(ltp_unquant)(exc, exc32, pit_min, pit_max, ol_pitch_coef, SUBMODE(ltp_params), | 1608 | 44.4k | NB_SUBFRAME_SIZE, &pitch, &pitch_gain[0], bits, stack, | 1609 | 44.4k | st->count_lost, offset, st->last_pitch_gain, 0); | 1610 | | | 1611 | | /* Ensuring that things aren't blowing up as would happen if e.g. an encoder is | 1612 | | crafting packets to make us produce NaNs and slow down the decoder (vague DoS threat). | 1613 | | We can probably be even more aggressive and limit to 15000 or so. */ | 1614 | 44.4k | sanitize_values32(exc32, NEG32(QCONST32(32000,SIG_SHIFT-1)), QCONST32(32000,SIG_SHIFT-1), NB_SUBFRAME_SIZE); | 1615 | | | 1616 | 44.4k | tmp = gain_3tap_to_1tap(pitch_gain); | 1617 | | | 1618 | 44.4k | pitch_average += tmp; | 1619 | 44.4k | if ((tmp>best_pitch_gain&&ABS(2*best_pitch-pitch)>=3&&ABS(3*best_pitch-pitch)>=4&&ABS(4*best_pitch-pitch)>=5) | 1620 | 44.4k | || (tmp>MULT16_16_Q15(QCONST16(.6,15),best_pitch_gain)&&(ABS(best_pitch-2*pitch)<3||ABS(best_pitch-3*pitch)<4||ABS(best_pitch-4*pitch)<5)) | 1621 | 44.4k | || (MULT16_16_Q15(QCONST16(.67,15),tmp)>best_pitch_gain&&(ABS(2*best_pitch-pitch)<3||ABS(3*best_pitch-pitch)<4||ABS(4*best_pitch-pitch)<5)) ) | 1622 | 6.10k | { | 1623 | 6.10k | best_pitch = pitch; | 1624 | 6.10k | if (tmp > best_pitch_gain) | 1625 | 5.47k | best_pitch_gain = tmp; | 1626 | 6.10k | } | 1627 | 44.4k | } | 1628 | | | 1629 | | /* Unquantize the innovation */ | 1630 | 44.4k | { | 1631 | 44.4k | int q_energy; | 1632 | 44.4k | spx_word32_t ener; | 1633 | | | 1634 | 44.4k | SPEEX_MEMSET(innov, 0, NB_SUBFRAME_SIZE); | 1635 | | | 1636 | | /* Decode sub-frame gain correction */ | 1637 | 44.4k | if (SUBMODE(have_subframe_gain)==3) | 1638 | 10.9k | { | 1639 | 10.9k | q_energy = speex_bits_unpack_unsigned(bits, 3); | 1640 | 10.9k | ener = MULT16_32_Q14(exc_gain_quant_scal3[q_energy],ol_gain); | 1641 | 33.4k | } else if (SUBMODE(have_subframe_gain)==1) | 1642 | 14.5k | { | 1643 | 14.5k | q_energy = speex_bits_unpack_unsigned(bits, 1); | 1644 | 14.5k | ener = MULT16_32_Q14(exc_gain_quant_scal1[q_energy],ol_gain); | 1645 | 18.8k | } else { | 1646 | 18.8k | ener = ol_gain; | 1647 | 18.8k | } | 1648 | | | 1649 | 44.4k | speex_assert (SUBMODE(innovation_unquant)); | 1650 | 44.4k | { | 1651 | | /*Fixed codebook contribution*/ | 1652 | 44.4k | SUBMODE(innovation_unquant)(innov, SUBMODE(innovation_params), NB_SUBFRAME_SIZE, bits, stack, &st->seed); | 1653 | | /* De-normalize innovation and update excitation */ | 1654 | | | 1655 | 44.4k | signal_mul(innov, innov, ener, NB_SUBFRAME_SIZE); | 1656 | | | 1657 | | /* Decode second codebook (only for some modes) */ | 1658 | 44.4k | if (SUBMODE(double_codebook)) | 1659 | 2.47k | { | 1660 | 2.47k | char *tmp_stack=stack; | 1661 | 2.47k | VARDECL(spx_sig_t *innov2); | 1662 | 2.47k | ALLOC(innov2, NB_SUBFRAME_SIZE, spx_sig_t); | 1663 | 2.47k | SPEEX_MEMSET(innov2, 0, NB_SUBFRAME_SIZE); | 1664 | 2.47k | SUBMODE(innovation_unquant)(innov2, SUBMODE(innovation_params), NB_SUBFRAME_SIZE, bits, stack, &st->seed); | 1665 | 2.47k | signal_mul(innov2, innov2, MULT16_32_Q15(QCONST16(0.454545f,15),ener), NB_SUBFRAME_SIZE); | 1666 | 101k | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1667 | 99.0k | innov[i] = ADD32(innov[i], innov2[i]); | 1668 | 2.47k | stack = tmp_stack; | 1669 | 2.47k | } | 1670 | 1.82M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1671 | 1.77M | exc[i]=EXTRACT16(SATURATE32(PSHR32(ADD32(SHL32(exc32[i],1),innov[i]),SIG_SHIFT),32767)); | 1672 | | /*print_vec(exc, 40, "innov");*/ | 1673 | 44.4k | if (innov_save) | 1674 | 30.7k | { | 1675 | 1.26M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1676 | 1.23M | innov_save[i] = EXTRACT16(PSHR32(innov[i], SIG_SHIFT)); | 1677 | 30.7k | } | 1678 | 44.4k | } | 1679 | | | 1680 | | /*Vocoder mode*/ | 1681 | 44.4k | if (st->submodeID==1) | 1682 | 7.66k | { | 1683 | 7.66k | spx_word16_t g=ol_pitch_coef; | 1684 | 7.66k | g=MULT16_16_P14(QCONST16(1.5f,14),(g-QCONST16(.2f,6))); | 1685 | 7.66k | if (g<0) | 1686 | 5.36k | g=0; | 1687 | 7.66k | if (g>GAIN_SCALING) | 1688 | 748 | g=GAIN_SCALING; | 1689 | | | 1690 | 7.66k | SPEEX_MEMSET(exc, 0, NB_SUBFRAME_SIZE); | 1691 | 20.4k | while (st->voc_offset<NB_SUBFRAME_SIZE) | 1692 | 12.7k | { | 1693 | | /* exc[st->voc_offset]= g*sqrt(2*ol_pitch)*ol_gain; | 1694 | | Not quite sure why we need the factor of two in the sqrt */ | 1695 | 12.7k | if (st->voc_offset>=0) | 1696 | 12.7k | exc[st->voc_offset]=MULT16_16(spx_sqrt(MULT16_16_16(2,ol_pitch)),EXTRACT16(PSHR32(MULT16_16(g,PSHR32(ol_gain,SIG_SHIFT)),6))); | 1697 | 12.7k | st->voc_offset+=ol_pitch; | 1698 | 12.7k | } | 1699 | 7.66k | st->voc_offset -= NB_SUBFRAME_SIZE; | 1700 | | | 1701 | 314k | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1702 | 306k | { | 1703 | 306k | spx_word16_t exci=exc[i]; | 1704 | 306k | exc[i]= ADD16(ADD16(MULT16_16_Q15(QCONST16(.7f,15),exc[i]) , MULT16_16_Q15(QCONST16(.3f,15),st->voc_m1)), | 1705 | 306k | SUB16(MULT16_16_Q15(Q15_ONE-MULT16_16_16(QCONST16(.85f,9),g),EXTRACT16(PSHR32(innov[i],SIG_SHIFT))), | 1706 | 306k | MULT16_16_Q15(MULT16_16_16(QCONST16(.15f,9),g),EXTRACT16(PSHR32(st->voc_m2,SIG_SHIFT))) | 1707 | 306k | )); | 1708 | 306k | st->voc_m1 = exci; | 1709 | 306k | st->voc_m2=innov[i]; | 1710 | 306k | st->voc_mean = EXTRACT16(PSHR32(ADD32(MULT16_16(QCONST16(.8f,15),st->voc_mean), MULT16_16(QCONST16(.2f,15),exc[i])), 15)); | 1711 | 306k | exc[i]-=st->voc_mean; | 1712 | 306k | } | 1713 | 7.66k | } | 1714 | | | 1715 | 44.4k | } | 1716 | 44.4k | } | 1717 | | | 1718 | 11.1k | ALLOC(interp_qlsp, NB_ORDER, spx_lsp_t); | 1719 | | | 1720 | 11.1k | if (st->lpc_enh_enabled && SUBMODE(comb_gain)>0 && !st->count_lost) | 1721 | 9.18k | { | 1722 | 9.18k | multicomb(st->exc-NB_SUBFRAME_SIZE, out, st->interp_qlpc, NB_ORDER, 2*NB_SUBFRAME_SIZE, best_pitch, 40, SUBMODE(comb_gain), stack); | 1723 | 9.18k | multicomb(st->exc+NB_SUBFRAME_SIZE, out+2*NB_SUBFRAME_SIZE, st->interp_qlpc, NB_ORDER, 2*NB_SUBFRAME_SIZE, best_pitch, 40, SUBMODE(comb_gain), stack); | 1724 | 9.18k | } else { | 1725 | 1.91k | SPEEX_COPY(out, &st->exc[-NB_SUBFRAME_SIZE], NB_FRAME_SIZE); | 1726 | 1.91k | } | 1727 | | | 1728 | | /* If the last packet was lost, re-scale the excitation to obtain the same energy as encoded in ol_gain */ | 1729 | 11.1k | if (st->count_lost) | 1730 | 0 | { | 1731 | 0 | spx_word16_t exc_ener; | 1732 | 0 | spx_word32_t gain32; | 1733 | 0 | spx_word16_t gain; | 1734 | 0 | exc_ener = compute_rms16 (st->exc, NB_FRAME_SIZE); | 1735 | 0 | gain32 = PDIV32(ol_gain, ADD16(exc_ener,1)); | 1736 | | #ifdef FIXED_POINT | 1737 | | if (gain32 > 32767) | 1738 | | gain32 = 32767; | 1739 | | gain = EXTRACT16(gain32); | 1740 | | #else | 1741 | 0 | if (gain32 > 2) | 1742 | 0 | gain32=2; | 1743 | 0 | gain = gain32; | 1744 | 0 | #endif | 1745 | 0 | for (i=0;i<NB_FRAME_SIZE;i++) | 1746 | 0 | { | 1747 | 0 | st->exc[i] = MULT16_16_Q14(gain, st->exc[i]); | 1748 | 0 | out[i]=st->exc[i-NB_SUBFRAME_SIZE]; | 1749 | 0 | } | 1750 | 0 | } | 1751 | | | 1752 | | /*Loop on subframes */ | 1753 | 55.5k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) | 1754 | 44.4k | { | 1755 | 44.4k | int offset; | 1756 | 44.4k | spx_word16_t *sp; | 1757 | | | 1758 | | /* Offset relative to start of frame */ | 1759 | 44.4k | offset = NB_SUBFRAME_SIZE*sub; | 1760 | | /* Original signal */ | 1761 | 44.4k | sp=out+offset; | 1762 | | | 1763 | | /* LSP interpolation (quantized and unquantized) */ | 1764 | 44.4k | lsp_interpolate(st->old_qlsp, qlsp, interp_qlsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); | 1765 | | | 1766 | | /* Compute interpolated LPCs (unquantized) */ | 1767 | 44.4k | lsp_to_lpc(interp_qlsp, ak, NB_ORDER, stack); | 1768 | | | 1769 | | /* Compute analysis filter at w=pi */ | 1770 | 44.4k | { | 1771 | 44.4k | spx_word32_t pi_g=LPC_SCALING; | 1772 | 266k | for (i=0;i<NB_ORDER;i+=2) | 1773 | 222k | { | 1774 | | /*pi_g += -st->interp_qlpc[i] + st->interp_qlpc[i+1];*/ | 1775 | 222k | pi_g = ADD32(pi_g, SUB32(EXTEND32(ak[i+1]),EXTEND32(ak[i]))); | 1776 | 222k | } | 1777 | 44.4k | st->pi_gain[sub] = pi_g; | 1778 | 44.4k | } | 1779 | | | 1780 | 44.4k | iir_mem16(sp, st->interp_qlpc, sp, NB_SUBFRAME_SIZE, NB_ORDER, | 1781 | 44.4k | st->mem_sp, stack); | 1782 | | | 1783 | 488k | for (i=0;i<NB_ORDER;i++) | 1784 | 444k | st->interp_qlpc[i] = ak[i]; | 1785 | | | 1786 | 44.4k | } | 1787 | | | 1788 | 11.1k | if (st->highpass_enabled) | 1789 | 11.1k | highpass(out, out, NB_FRAME_SIZE, (st->isWideband?HIGHPASS_WIDEBAND:HIGHPASS_NARROWBAND)|HIGHPASS_OUTPUT, st->mem_hp); | 1790 | | /*for (i=0;i<NB_FRAME_SIZE;i++) | 1791 | | printf ("%d\n", (int)st->frame[i]);*/ | 1792 | | | 1793 | | /* Tracking output level */ | 1794 | 11.1k | st->level = 1+PSHR32(ol_gain,SIG_SHIFT); | 1795 | 11.1k | st->max_level = MAX16(MULT16_16_Q15(QCONST16(.99f,15), st->max_level), st->level); | 1796 | 11.1k | st->min_level = MIN16(ADD16(1,MULT16_16_Q14(QCONST16(1.01f,14), st->min_level)), st->level); | 1797 | 11.1k | if (st->max_level < st->min_level+1) | 1798 | 3.41k | st->max_level = st->min_level+1; | 1799 | | /*printf ("%f %f %f %d\n", og, st->min_level, st->max_level, update);*/ | 1800 | | | 1801 | | /* Store the LSPs for interpolation in the next frame */ | 1802 | 122k | for (i=0;i<NB_ORDER;i++) | 1803 | 111k | st->old_qlsp[i] = qlsp[i]; | 1804 | | | 1805 | | /* The next frame will not be the first (Duh!) */ | 1806 | 11.1k | st->first = 0; | 1807 | 11.1k | st->count_lost=0; | 1808 | 11.1k | st->last_pitch = best_pitch; | 1809 | | #ifdef FIXED_POINT | 1810 | | st->last_pitch_gain = PSHR16(pitch_average,2); | 1811 | | #else | 1812 | 11.1k | st->last_pitch_gain = .25*pitch_average; | 1813 | 11.1k | #endif | 1814 | 11.1k | st->pitch_gain_buf[st->pitch_gain_buf_idx++] = st->last_pitch_gain; | 1815 | 11.1k | if (st->pitch_gain_buf_idx > 2) /* rollover */ | 1816 | 3.43k | st->pitch_gain_buf_idx = 0; | 1817 | | | 1818 | 11.1k | st->last_ol_gain = ol_gain; | 1819 | | | 1820 | 11.1k | return 0; | 1821 | 20.7k | } |
Line | Count | Source | 1331 | 36.7k | { | 1332 | 36.7k | DecState *st; | 1333 | 36.7k | int i, sub; | 1334 | 36.7k | int pitch; | 1335 | 36.7k | spx_word16_t pitch_gain[3]; | 1336 | 36.7k | spx_word32_t ol_gain=0; | 1337 | 36.7k | int ol_pitch=0; | 1338 | 36.7k | spx_word16_t ol_pitch_coef=0; | 1339 | 36.7k | int best_pitch=40; | 1340 | 36.7k | spx_word16_t best_pitch_gain=0; | 1341 | 36.7k | int wideband; | 1342 | 36.7k | int m; | 1343 | 36.7k | char *stack; | 1344 | 36.7k | VARDECL(spx_sig_t *innov); | 1345 | 36.7k | VARDECL(spx_word32_t *exc32); | 1346 | 36.7k | VARDECL(spx_coef_t *ak); | 1347 | 36.7k | VARDECL(spx_lsp_t *qlsp); | 1348 | 36.7k | spx_word16_t pitch_average=0; | 1349 | | | 1350 | 36.7k | spx_word16_t *out = (spx_word16_t*)vout; | 1351 | 36.7k | VARDECL(spx_lsp_t *interp_qlsp); | 1352 | | | 1353 | 36.7k | st=(DecState*)state; | 1354 | 36.7k | stack=st->stack; | 1355 | | | 1356 | 36.7k | st->exc = st->excBuf + 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 6; | 1357 | | | 1358 | | /* Check if we're in DTX mode*/ | 1359 | 36.7k | if (!bits && st->dtx_enabled) | 1360 | 0 | { | 1361 | 0 | st->submodeID=0; | 1362 | 0 | } else | 1363 | 36.7k | { | 1364 | | /* If bits is NULL, consider the packet to be lost (what could we do anyway) */ | 1365 | 36.7k | if (!bits) | 1366 | 0 | { | 1367 | 0 | nb_decode_lost(st, out, stack); | 1368 | 0 | return 0; | 1369 | 0 | } | 1370 | | | 1371 | 36.7k | if (st->encode_submode) | 1372 | 36.7k | { | 1373 | | | 1374 | | /* Search for next narrowband block (handle requests, skip wideband blocks) */ | 1375 | 39.2k | do { | 1376 | 39.2k | if (speex_bits_remaining(bits)<5) | 1377 | 5.61k | return -1; | 1378 | 33.6k | wideband = speex_bits_unpack_unsigned(bits, 1); | 1379 | 33.6k | if (wideband) /* Skip wideband block (for compatibility) */ | 1380 | 2.12k | { | 1381 | 2.12k | int submode; | 1382 | 2.12k | int advance; | 1383 | 2.12k | advance = submode = speex_bits_unpack_unsigned(bits, SB_SUBMODE_BITS); | 1384 | | /*speex_mode_query(&speex_wb_mode, SPEEX_SUBMODE_BITS_PER_FRAME, &advance);*/ | 1385 | 2.12k | advance = wb_skip_table[submode]; | 1386 | 2.12k | if (advance < 0) | 1387 | 0 | { | 1388 | 0 | speex_notify("Invalid mode encountered. The stream is corrupted."); | 1389 | 0 | return -2; | 1390 | 0 | } | 1391 | 2.12k | advance -= (SB_SUBMODE_BITS+1); | 1392 | 2.12k | speex_bits_advance(bits, advance); | 1393 | | | 1394 | 2.12k | if (speex_bits_remaining(bits)<5) | 1395 | 498 | return -1; | 1396 | 1.62k | wideband = speex_bits_unpack_unsigned(bits, 1); | 1397 | 1.62k | if (wideband) | 1398 | 1.26k | { | 1399 | 1.26k | advance = submode = speex_bits_unpack_unsigned(bits, SB_SUBMODE_BITS); | 1400 | | /*speex_mode_query(&speex_wb_mode, SPEEX_SUBMODE_BITS_PER_FRAME, &advance);*/ | 1401 | 1.26k | advance = wb_skip_table[submode]; | 1402 | 1.26k | if (advance < 0) | 1403 | 0 | { | 1404 | 0 | speex_notify("Invalid mode encountered. The stream is corrupted."); | 1405 | 0 | return -2; | 1406 | 0 | } | 1407 | 1.26k | advance -= (SB_SUBMODE_BITS+1); | 1408 | 1.26k | speex_bits_advance(bits, advance); | 1409 | 1.26k | wideband = speex_bits_unpack_unsigned(bits, 1); | 1410 | 1.26k | if (wideband) | 1411 | 1.03k | { | 1412 | 1.03k | speex_notify("More than two wideband layers found. The stream is corrupted."); | 1413 | 1.03k | return -2; | 1414 | 1.03k | } | 1415 | | | 1416 | 1.26k | } | 1417 | 1.62k | } | 1418 | 32.0k | if (speex_bits_remaining(bits)<4) | 1419 | 209 | return -1; | 1420 | | /* FIXME: Check for overflow */ | 1421 | 31.8k | m = speex_bits_unpack_unsigned(bits, 4); | 1422 | 31.8k | if (m==15) /* We found a terminator */ | 1423 | 537 | { | 1424 | 537 | return -1; | 1425 | 31.3k | } else if (m==14) /* Speex in-band request */ | 1426 | 2.13k | { | 1427 | 2.13k | int ret = speex_inband_handler(bits, st->speex_callbacks, state); | 1428 | 2.13k | if (ret) | 1429 | 0 | return ret; | 1430 | 29.1k | } else if (m==13) /* User in-band request */ | 1431 | 334 | { | 1432 | 334 | int ret = st->user_callback.func(bits, state, st->user_callback.data); | 1433 | 334 | if (ret) | 1434 | 0 | return ret; | 1435 | 28.8k | } else if (m>8) /* Invalid mode */ | 1436 | 1.81k | { | 1437 | 1.81k | speex_notify("Invalid mode encountered. The stream is corrupted."); | 1438 | 1.81k | return -2; | 1439 | 1.81k | } | 1440 | | | 1441 | 31.8k | } while (m>8); | 1442 | | | 1443 | | /* Get the sub-mode that was used */ | 1444 | 27.0k | st->submodeID = m; | 1445 | 27.0k | } | 1446 | | | 1447 | 36.7k | } | 1448 | | | 1449 | | /* Shift all buffers by one frame */ | 1450 | 27.0k | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 12); | 1451 | | | 1452 | | /* If null mode (no transmission), just set a couple things to zero*/ | 1453 | 27.0k | if (st->submodes[st->submodeID] == NULL) | 1454 | 13.2k | { | 1455 | 13.2k | VARDECL(spx_coef_t *lpc); | 1456 | 13.2k | ALLOC(lpc, NB_ORDER, spx_coef_t); | 1457 | 13.2k | bw_lpc(QCONST16(0.93f,15), st->interp_qlpc, lpc, NB_ORDER); | 1458 | 13.2k | { | 1459 | 13.2k | spx_word16_t innov_gain=0; | 1460 | | /* FIXME: This was innov, not exc */ | 1461 | 13.2k | innov_gain = compute_rms16(st->exc, NB_FRAME_SIZE); | 1462 | 2.12M | for (i=0;i<NB_FRAME_SIZE;i++) | 1463 | 2.11M | st->exc[i]=speex_rand(innov_gain, &st->seed); | 1464 | 13.2k | } | 1465 | | | 1466 | | | 1467 | 13.2k | st->first=1; | 1468 | | | 1469 | | /* Final signal synthesis from excitation */ | 1470 | 13.2k | iir_mem16(st->exc, lpc, out, NB_FRAME_SIZE, NB_ORDER, st->mem_sp, stack); | 1471 | | | 1472 | | /* Normally this is written to later but since this is returning early, | 1473 | | avoid reading uninitialized memory in caller */ | 1474 | 13.2k | if (st->innov_save) | 1475 | 10.5k | SPEEX_MEMSET(st->innov_save, 0, NB_NB_SUBFRAMES*NB_SUBFRAME_SIZE); | 1476 | | | 1477 | 13.2k | st->count_lost=0; | 1478 | 13.2k | return 0; | 1479 | 13.2k | } | 1480 | | | 1481 | 13.8k | ALLOC(qlsp, NB_ORDER, spx_lsp_t); | 1482 | | | 1483 | | /* Unquantize LSPs */ | 1484 | 13.8k | SUBMODE(lsp_unquant)(qlsp, NB_ORDER, bits); | 1485 | | | 1486 | | /*Damp memory if a frame was lost and the LSP changed too much*/ | 1487 | 13.8k | if (st->count_lost) | 1488 | 0 | { | 1489 | 0 | spx_word16_t fact; | 1490 | 0 | spx_word32_t lsp_dist=0; | 1491 | 0 | for (i=0;i<NB_ORDER;i++) | 1492 | 0 | lsp_dist = ADD32(lsp_dist, EXTEND32(ABS(st->old_qlsp[i] - qlsp[i]))); | 1493 | 0 | #ifdef FIXED_POINT | 1494 | 0 | fact = SHR16(19661,SHR32(lsp_dist,LSP_SHIFT+2)); | 1495 | | #else | 1496 | | fact = .6*exp(-.2*lsp_dist); | 1497 | | #endif | 1498 | 0 | for (i=0;i<NB_ORDER;i++) | 1499 | 0 | st->mem_sp[i] = MULT16_32_Q15(fact,st->mem_sp[i]); | 1500 | 0 | } | 1501 | | | 1502 | | | 1503 | | /* Handle first frame and lost-packet case */ | 1504 | 13.8k | if (st->first || st->count_lost) | 1505 | 2.94k | { | 1506 | 32.3k | for (i=0;i<NB_ORDER;i++) | 1507 | 29.4k | st->old_qlsp[i] = qlsp[i]; | 1508 | 2.94k | } | 1509 | | | 1510 | | /* Get open-loop pitch estimation for low bit-rate pitch coding */ | 1511 | 13.8k | if (SUBMODE(lbr_pitch)!=-1) | 1512 | 5.42k | { | 1513 | 5.42k | ol_pitch = NB_PITCH_START+speex_bits_unpack_unsigned(bits, 7); | 1514 | 5.42k | } | 1515 | | | 1516 | 13.8k | if (SUBMODE(forced_pitch_gain)) | 1517 | 4.32k | { | 1518 | 4.32k | int quant; | 1519 | 4.32k | quant = speex_bits_unpack_unsigned(bits, 4); | 1520 | 4.32k | ol_pitch_coef=MULT16_16_P15(QCONST16(0.066667,15),SHL16(quant,GAIN_SHIFT)); | 1521 | 4.32k | } | 1522 | | | 1523 | | /* Get global excitation gain */ | 1524 | 13.8k | { | 1525 | 13.8k | int qe; | 1526 | 13.8k | qe = speex_bits_unpack_unsigned(bits, 5); | 1527 | 13.8k | #ifdef FIXED_POINT | 1528 | | /* FIXME: Perhaps we could slightly lower the gain here when the output is going to saturate? */ | 1529 | 13.8k | ol_gain = MULT16_32_Q15(28406,ol_gain_table[qe]); | 1530 | | #else | 1531 | | ol_gain = SIG_SCALING*exp(qe/3.5); | 1532 | | #endif | 1533 | 13.8k | } | 1534 | | | 1535 | 13.8k | ALLOC(ak, NB_ORDER, spx_coef_t); | 1536 | 13.8k | ALLOC(innov, NB_SUBFRAME_SIZE, spx_sig_t); | 1537 | 13.8k | ALLOC(exc32, NB_SUBFRAME_SIZE, spx_word32_t); | 1538 | | | 1539 | 13.8k | if (st->submodeID==1) | 1540 | 2.31k | { | 1541 | 2.31k | int extra; | 1542 | 2.31k | extra = speex_bits_unpack_unsigned(bits, 4); | 1543 | | | 1544 | 2.31k | if (extra==15) | 1545 | 516 | st->dtx_enabled=1; | 1546 | 1.80k | else | 1547 | 1.80k | st->dtx_enabled=0; | 1548 | 2.31k | } | 1549 | 13.8k | if (st->submodeID>1) | 1550 | 11.5k | st->dtx_enabled=0; | 1551 | | | 1552 | | /*Loop on subframes */ | 1553 | 69.1k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) | 1554 | 55.3k | { | 1555 | 55.3k | int offset; | 1556 | 55.3k | spx_word16_t *exc; | 1557 | 55.3k | spx_word16_t *innov_save = NULL; | 1558 | 55.3k | spx_word16_t tmp; | 1559 | | | 1560 | | /* Offset relative to start of frame */ | 1561 | 55.3k | offset = NB_SUBFRAME_SIZE*sub; | 1562 | | /* Excitation */ | 1563 | 55.3k | exc=st->exc+offset; | 1564 | | /* Original signal */ | 1565 | 55.3k | if (st->innov_save) | 1566 | 40.1k | innov_save = st->innov_save+offset; | 1567 | | | 1568 | | | 1569 | | /* Reset excitation */ | 1570 | 55.3k | SPEEX_MEMSET(exc, 0, NB_SUBFRAME_SIZE); | 1571 | | | 1572 | | /*Adaptive codebook contribution*/ | 1573 | 55.3k | speex_assert (SUBMODE(ltp_unquant)); | 1574 | 55.3k | { | 1575 | 55.3k | int pit_min, pit_max; | 1576 | | /* Handle pitch constraints if any */ | 1577 | 55.3k | if (SUBMODE(lbr_pitch) != -1) | 1578 | 21.7k | { | 1579 | 21.7k | int margin; | 1580 | 21.7k | margin = SUBMODE(lbr_pitch); | 1581 | 21.7k | if (margin) | 1582 | 0 | { | 1583 | | /* GT - need optimization? | 1584 | | if (ol_pitch < NB_PITCH_START+margin-1) | 1585 | | ol_pitch=NB_PITCH_START+margin-1; | 1586 | | if (ol_pitch > NB_PITCH_END-margin) | 1587 | | ol_pitch=NB_PITCH_END-margin; | 1588 | | pit_min = ol_pitch-margin+1; | 1589 | | pit_max = ol_pitch+margin; | 1590 | | */ | 1591 | 0 | pit_min = ol_pitch-margin+1; | 1592 | 0 | if (pit_min < NB_PITCH_START) | 1593 | 0 | pit_min = NB_PITCH_START; | 1594 | 0 | pit_max = ol_pitch+margin; | 1595 | 0 | if (pit_max > NB_PITCH_END) | 1596 | 0 | pit_max = NB_PITCH_END; | 1597 | 21.7k | } else { | 1598 | 21.7k | pit_min = pit_max = ol_pitch; | 1599 | 21.7k | } | 1600 | 33.6k | } else { | 1601 | 33.6k | pit_min = NB_PITCH_START; | 1602 | 33.6k | pit_max = NB_PITCH_END; | 1603 | 33.6k | } | 1604 | | | 1605 | | | 1606 | | | 1607 | 55.3k | SUBMODE(ltp_unquant)(exc, exc32, pit_min, pit_max, ol_pitch_coef, SUBMODE(ltp_params), | 1608 | 55.3k | NB_SUBFRAME_SIZE, &pitch, &pitch_gain[0], bits, stack, | 1609 | 55.3k | st->count_lost, offset, st->last_pitch_gain, 0); | 1610 | | | 1611 | | /* Ensuring that things aren't blowing up as would happen if e.g. an encoder is | 1612 | | crafting packets to make us produce NaNs and slow down the decoder (vague DoS threat). | 1613 | | We can probably be even more aggressive and limit to 15000 or so. */ | 1614 | 55.3k | sanitize_values32(exc32, NEG32(QCONST32(32000,SIG_SHIFT-1)), QCONST32(32000,SIG_SHIFT-1), NB_SUBFRAME_SIZE); | 1615 | | | 1616 | 55.3k | tmp = gain_3tap_to_1tap(pitch_gain); | 1617 | | | 1618 | 55.3k | pitch_average += tmp; | 1619 | 55.3k | if ((tmp>best_pitch_gain&&ABS(2*best_pitch-pitch)>=3&&ABS(3*best_pitch-pitch)>=4&&ABS(4*best_pitch-pitch)>=5) | 1620 | 55.3k | || (tmp>MULT16_16_Q15(QCONST16(.6,15),best_pitch_gain)&&(ABS(best_pitch-2*pitch)<3||ABS(best_pitch-3*pitch)<4||ABS(best_pitch-4*pitch)<5)) | 1621 | 55.3k | || (MULT16_16_Q15(QCONST16(.67,15),tmp)>best_pitch_gain&&(ABS(2*best_pitch-pitch)<3||ABS(3*best_pitch-pitch)<4||ABS(4*best_pitch-pitch)<5)) ) | 1622 | 9.27k | { | 1623 | 9.27k | best_pitch = pitch; | 1624 | 9.27k | if (tmp > best_pitch_gain) | 1625 | 8.56k | best_pitch_gain = tmp; | 1626 | 9.27k | } | 1627 | 55.3k | } | 1628 | | | 1629 | | /* Unquantize the innovation */ | 1630 | 55.3k | { | 1631 | 55.3k | int q_energy; | 1632 | 55.3k | spx_word32_t ener; | 1633 | | | 1634 | 55.3k | SPEEX_MEMSET(innov, 0, NB_SUBFRAME_SIZE); | 1635 | | | 1636 | | /* Decode sub-frame gain correction */ | 1637 | 55.3k | if (SUBMODE(have_subframe_gain)==3) | 1638 | 11.5k | { | 1639 | 11.5k | q_energy = speex_bits_unpack_unsigned(bits, 3); | 1640 | 11.5k | ener = MULT16_32_Q14(exc_gain_quant_scal3[q_energy],ol_gain); | 1641 | 43.7k | } else if (SUBMODE(have_subframe_gain)==1) | 1642 | 22.0k | { | 1643 | 22.0k | q_energy = speex_bits_unpack_unsigned(bits, 1); | 1644 | 22.0k | ener = MULT16_32_Q14(exc_gain_quant_scal1[q_energy],ol_gain); | 1645 | 22.0k | } else { | 1646 | 21.7k | ener = ol_gain; | 1647 | 21.7k | } | 1648 | | | 1649 | 55.3k | speex_assert (SUBMODE(innovation_unquant)); | 1650 | 55.3k | { | 1651 | | /*Fixed codebook contribution*/ | 1652 | 55.3k | SUBMODE(innovation_unquant)(innov, SUBMODE(innovation_params), NB_SUBFRAME_SIZE, bits, stack, &st->seed); | 1653 | | /* De-normalize innovation and update excitation */ | 1654 | | | 1655 | 55.3k | signal_mul(innov, innov, ener, NB_SUBFRAME_SIZE); | 1656 | | | 1657 | | /* Decode second codebook (only for some modes) */ | 1658 | 55.3k | if (SUBMODE(double_codebook)) | 1659 | 3.40k | { | 1660 | 3.40k | char *tmp_stack=stack; | 1661 | 3.40k | VARDECL(spx_sig_t *innov2); | 1662 | 3.40k | ALLOC(innov2, NB_SUBFRAME_SIZE, spx_sig_t); | 1663 | 3.40k | SPEEX_MEMSET(innov2, 0, NB_SUBFRAME_SIZE); | 1664 | 3.40k | SUBMODE(innovation_unquant)(innov2, SUBMODE(innovation_params), NB_SUBFRAME_SIZE, bits, stack, &st->seed); | 1665 | 3.40k | signal_mul(innov2, innov2, MULT16_32_Q15(QCONST16(0.454545f,15),ener), NB_SUBFRAME_SIZE); | 1666 | 139k | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1667 | 136k | innov[i] = ADD32(innov[i], innov2[i]); | 1668 | 3.40k | stack = tmp_stack; | 1669 | 3.40k | } | 1670 | 2.26M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1671 | 2.21M | exc[i]=EXTRACT16(SATURATE32(PSHR32(ADD32(SHL32(exc32[i],1),innov[i]),SIG_SHIFT),32767)); | 1672 | | /*print_vec(exc, 40, "innov");*/ | 1673 | 55.3k | if (innov_save) | 1674 | 40.1k | { | 1675 | 1.64M | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1676 | 1.60M | innov_save[i] = EXTRACT16(PSHR32(innov[i], SIG_SHIFT)); | 1677 | 40.1k | } | 1678 | 55.3k | } | 1679 | | | 1680 | | /*Vocoder mode*/ | 1681 | 55.3k | if (st->submodeID==1) | 1682 | 9.26k | { | 1683 | 9.26k | spx_word16_t g=ol_pitch_coef; | 1684 | 9.26k | g=MULT16_16_P14(QCONST16(1.5f,14),(g-QCONST16(.2f,6))); | 1685 | 9.26k | if (g<0) | 1686 | 4.13k | g=0; | 1687 | 9.26k | if (g>GAIN_SCALING) | 1688 | 1.47k | g=GAIN_SCALING; | 1689 | | | 1690 | 9.26k | SPEEX_MEMSET(exc, 0, NB_SUBFRAME_SIZE); | 1691 | 19.7k | while (st->voc_offset<NB_SUBFRAME_SIZE) | 1692 | 10.4k | { | 1693 | | /* exc[st->voc_offset]= g*sqrt(2*ol_pitch)*ol_gain; | 1694 | | Not quite sure why we need the factor of two in the sqrt */ | 1695 | 10.4k | if (st->voc_offset>=0) | 1696 | 10.4k | exc[st->voc_offset]=MULT16_16(spx_sqrt(MULT16_16_16(2,ol_pitch)),EXTRACT16(PSHR32(MULT16_16(g,PSHR32(ol_gain,SIG_SHIFT)),6))); | 1697 | 10.4k | st->voc_offset+=ol_pitch; | 1698 | 10.4k | } | 1699 | 9.26k | st->voc_offset -= NB_SUBFRAME_SIZE; | 1700 | | | 1701 | 379k | for (i=0;i<NB_SUBFRAME_SIZE;i++) | 1702 | 370k | { | 1703 | 370k | spx_word16_t exci=exc[i]; | 1704 | 370k | exc[i]= ADD16(ADD16(MULT16_16_Q15(QCONST16(.7f,15),exc[i]) , MULT16_16_Q15(QCONST16(.3f,15),st->voc_m1)), | 1705 | 370k | SUB16(MULT16_16_Q15(Q15_ONE-MULT16_16_16(QCONST16(.85f,9),g),EXTRACT16(PSHR32(innov[i],SIG_SHIFT))), | 1706 | 370k | MULT16_16_Q15(MULT16_16_16(QCONST16(.15f,9),g),EXTRACT16(PSHR32(st->voc_m2,SIG_SHIFT))) | 1707 | 370k | )); | 1708 | 370k | st->voc_m1 = exci; | 1709 | 370k | st->voc_m2=innov[i]; | 1710 | 370k | st->voc_mean = EXTRACT16(PSHR32(ADD32(MULT16_16(QCONST16(.8f,15),st->voc_mean), MULT16_16(QCONST16(.2f,15),exc[i])), 15)); | 1711 | 370k | exc[i]-=st->voc_mean; | 1712 | 370k | } | 1713 | 9.26k | } | 1714 | | | 1715 | 55.3k | } | 1716 | 55.3k | } | 1717 | | | 1718 | 13.8k | ALLOC(interp_qlsp, NB_ORDER, spx_lsp_t); | 1719 | | | 1720 | 13.8k | if (st->lpc_enh_enabled && SUBMODE(comb_gain)>0 && !st->count_lost) | 1721 | 11.5k | { | 1722 | 11.5k | multicomb(st->exc-NB_SUBFRAME_SIZE, out, st->interp_qlpc, NB_ORDER, 2*NB_SUBFRAME_SIZE, best_pitch, 40, SUBMODE(comb_gain), stack); | 1723 | 11.5k | multicomb(st->exc+NB_SUBFRAME_SIZE, out+2*NB_SUBFRAME_SIZE, st->interp_qlpc, NB_ORDER, 2*NB_SUBFRAME_SIZE, best_pitch, 40, SUBMODE(comb_gain), stack); | 1724 | 11.5k | } else { | 1725 | 2.31k | SPEEX_COPY(out, &st->exc[-NB_SUBFRAME_SIZE], NB_FRAME_SIZE); | 1726 | 2.31k | } | 1727 | | | 1728 | | /* If the last packet was lost, re-scale the excitation to obtain the same energy as encoded in ol_gain */ | 1729 | 13.8k | if (st->count_lost) | 1730 | 0 | { | 1731 | 0 | spx_word16_t exc_ener; | 1732 | 0 | spx_word32_t gain32; | 1733 | 0 | spx_word16_t gain; | 1734 | 0 | exc_ener = compute_rms16 (st->exc, NB_FRAME_SIZE); | 1735 | 0 | gain32 = PDIV32(ol_gain, ADD16(exc_ener,1)); | 1736 | 0 | #ifdef FIXED_POINT | 1737 | 0 | if (gain32 > 32767) | 1738 | 0 | gain32 = 32767; | 1739 | 0 | gain = EXTRACT16(gain32); | 1740 | | #else | 1741 | | if (gain32 > 2) | 1742 | | gain32=2; | 1743 | | gain = gain32; | 1744 | | #endif | 1745 | 0 | for (i=0;i<NB_FRAME_SIZE;i++) | 1746 | 0 | { | 1747 | 0 | st->exc[i] = MULT16_16_Q14(gain, st->exc[i]); | 1748 | 0 | out[i]=st->exc[i-NB_SUBFRAME_SIZE]; | 1749 | 0 | } | 1750 | 0 | } | 1751 | | | 1752 | | /*Loop on subframes */ | 1753 | 69.1k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) | 1754 | 55.3k | { | 1755 | 55.3k | int offset; | 1756 | 55.3k | spx_word16_t *sp; | 1757 | | | 1758 | | /* Offset relative to start of frame */ | 1759 | 55.3k | offset = NB_SUBFRAME_SIZE*sub; | 1760 | | /* Original signal */ | 1761 | 55.3k | sp=out+offset; | 1762 | | | 1763 | | /* LSP interpolation (quantized and unquantized) */ | 1764 | 55.3k | lsp_interpolate(st->old_qlsp, qlsp, interp_qlsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); | 1765 | | | 1766 | | /* Compute interpolated LPCs (unquantized) */ | 1767 | 55.3k | lsp_to_lpc(interp_qlsp, ak, NB_ORDER, stack); | 1768 | | | 1769 | | /* Compute analysis filter at w=pi */ | 1770 | 55.3k | { | 1771 | 55.3k | spx_word32_t pi_g=LPC_SCALING; | 1772 | 331k | for (i=0;i<NB_ORDER;i+=2) | 1773 | 276k | { | 1774 | | /*pi_g += -st->interp_qlpc[i] + st->interp_qlpc[i+1];*/ | 1775 | 276k | pi_g = ADD32(pi_g, SUB32(EXTEND32(ak[i+1]),EXTEND32(ak[i]))); | 1776 | 276k | } | 1777 | 55.3k | st->pi_gain[sub] = pi_g; | 1778 | 55.3k | } | 1779 | | | 1780 | 55.3k | iir_mem16(sp, st->interp_qlpc, sp, NB_SUBFRAME_SIZE, NB_ORDER, | 1781 | 55.3k | st->mem_sp, stack); | 1782 | | | 1783 | 608k | for (i=0;i<NB_ORDER;i++) | 1784 | 553k | st->interp_qlpc[i] = ak[i]; | 1785 | | | 1786 | 55.3k | } | 1787 | | | 1788 | 13.8k | if (st->highpass_enabled) | 1789 | 13.8k | highpass(out, out, NB_FRAME_SIZE, (st->isWideband?HIGHPASS_WIDEBAND:HIGHPASS_NARROWBAND)|HIGHPASS_OUTPUT, st->mem_hp); | 1790 | | /*for (i=0;i<NB_FRAME_SIZE;i++) | 1791 | | printf ("%d\n", (int)st->frame[i]);*/ | 1792 | | | 1793 | | /* Tracking output level */ | 1794 | 13.8k | st->level = 1+PSHR32(ol_gain,SIG_SHIFT); | 1795 | 13.8k | st->max_level = MAX16(MULT16_16_Q15(QCONST16(.99f,15), st->max_level), st->level); | 1796 | 13.8k | st->min_level = MIN16(ADD16(1,MULT16_16_Q14(QCONST16(1.01f,14), st->min_level)), st->level); | 1797 | 13.8k | if (st->max_level < st->min_level+1) | 1798 | 4.57k | st->max_level = st->min_level+1; | 1799 | | /*printf ("%f %f %f %d\n", og, st->min_level, st->max_level, update);*/ | 1800 | | | 1801 | | /* Store the LSPs for interpolation in the next frame */ | 1802 | 152k | for (i=0;i<NB_ORDER;i++) | 1803 | 138k | st->old_qlsp[i] = qlsp[i]; | 1804 | | | 1805 | | /* The next frame will not be the first (Duh!) */ | 1806 | 13.8k | st->first = 0; | 1807 | 13.8k | st->count_lost=0; | 1808 | 13.8k | st->last_pitch = best_pitch; | 1809 | 13.8k | #ifdef FIXED_POINT | 1810 | 13.8k | st->last_pitch_gain = PSHR16(pitch_average,2); | 1811 | | #else | 1812 | | st->last_pitch_gain = .25*pitch_average; | 1813 | | #endif | 1814 | 13.8k | st->pitch_gain_buf[st->pitch_gain_buf_idx++] = st->last_pitch_gain; | 1815 | 13.8k | if (st->pitch_gain_buf_idx > 2) /* rollover */ | 1816 | 3.97k | st->pitch_gain_buf_idx = 0; | 1817 | | | 1818 | 13.8k | st->last_ol_gain = ol_gain; | 1819 | | | 1820 | 13.8k | return 0; | 1821 | 27.0k | } |
|
1822 | | #endif /* DISABLE_DECODER */ |
1823 | | |