/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 | 833k | #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 | | #define LSP_MARGIN 16 |
80 | | #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 | 84.6k | #define LSP_MARGIN .002f |
91 | 0 | #define LSP_DELTA1 .2f |
92 | | #define LSP_DELTA2 .05f |
93 | | |
94 | | #endif |
95 | | |
96 | | #ifdef VORBIS_PSYCHO |
97 | | #define EXTRA_BUFFER 100 |
98 | | #else |
99 | | #define EXTRA_BUFFER 0 |
100 | | #endif |
101 | | |
102 | | |
103 | | extern const spx_word16_t lag_window[]; |
104 | | extern const spx_word16_t lpc_window[]; |
105 | | |
106 | | #ifndef DISABLE_ENCODER |
107 | | void *nb_encoder_init(const SpeexMode *m) |
108 | 0 | { |
109 | 0 | EncState *st; |
110 | 0 | const SpeexNBMode *mode; |
111 | 0 | int i; |
112 | |
|
113 | 0 | mode=(const SpeexNBMode *)m->mode; |
114 | 0 | st = (EncState*)speex_alloc(sizeof(EncState)); |
115 | 0 | if (!st) |
116 | 0 | return NULL; |
117 | 0 | #if defined(VAR_ARRAYS) || defined (USE_ALLOCA) |
118 | 0 | st->stack = NULL; |
119 | | #else |
120 | | st->stack = (char*)speex_alloc_scratch(NB_ENC_STACK); |
121 | | #endif |
122 | |
|
123 | 0 | st->mode=m; |
124 | |
|
125 | 0 | st->gamma1=mode->gamma1; |
126 | 0 | st->gamma2=mode->gamma2; |
127 | 0 | st->lpc_floor = mode->lpc_floor; |
128 | |
|
129 | 0 | st->submodes=mode->submodes; |
130 | 0 | st->submodeID=st->submodeSelect=mode->defaultSubmode; |
131 | 0 | st->bounded_pitch = 1; |
132 | |
|
133 | 0 | st->encode_submode = 1; |
134 | |
|
135 | | #ifdef VORBIS_PSYCHO |
136 | | st->psy = vorbis_psy_init(8000, 256); |
137 | | st->curve = (float*)speex_alloc(128*sizeof(float)); |
138 | | st->old_curve = (float*)speex_alloc(128*sizeof(float)); |
139 | | st->psy_window = (float*)speex_alloc(256*sizeof(float)); |
140 | | #endif |
141 | |
|
142 | 0 | st->cumul_gain = 1024; |
143 | |
|
144 | 0 | st->window= lpc_window; |
145 | | |
146 | | /* Create the window for autocorrelation (lag-windowing) */ |
147 | 0 | st->lagWindow = lag_window; |
148 | |
|
149 | 0 | st->first = 1; |
150 | 0 | for (i=0;i<NB_ORDER;i++) |
151 | 0 | st->old_lsp[i]= DIV32(MULT16_16(QCONST16(3.1415927f, LSP_SHIFT), i+1), NB_ORDER+1); |
152 | |
|
153 | 0 | st->innov_rms_save = NULL; |
154 | |
|
155 | 0 | #ifndef DISABLE_VBR |
156 | 0 | vbr_init(&st->vbr); |
157 | 0 | st->vbr_quality = 8; |
158 | 0 | st->vbr_enabled = 0; |
159 | 0 | st->vbr_max = 0; |
160 | 0 | st->vad_enabled = 0; |
161 | 0 | st->dtx_enabled = 0; |
162 | 0 | st->dtx_count=0; |
163 | 0 | st->abr_enabled = 0; |
164 | 0 | st->abr_drift = 0; |
165 | 0 | st->abr_drift2 = 0; |
166 | 0 | #endif /* #ifndef DISABLE_VBR */ |
167 | |
|
168 | 0 | st->plc_tuning = 2; |
169 | 0 | st->complexity=2; |
170 | 0 | st->sampling_rate=8000; |
171 | 0 | st->isWideband = 0; |
172 | 0 | st->highpass_enabled = 1; |
173 | |
|
174 | | #ifdef ENABLE_VALGRIND |
175 | | VALGRIND_MAKE_MEM_DEFINED(st, NB_ENC_STACK); |
176 | | #endif |
177 | 0 | return st; |
178 | 0 | } |
179 | | |
180 | | void nb_encoder_destroy(void *state) |
181 | 0 | { |
182 | 0 | EncState *st=(EncState *)state; |
183 | | /* Free all allocated memory */ |
184 | | #if !(defined(VAR_ARRAYS) || defined (USE_ALLOCA)) |
185 | | speex_free_scratch(st->stack); |
186 | | #endif |
187 | |
|
188 | 0 | #ifndef DISABLE_VBR |
189 | 0 | vbr_destroy(&st->vbr); |
190 | 0 | #endif /* #ifndef DISABLE_VBR */ |
191 | |
|
192 | | #ifdef VORBIS_PSYCHO |
193 | | vorbis_psy_destroy(st->psy); |
194 | | speex_free (st->curve); |
195 | | speex_free (st->old_curve); |
196 | | speex_free (st->psy_window); |
197 | | #endif |
198 | | |
199 | | /*Free state memory... should be last*/ |
200 | 0 | speex_free(st); |
201 | 0 | } |
202 | | |
203 | | |
204 | | int nb_encoder_ctl(void *state, int request, void *ptr) |
205 | 0 | { |
206 | 0 | EncState *st; |
207 | 0 | st=(EncState*)state; |
208 | 0 | switch(request) |
209 | 0 | { |
210 | 0 | case SPEEX_GET_FRAME_SIZE: |
211 | 0 | (*(spx_int32_t*)ptr) = NB_FRAME_SIZE; |
212 | 0 | break; |
213 | 0 | case SPEEX_SET_LOW_MODE: |
214 | 0 | case SPEEX_SET_MODE: |
215 | 0 | st->submodeSelect = st->submodeID = (*(spx_int32_t*)ptr); |
216 | 0 | break; |
217 | 0 | case SPEEX_GET_LOW_MODE: |
218 | 0 | case SPEEX_GET_MODE: |
219 | 0 | (*(spx_int32_t*)ptr) = st->submodeID; |
220 | 0 | break; |
221 | 0 | #ifndef DISABLE_VBR |
222 | 0 | case SPEEX_SET_VBR: |
223 | 0 | st->vbr_enabled = (*(spx_int32_t*)ptr); |
224 | 0 | break; |
225 | 0 | case SPEEX_GET_VBR: |
226 | 0 | (*(spx_int32_t*)ptr) = st->vbr_enabled; |
227 | 0 | break; |
228 | 0 | case SPEEX_SET_VAD: |
229 | 0 | st->vad_enabled = (*(spx_int32_t*)ptr); |
230 | 0 | break; |
231 | 0 | case SPEEX_GET_VAD: |
232 | 0 | (*(spx_int32_t*)ptr) = st->vad_enabled; |
233 | 0 | break; |
234 | 0 | case SPEEX_SET_DTX: |
235 | 0 | st->dtx_enabled = (*(spx_int32_t*)ptr); |
236 | 0 | break; |
237 | 0 | case SPEEX_GET_DTX: |
238 | 0 | (*(spx_int32_t*)ptr) = st->dtx_enabled; |
239 | 0 | break; |
240 | 0 | case SPEEX_SET_ABR: |
241 | 0 | st->abr_enabled = (*(spx_int32_t*)ptr); |
242 | 0 | st->vbr_enabled = st->abr_enabled!=0; |
243 | 0 | if (st->vbr_enabled) |
244 | 0 | { |
245 | 0 | spx_int32_t i=10; |
246 | 0 | spx_int32_t rate, target; |
247 | 0 | float vbr_qual; |
248 | 0 | target = (*(spx_int32_t*)ptr); |
249 | 0 | while (i>=0) |
250 | 0 | { |
251 | 0 | speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i); |
252 | 0 | speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate); |
253 | 0 | if (rate <= target) |
254 | 0 | break; |
255 | 0 | i--; |
256 | 0 | } |
257 | 0 | vbr_qual=i; |
258 | 0 | if (vbr_qual<0) |
259 | 0 | vbr_qual=0; |
260 | 0 | speex_encoder_ctl(st, SPEEX_SET_VBR_QUALITY, &vbr_qual); |
261 | 0 | st->abr_count=0; |
262 | 0 | st->abr_drift=0; |
263 | 0 | st->abr_drift2=0; |
264 | 0 | } |
265 | |
|
266 | 0 | break; |
267 | 0 | case SPEEX_GET_ABR: |
268 | 0 | (*(spx_int32_t*)ptr) = st->abr_enabled; |
269 | 0 | break; |
270 | 0 | #endif /* #ifndef DISABLE_VBR */ |
271 | 0 | #if !defined(DISABLE_VBR) && !defined(DISABLE_FLOAT_API) |
272 | 0 | case SPEEX_SET_VBR_QUALITY: |
273 | 0 | st->vbr_quality = (*(float*)ptr); |
274 | 0 | break; |
275 | 0 | case SPEEX_GET_VBR_QUALITY: |
276 | 0 | (*(float*)ptr) = st->vbr_quality; |
277 | 0 | break; |
278 | 0 | #endif /* !defined(DISABLE_VBR) && !defined(DISABLE_FLOAT_API) */ |
279 | 0 | case SPEEX_SET_QUALITY: |
280 | 0 | { |
281 | 0 | int quality = (*(spx_int32_t*)ptr); |
282 | 0 | if (quality < 0) |
283 | 0 | quality = 0; |
284 | 0 | if (quality > 10) |
285 | 0 | quality = 10; |
286 | 0 | st->submodeSelect = st->submodeID = ((const SpeexNBMode*)(st->mode->mode))->quality_map[quality]; |
287 | 0 | } |
288 | 0 | break; |
289 | 0 | case SPEEX_SET_COMPLEXITY: |
290 | 0 | st->complexity = (*(spx_int32_t*)ptr); |
291 | 0 | if (st->complexity<0) |
292 | 0 | st->complexity=0; |
293 | 0 | break; |
294 | 0 | case SPEEX_GET_COMPLEXITY: |
295 | 0 | (*(spx_int32_t*)ptr) = st->complexity; |
296 | 0 | break; |
297 | 0 | case SPEEX_SET_BITRATE: |
298 | 0 | { |
299 | 0 | spx_int32_t i=10; |
300 | 0 | spx_int32_t rate, target; |
301 | 0 | target = (*(spx_int32_t*)ptr); |
302 | 0 | while (i>=0) |
303 | 0 | { |
304 | 0 | speex_encoder_ctl(st, SPEEX_SET_QUALITY, &i); |
305 | 0 | speex_encoder_ctl(st, SPEEX_GET_BITRATE, &rate); |
306 | 0 | if (rate <= target) |
307 | 0 | break; |
308 | 0 | i--; |
309 | 0 | } |
310 | 0 | } |
311 | 0 | break; |
312 | 0 | case SPEEX_GET_BITRATE: |
313 | 0 | if (st->submodes[st->submodeID]) |
314 | 0 | (*(spx_int32_t*)ptr) = st->sampling_rate*SUBMODE(bits_per_frame)/NB_FRAME_SIZE; |
315 | 0 | else |
316 | 0 | (*(spx_int32_t*)ptr) = st->sampling_rate*(NB_SUBMODE_BITS+1)/NB_FRAME_SIZE; |
317 | 0 | break; |
318 | 0 | case SPEEX_SET_SAMPLING_RATE: |
319 | 0 | st->sampling_rate = (*(spx_int32_t*)ptr); |
320 | 0 | break; |
321 | 0 | case SPEEX_GET_SAMPLING_RATE: |
322 | 0 | (*(spx_int32_t*)ptr)=st->sampling_rate; |
323 | 0 | break; |
324 | 0 | case SPEEX_RESET_STATE: |
325 | 0 | { |
326 | 0 | int i; |
327 | 0 | st->bounded_pitch = 1; |
328 | 0 | st->first = 1; |
329 | 0 | for (i=0;i<NB_ORDER;i++) |
330 | 0 | st->old_lsp[i]= DIV32(MULT16_16(QCONST16(3.1415927f, LSP_SHIFT), i+1), NB_ORDER+1); |
331 | 0 | for (i=0;i<NB_ORDER;i++) |
332 | 0 | st->mem_sw[i]=st->mem_sw_whole[i]=st->mem_sp[i]=st->mem_exc[i]=0; |
333 | 0 | for (i=0;i<NB_FRAME_SIZE+NB_PITCH_END+1;i++) |
334 | 0 | st->excBuf[i]=st->swBuf[i]=0; |
335 | 0 | for (i=0;i<NB_WINDOW_SIZE-NB_FRAME_SIZE;i++) |
336 | 0 | st->winBuf[i]=0; |
337 | 0 | } |
338 | 0 | break; |
339 | 0 | case SPEEX_SET_SUBMODE_ENCODING: |
340 | 0 | st->encode_submode = (*(spx_int32_t*)ptr); |
341 | 0 | break; |
342 | 0 | case SPEEX_GET_SUBMODE_ENCODING: |
343 | 0 | (*(spx_int32_t*)ptr) = st->encode_submode; |
344 | 0 | break; |
345 | 0 | case SPEEX_GET_LOOKAHEAD: |
346 | 0 | (*(spx_int32_t*)ptr)=(NB_WINDOW_SIZE-NB_FRAME_SIZE); |
347 | 0 | break; |
348 | 0 | case SPEEX_SET_PLC_TUNING: |
349 | 0 | st->plc_tuning = (*(spx_int32_t*)ptr); |
350 | 0 | if (st->plc_tuning>100) |
351 | 0 | st->plc_tuning=100; |
352 | 0 | break; |
353 | 0 | case SPEEX_GET_PLC_TUNING: |
354 | 0 | (*(spx_int32_t*)ptr)=(st->plc_tuning); |
355 | 0 | break; |
356 | 0 | #ifndef DISABLE_VBR |
357 | 0 | case SPEEX_SET_VBR_MAX_BITRATE: |
358 | 0 | st->vbr_max = (*(spx_int32_t*)ptr); |
359 | 0 | break; |
360 | 0 | case SPEEX_GET_VBR_MAX_BITRATE: |
361 | 0 | (*(spx_int32_t*)ptr) = st->vbr_max; |
362 | 0 | break; |
363 | 0 | #endif /* #ifndef DISABLE_VBR */ |
364 | 0 | case SPEEX_SET_HIGHPASS: |
365 | 0 | st->highpass_enabled = (*(spx_int32_t*)ptr); |
366 | 0 | break; |
367 | 0 | case SPEEX_GET_HIGHPASS: |
368 | 0 | (*(spx_int32_t*)ptr) = st->highpass_enabled; |
369 | 0 | break; |
370 | | |
371 | | /* This is all internal stuff past this point */ |
372 | 0 | case SPEEX_GET_PI_GAIN: |
373 | 0 | { |
374 | 0 | int i; |
375 | 0 | spx_word32_t *g = (spx_word32_t*)ptr; |
376 | 0 | for (i=0;i<NB_NB_SUBFRAMES;i++) |
377 | 0 | g[i]=st->pi_gain[i]; |
378 | 0 | } |
379 | 0 | break; |
380 | 0 | case SPEEX_GET_EXC: |
381 | 0 | { |
382 | 0 | int i; |
383 | 0 | for (i=0;i<NB_NB_SUBFRAMES;i++) |
384 | 0 | ((spx_word16_t*)ptr)[i] = compute_rms16(st->exc+i*NB_SUBFRAME_SIZE, NB_SUBFRAME_SIZE); |
385 | 0 | } |
386 | 0 | break; |
387 | 0 | #ifndef DISABLE_VBR |
388 | 0 | case SPEEX_GET_RELATIVE_QUALITY: |
389 | 0 | (*(float*)ptr)=st->relative_quality; |
390 | 0 | break; |
391 | 0 | #endif /* #ifndef DISABLE_VBR */ |
392 | 0 | case SPEEX_SET_INNOVATION_SAVE: |
393 | 0 | st->innov_rms_save = (spx_word16_t*)ptr; |
394 | 0 | break; |
395 | 0 | case SPEEX_SET_WIDEBAND: |
396 | 0 | st->isWideband = *((spx_int32_t*)ptr); |
397 | 0 | break; |
398 | 0 | case SPEEX_GET_STACK: |
399 | 0 | *((char**)ptr) = st->stack; |
400 | 0 | break; |
401 | 0 | default: |
402 | 0 | speex_warning_int("Unknown nb_ctl request: ", request); |
403 | 0 | return -1; |
404 | 0 | } |
405 | 0 | return 0; |
406 | 0 | } |
407 | | |
408 | | |
409 | | int nb_encode(void *state, void *vin, SpeexBits *bits) |
410 | 0 | { |
411 | 0 | EncState *st; |
412 | 0 | int i, sub, roots; |
413 | 0 | int ol_pitch; |
414 | 0 | spx_word16_t ol_pitch_coef; |
415 | 0 | spx_word32_t ol_gain; |
416 | 0 | VARDECL(spx_word16_t *target); |
417 | 0 | VARDECL(spx_sig_t *innov); |
418 | 0 | VARDECL(spx_word32_t *exc32); |
419 | 0 | VARDECL(spx_mem_t *mem); |
420 | 0 | VARDECL(spx_coef_t *bw_lpc1); |
421 | 0 | VARDECL(spx_coef_t *bw_lpc2); |
422 | 0 | VARDECL(spx_coef_t *lpc); |
423 | 0 | VARDECL(spx_lsp_t *lsp); |
424 | 0 | VARDECL(spx_lsp_t *qlsp); |
425 | 0 | VARDECL(spx_lsp_t *interp_lsp); |
426 | 0 | VARDECL(spx_lsp_t *interp_qlsp); |
427 | 0 | VARDECL(spx_coef_t *interp_lpc); |
428 | 0 | VARDECL(spx_coef_t *interp_qlpc); |
429 | 0 | char *stack; |
430 | 0 | VARDECL(spx_word16_t *syn_resp); |
431 | |
|
432 | 0 | spx_word32_t ener=0; |
433 | 0 | spx_word16_t fine_gain; |
434 | 0 | spx_word16_t *in = (spx_word16_t*)vin; |
435 | |
|
436 | 0 | st=(EncState *)state; |
437 | 0 | stack=st->stack; |
438 | |
|
439 | 0 | ALLOC(lpc, NB_ORDER, spx_coef_t); |
440 | 0 | ALLOC(bw_lpc1, NB_ORDER, spx_coef_t); |
441 | 0 | ALLOC(bw_lpc2, NB_ORDER, spx_coef_t); |
442 | 0 | ALLOC(lsp, NB_ORDER, spx_lsp_t); |
443 | 0 | ALLOC(qlsp, NB_ORDER, spx_lsp_t); |
444 | 0 | ALLOC(interp_lsp, NB_ORDER, spx_lsp_t); |
445 | 0 | ALLOC(interp_qlsp, NB_ORDER, spx_lsp_t); |
446 | 0 | ALLOC(interp_lpc, NB_ORDER, spx_coef_t); |
447 | 0 | ALLOC(interp_qlpc, NB_ORDER, spx_coef_t); |
448 | |
|
449 | 0 | st->exc = st->excBuf + NB_PITCH_END + 2; |
450 | 0 | st->sw = st->swBuf + NB_PITCH_END + 2; |
451 | | /* Move signals 1 frame towards the past */ |
452 | 0 | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, NB_PITCH_END+2); |
453 | 0 | SPEEX_MOVE(st->swBuf, st->swBuf+NB_FRAME_SIZE, NB_PITCH_END+2); |
454 | |
|
455 | 0 | if (st->highpass_enabled) |
456 | 0 | highpass(in, in, NB_FRAME_SIZE, (st->isWideband?HIGHPASS_WIDEBAND:HIGHPASS_NARROWBAND)|HIGHPASS_INPUT, st->mem_hp); |
457 | |
|
458 | 0 | { |
459 | 0 | VARDECL(spx_word16_t *w_sig); |
460 | 0 | VARDECL(spx_word16_t *autocorr); |
461 | 0 | ALLOC(w_sig, NB_WINDOW_SIZE, spx_word16_t); |
462 | 0 | ALLOC(autocorr, NB_ORDER+1, spx_word16_t); |
463 | | /* Window for analysis */ |
464 | 0 | for (i=0;i<NB_WINDOW_SIZE-NB_FRAME_SIZE;i++) |
465 | 0 | w_sig[i] = MULT16_16_Q15(st->winBuf[i],st->window[i]); |
466 | 0 | for (;i<NB_WINDOW_SIZE;i++) |
467 | 0 | w_sig[i] = MULT16_16_Q15(in[i-NB_WINDOW_SIZE+NB_FRAME_SIZE],st->window[i]); |
468 | | /* Compute auto-correlation */ |
469 | 0 | _spx_autocorr(w_sig, autocorr, NB_ORDER+1, NB_WINDOW_SIZE); |
470 | 0 | autocorr[0] = ADD16(autocorr[0],MULT16_16_Q15(autocorr[0],st->lpc_floor)); /* Noise floor in auto-correlation domain */ |
471 | | |
472 | | /* Lag windowing: equivalent to filtering in the power-spectrum domain */ |
473 | 0 | for (i=0;i<NB_ORDER+1;i++) |
474 | 0 | autocorr[i] = MULT16_16_Q15(autocorr[i],st->lagWindow[i]); |
475 | 0 | autocorr[0] = ADD16(autocorr[0],1); |
476 | | |
477 | | /* Levinson-Durbin */ |
478 | 0 | _spx_lpc(lpc, autocorr, NB_ORDER); |
479 | | /* LPC to LSPs (x-domain) transform */ |
480 | 0 | roots=lpc_to_lsp (lpc, NB_ORDER, lsp, 10, LSP_DELTA1, stack); |
481 | | /* Check if we found all the roots */ |
482 | 0 | if (roots!=NB_ORDER) |
483 | 0 | { |
484 | | /*If we can't find all LSP's, do some damage control and use previous filter*/ |
485 | 0 | for (i=0;i<NB_ORDER;i++) |
486 | 0 | { |
487 | 0 | lsp[i]=st->old_lsp[i]; |
488 | 0 | } |
489 | 0 | } |
490 | 0 | } |
491 | | |
492 | | |
493 | | |
494 | | |
495 | | /* Whole frame analysis (open-loop estimation of pitch and excitation gain) */ |
496 | 0 | { |
497 | 0 | int diff = NB_WINDOW_SIZE-NB_FRAME_SIZE; |
498 | 0 | if (st->first) |
499 | 0 | for (i=0;i<NB_ORDER;i++) |
500 | 0 | interp_lsp[i] = lsp[i]; |
501 | 0 | else |
502 | 0 | lsp_interpolate(st->old_lsp, lsp, interp_lsp, NB_ORDER, NB_NB_SUBFRAMES, NB_NB_SUBFRAMES<<1, LSP_MARGIN); |
503 | | |
504 | | /* Compute interpolated LPCs (unquantized) for whole frame*/ |
505 | 0 | lsp_to_lpc(interp_lsp, interp_lpc, NB_ORDER,stack); |
506 | | |
507 | | |
508 | | /*Open-loop pitch*/ |
509 | 0 | if (!st->submodes[st->submodeID] || (st->complexity>2 && SUBMODE(have_subframe_gain)<3) || SUBMODE(forced_pitch_gain) || SUBMODE(lbr_pitch) != -1 |
510 | 0 | #ifndef DISABLE_VBR |
511 | 0 | || st->vbr_enabled || st->vad_enabled |
512 | 0 | #endif |
513 | 0 | ) |
514 | 0 | { |
515 | 0 | int nol_pitch[6]; |
516 | 0 | spx_word16_t nol_pitch_coef[6]; |
517 | |
|
518 | 0 | bw_lpc(0.9, interp_lpc, bw_lpc1, NB_ORDER); |
519 | 0 | bw_lpc(0.55, interp_lpc, bw_lpc2, NB_ORDER); |
520 | |
|
521 | 0 | SPEEX_COPY(st->sw, st->winBuf, diff); |
522 | 0 | SPEEX_COPY(st->sw+diff, in, NB_FRAME_SIZE-diff); |
523 | 0 | filter10(st->sw, bw_lpc1, bw_lpc2, st->sw, NB_FRAME_SIZE, st->mem_sw_whole, stack); |
524 | |
|
525 | 0 | open_loop_nbest_pitch(st->sw, NB_PITCH_START, NB_PITCH_END, NB_FRAME_SIZE, |
526 | 0 | nol_pitch, nol_pitch_coef, 6, stack); |
527 | 0 | ol_pitch=nol_pitch[0]; |
528 | 0 | ol_pitch_coef = nol_pitch_coef[0]; |
529 | | /*Try to remove pitch multiples*/ |
530 | 0 | for (i=1;i<6;i++) |
531 | 0 | { |
532 | | #ifdef FIXED_POINT |
533 | | if ((nol_pitch_coef[i]>MULT16_16_Q15(nol_pitch_coef[0],27853)) && |
534 | | #else |
535 | 0 | if ((nol_pitch_coef[i]>.85*nol_pitch_coef[0]) && |
536 | 0 | #endif |
537 | 0 | (ABS(2*nol_pitch[i]-ol_pitch)<=2 || ABS(3*nol_pitch[i]-ol_pitch)<=3 || |
538 | 0 | ABS(4*nol_pitch[i]-ol_pitch)<=4 || ABS(5*nol_pitch[i]-ol_pitch)<=5)) |
539 | 0 | { |
540 | | /*ol_pitch_coef=nol_pitch_coef[i];*/ |
541 | 0 | ol_pitch = nol_pitch[i]; |
542 | 0 | } |
543 | 0 | } |
544 | | /*if (ol_pitch>50) |
545 | | ol_pitch/=2;*/ |
546 | | /*ol_pitch_coef = sqrt(ol_pitch_coef);*/ |
547 | |
|
548 | 0 | } else { |
549 | 0 | ol_pitch=0; |
550 | 0 | ol_pitch_coef=0; |
551 | 0 | } |
552 | | |
553 | | /*Compute "real" excitation*/ |
554 | | /*SPEEX_COPY(st->exc, st->winBuf, diff); |
555 | | SPEEX_COPY(st->exc+diff, in, NB_FRAME_SIZE-diff);*/ |
556 | 0 | fir_mem16(st->winBuf, interp_lpc, st->exc, diff, NB_ORDER, st->mem_exc, stack); |
557 | 0 | fir_mem16(in, interp_lpc, st->exc+diff, NB_FRAME_SIZE-diff, NB_ORDER, st->mem_exc, stack); |
558 | | |
559 | | /* Compute open-loop excitation gain */ |
560 | 0 | { |
561 | 0 | spx_word16_t g = compute_rms16(st->exc, NB_FRAME_SIZE); |
562 | 0 | if (st->submodeID!=1 && ol_pitch>0) |
563 | 0 | ol_gain = MULT16_16(g, MULT16_16_Q14(QCONST16(1.1,14), |
564 | 0 | spx_sqrt(QCONST32(1.,28)-MULT16_32_Q15(QCONST16(.8,15),SHL32(MULT16_16(ol_pitch_coef,ol_pitch_coef),16))))); |
565 | 0 | else |
566 | 0 | ol_gain = SHL32(EXTEND32(g),SIG_SHIFT); |
567 | 0 | } |
568 | 0 | } |
569 | |
|
570 | | #ifdef VORBIS_PSYCHO |
571 | | SPEEX_MOVE(st->psy_window, st->psy_window+NB_FRAME_SIZE, 256-NB_FRAME_SIZE); |
572 | | SPEEX_COPY(&st->psy_window[256-NB_FRAME_SIZE], in, NB_FRAME_SIZE); |
573 | | compute_curve(st->psy, st->psy_window, st->curve); |
574 | | /*print_vec(st->curve, 128, "curve");*/ |
575 | | if (st->first) |
576 | | SPEEX_COPY(st->old_curve, st->curve, 128); |
577 | | #endif |
578 | | |
579 | | /*VBR stuff*/ |
580 | 0 | #ifndef DISABLE_VBR |
581 | 0 | if (st->vbr_enabled||st->vad_enabled) |
582 | 0 | { |
583 | 0 | float lsp_dist=0; |
584 | 0 | for (i=0;i<NB_ORDER;i++) |
585 | 0 | lsp_dist += (st->old_lsp[i] - lsp[i])*(st->old_lsp[i] - lsp[i]); |
586 | 0 | lsp_dist /= LSP_SCALING*LSP_SCALING; |
587 | |
|
588 | 0 | if (st->abr_enabled) |
589 | 0 | { |
590 | 0 | float qual_change=0; |
591 | 0 | if (st->abr_drift2 * st->abr_drift > 0) |
592 | 0 | { |
593 | | /* Only adapt if long-term and short-term drift are the same sign */ |
594 | 0 | qual_change = -.00001*st->abr_drift/(1+st->abr_count); |
595 | 0 | if (qual_change>.05) |
596 | 0 | qual_change=.05; |
597 | 0 | if (qual_change<-.05) |
598 | 0 | qual_change=-.05; |
599 | 0 | } |
600 | 0 | st->vbr_quality += qual_change; |
601 | 0 | if (st->vbr_quality>10) |
602 | 0 | st->vbr_quality=10; |
603 | 0 | if (st->vbr_quality<0) |
604 | 0 | st->vbr_quality=0; |
605 | 0 | } |
606 | |
|
607 | 0 | st->relative_quality = vbr_analysis(&st->vbr, in, NB_FRAME_SIZE, ol_pitch, GAIN_SCALING_1*ol_pitch_coef); |
608 | | /*if (delta_qual<0)*/ |
609 | | /* delta_qual*=.1*(3+st->vbr_quality);*/ |
610 | 0 | if (st->vbr_enabled) |
611 | 0 | { |
612 | 0 | spx_int32_t mode; |
613 | 0 | int choice=0; |
614 | 0 | float min_diff=100; |
615 | 0 | mode = 8; |
616 | 0 | while (mode) |
617 | 0 | { |
618 | 0 | int v1; |
619 | 0 | float thresh; |
620 | 0 | v1=(int)floor(st->vbr_quality); |
621 | 0 | if (v1==10) |
622 | 0 | thresh = vbr_nb_thresh[mode][v1]; |
623 | 0 | else |
624 | 0 | thresh = (st->vbr_quality-v1)*vbr_nb_thresh[mode][v1+1] + (1+v1-st->vbr_quality)*vbr_nb_thresh[mode][v1]; |
625 | 0 | if (st->relative_quality > thresh && |
626 | 0 | st->relative_quality-thresh<min_diff) |
627 | 0 | { |
628 | 0 | choice = mode; |
629 | 0 | min_diff = st->relative_quality-thresh; |
630 | 0 | } |
631 | 0 | mode--; |
632 | 0 | } |
633 | 0 | mode=choice; |
634 | 0 | if (mode==0) |
635 | 0 | { |
636 | 0 | if (st->dtx_count==0 || lsp_dist>.05 || !st->dtx_enabled || st->dtx_count>20) |
637 | 0 | { |
638 | 0 | mode=1; |
639 | 0 | st->dtx_count=1; |
640 | 0 | } else { |
641 | 0 | mode=0; |
642 | 0 | st->dtx_count++; |
643 | 0 | } |
644 | 0 | } else { |
645 | 0 | st->dtx_count=0; |
646 | 0 | } |
647 | |
|
648 | 0 | speex_encoder_ctl(state, SPEEX_SET_MODE, &mode); |
649 | 0 | if (st->vbr_max>0) |
650 | 0 | { |
651 | 0 | spx_int32_t rate; |
652 | 0 | speex_encoder_ctl(state, SPEEX_GET_BITRATE, &rate); |
653 | 0 | if (rate > st->vbr_max) |
654 | 0 | { |
655 | 0 | rate = st->vbr_max; |
656 | 0 | speex_encoder_ctl(state, SPEEX_SET_BITRATE, &rate); |
657 | 0 | } |
658 | 0 | } |
659 | |
|
660 | 0 | if (st->abr_enabled) |
661 | 0 | { |
662 | 0 | spx_int32_t bitrate; |
663 | 0 | speex_encoder_ctl(state, SPEEX_GET_BITRATE, &bitrate); |
664 | 0 | st->abr_drift+=(bitrate-st->abr_enabled); |
665 | 0 | st->abr_drift2 = .95*st->abr_drift2 + .05*(bitrate-st->abr_enabled); |
666 | 0 | st->abr_count += 1.0; |
667 | 0 | } |
668 | |
|
669 | 0 | } else { |
670 | | /*VAD only case*/ |
671 | 0 | int mode; |
672 | 0 | if (st->relative_quality<2) |
673 | 0 | { |
674 | 0 | if (st->dtx_count==0 || lsp_dist>.05 || !st->dtx_enabled || st->dtx_count>20) |
675 | 0 | { |
676 | 0 | st->dtx_count=1; |
677 | 0 | mode=1; |
678 | 0 | } else { |
679 | 0 | mode=0; |
680 | 0 | st->dtx_count++; |
681 | 0 | } |
682 | 0 | } else { |
683 | 0 | st->dtx_count = 0; |
684 | 0 | mode=st->submodeSelect; |
685 | 0 | } |
686 | | /*speex_encoder_ctl(state, SPEEX_SET_MODE, &mode);*/ |
687 | 0 | st->submodeID=mode; |
688 | 0 | } |
689 | 0 | } else { |
690 | 0 | st->relative_quality = -1; |
691 | 0 | } |
692 | 0 | #endif /* #ifndef DISABLE_VBR */ |
693 | |
|
694 | 0 | if (st->encode_submode) |
695 | 0 | { |
696 | | /* First, transmit a zero for narrowband */ |
697 | 0 | speex_bits_pack(bits, 0, 1); |
698 | | |
699 | | /* Transmit the sub-mode we use for this frame */ |
700 | 0 | speex_bits_pack(bits, st->submodeID, NB_SUBMODE_BITS); |
701 | |
|
702 | 0 | } |
703 | | |
704 | | /* If null mode (no transmission), just set a couple things to zero*/ |
705 | 0 | if (st->submodes[st->submodeID] == NULL) |
706 | 0 | { |
707 | 0 | for (i=0;i<NB_FRAME_SIZE;i++) |
708 | 0 | st->exc[i]=st->sw[i]=VERY_SMALL; |
709 | |
|
710 | 0 | for (i=0;i<NB_ORDER;i++) |
711 | 0 | st->mem_sw[i]=0; |
712 | 0 | st->first=1; |
713 | 0 | st->bounded_pitch = 1; |
714 | |
|
715 | 0 | SPEEX_COPY(st->winBuf, in+2*NB_FRAME_SIZE-NB_WINDOW_SIZE, NB_WINDOW_SIZE-NB_FRAME_SIZE); |
716 | | |
717 | | /* Clear memory (no need to really compute it) */ |
718 | 0 | for (i=0;i<NB_ORDER;i++) |
719 | 0 | st->mem_sp[i] = 0; |
720 | 0 | return 0; |
721 | |
|
722 | 0 | } |
723 | | |
724 | | /* LSP Quantization */ |
725 | 0 | if (st->first) |
726 | 0 | { |
727 | 0 | for (i=0;i<NB_ORDER;i++) |
728 | 0 | st->old_lsp[i] = lsp[i]; |
729 | 0 | } |
730 | | |
731 | | |
732 | | /*Quantize LSPs*/ |
733 | 0 | #if 1 /*0 for unquantized*/ |
734 | 0 | SUBMODE(lsp_quant)(lsp, qlsp, NB_ORDER, bits); |
735 | | #else |
736 | | for (i=0;i<NB_ORDER;i++) |
737 | | qlsp[i]=lsp[i]; |
738 | | #endif |
739 | | |
740 | | /*If we use low bit-rate pitch mode, transmit open-loop pitch*/ |
741 | 0 | if (SUBMODE(lbr_pitch)!=-1) |
742 | 0 | { |
743 | 0 | speex_bits_pack(bits, ol_pitch-NB_PITCH_START, 7); |
744 | 0 | } |
745 | |
|
746 | 0 | if (SUBMODE(forced_pitch_gain)) |
747 | 0 | { |
748 | 0 | int quant; |
749 | | /* This just damps the pitch a bit, because it tends to be too aggressive when forced */ |
750 | 0 | ol_pitch_coef = MULT16_16_Q15(QCONST16(.9,15), ol_pitch_coef); |
751 | | #ifdef FIXED_POINT |
752 | | quant = PSHR16(MULT16_16_16(15, ol_pitch_coef),GAIN_SHIFT); |
753 | | #else |
754 | 0 | quant = (int)floor(.5+15*ol_pitch_coef*GAIN_SCALING_1); |
755 | 0 | #endif |
756 | 0 | if (quant>15) |
757 | 0 | quant=15; |
758 | 0 | if (quant<0) |
759 | 0 | quant=0; |
760 | 0 | speex_bits_pack(bits, quant, 4); |
761 | 0 | ol_pitch_coef=MULT16_16_P15(QCONST16(0.066667,15),SHL16(quant,GAIN_SHIFT)); |
762 | 0 | } |
763 | | |
764 | | |
765 | | /*Quantize and transmit open-loop excitation gain*/ |
766 | | #ifdef FIXED_POINT |
767 | | { |
768 | | int qe = scal_quant32(ol_gain, ol_gain_table, 32); |
769 | | /*ol_gain = exp(qe/3.5)*SIG_SCALING;*/ |
770 | | ol_gain = MULT16_32_Q15(28406,ol_gain_table[qe]); |
771 | | speex_bits_pack(bits, qe, 5); |
772 | | } |
773 | | #else |
774 | 0 | { |
775 | 0 | int qe = (int)(floor(.5+3.5*log(ol_gain*1.0/SIG_SCALING))); |
776 | 0 | if (qe<0) |
777 | 0 | qe=0; |
778 | 0 | if (qe>31) |
779 | 0 | qe=31; |
780 | 0 | ol_gain = exp(qe/3.5)*SIG_SCALING; |
781 | 0 | speex_bits_pack(bits, qe, 5); |
782 | 0 | } |
783 | 0 | #endif |
784 | | |
785 | | |
786 | | |
787 | | /* Special case for first frame */ |
788 | 0 | if (st->first) |
789 | 0 | { |
790 | 0 | for (i=0;i<NB_ORDER;i++) |
791 | 0 | st->old_qlsp[i] = qlsp[i]; |
792 | 0 | } |
793 | | |
794 | | /* Target signal */ |
795 | 0 | ALLOC(target, NB_SUBFRAME_SIZE, spx_word16_t); |
796 | 0 | ALLOC(innov, NB_SUBFRAME_SIZE, spx_sig_t); |
797 | 0 | ALLOC(exc32, NB_SUBFRAME_SIZE, spx_word32_t); |
798 | 0 | ALLOC(syn_resp, NB_SUBFRAME_SIZE, spx_word16_t); |
799 | 0 | ALLOC(mem, NB_ORDER, spx_mem_t); |
800 | | |
801 | | /* Loop on sub-frames */ |
802 | 0 | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) |
803 | 0 | { |
804 | 0 | int offset; |
805 | 0 | spx_word16_t *sw; |
806 | 0 | spx_word16_t *exc, *inBuf; |
807 | 0 | int pitch; |
808 | 0 | int response_bound = NB_SUBFRAME_SIZE; |
809 | | |
810 | | /* Offset relative to start of frame */ |
811 | 0 | offset = NB_SUBFRAME_SIZE*sub; |
812 | | /* Excitation */ |
813 | 0 | exc=st->exc+offset; |
814 | | /* Weighted signal */ |
815 | 0 | sw=st->sw+offset; |
816 | | |
817 | | /* LSP interpolation (quantized and unquantized) */ |
818 | 0 | lsp_interpolate(st->old_lsp, lsp, interp_lsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); |
819 | 0 | lsp_interpolate(st->old_qlsp, qlsp, interp_qlsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); |
820 | | |
821 | | /* Compute interpolated LPCs (quantized and unquantized) */ |
822 | 0 | lsp_to_lpc(interp_lsp, interp_lpc, NB_ORDER,stack); |
823 | |
|
824 | 0 | lsp_to_lpc(interp_qlsp, interp_qlpc, NB_ORDER, stack); |
825 | | |
826 | | /* Compute analysis filter gain at w=pi (for use in SB-CELP) */ |
827 | 0 | { |
828 | 0 | spx_word32_t pi_g=LPC_SCALING; |
829 | 0 | for (i=0;i<NB_ORDER;i+=2) |
830 | 0 | { |
831 | | /*pi_g += -st->interp_qlpc[i] + st->interp_qlpc[i+1];*/ |
832 | 0 | pi_g = ADD32(pi_g, SUB32(EXTEND32(interp_qlpc[i+1]),EXTEND32(interp_qlpc[i]))); |
833 | 0 | } |
834 | 0 | st->pi_gain[sub] = pi_g; |
835 | 0 | } |
836 | |
|
837 | | #ifdef VORBIS_PSYCHO |
838 | | { |
839 | | float curr_curve[128]; |
840 | | float fact = ((float)sub+1.0f)/NB_NB_SUBFRAMES; |
841 | | for (i=0;i<128;i++) |
842 | | curr_curve[i] = (1.0f-fact)*st->old_curve[i] + fact*st->curve[i]; |
843 | | curve_to_lpc(st->psy, curr_curve, bw_lpc1, bw_lpc2, 10); |
844 | | } |
845 | | #else |
846 | | /* Compute bandwidth-expanded (unquantized) LPCs for perceptual weighting */ |
847 | 0 | bw_lpc(st->gamma1, interp_lpc, bw_lpc1, NB_ORDER); |
848 | 0 | bw_lpc(st->gamma2, interp_lpc, bw_lpc2, NB_ORDER); |
849 | | /*print_vec(st->bw_lpc1, 10, "bw_lpc");*/ |
850 | 0 | #endif |
851 | | |
852 | | /*FIXME: This will break if we change the window size */ |
853 | 0 | speex_assert(NB_WINDOW_SIZE-NB_FRAME_SIZE == NB_SUBFRAME_SIZE); |
854 | 0 | if (sub==0) |
855 | 0 | inBuf = st->winBuf; |
856 | 0 | else |
857 | 0 | inBuf = &in[((sub-1)*NB_SUBFRAME_SIZE)]; |
858 | 0 | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
859 | 0 | sw[i] = inBuf[i]; |
860 | |
|
861 | 0 | if (st->complexity==0) |
862 | 0 | response_bound >>= 1; |
863 | 0 | compute_impulse_response(interp_qlpc, bw_lpc1, bw_lpc2, syn_resp, response_bound, NB_ORDER, stack); |
864 | 0 | for (i=response_bound;i<NB_SUBFRAME_SIZE;i++) |
865 | 0 | syn_resp[i]=VERY_SMALL; |
866 | | |
867 | | /* Compute zero response of A(z/g1) / ( A(z/g2) * A(z) ) */ |
868 | 0 | for (i=0;i<NB_ORDER;i++) |
869 | 0 | mem[i]=SHL32(st->mem_sp[i],1); |
870 | 0 | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
871 | 0 | exc[i] = VERY_SMALL; |
872 | | #ifdef SHORTCUTS2 |
873 | | iir_mem16(exc, interp_qlpc, exc, response_bound, NB_ORDER, mem, stack); |
874 | | for (i=0;i<NB_ORDER;i++) |
875 | | mem[i]=SHL32(st->mem_sw[i],1); |
876 | | filter10(exc, st->bw_lpc1, st->bw_lpc2, exc, response_bound, mem, stack); |
877 | | SPEEX_MEMSET(&exc[response_bound], 0, NB_SUBFRAME_SIZE-response_bound); |
878 | | #else |
879 | 0 | iir_mem16(exc, interp_qlpc, exc, NB_SUBFRAME_SIZE, NB_ORDER, mem, stack); |
880 | 0 | for (i=0;i<NB_ORDER;i++) |
881 | 0 | mem[i]=SHL32(st->mem_sw[i],1); |
882 | 0 | filter10(exc, bw_lpc1, bw_lpc2, exc, NB_SUBFRAME_SIZE, mem, stack); |
883 | 0 | #endif |
884 | | |
885 | | /* Compute weighted signal */ |
886 | 0 | for (i=0;i<NB_ORDER;i++) |
887 | 0 | mem[i]=st->mem_sw[i]; |
888 | 0 | filter10(sw, bw_lpc1, bw_lpc2, sw, NB_SUBFRAME_SIZE, mem, stack); |
889 | |
|
890 | 0 | if (st->complexity==0) |
891 | 0 | for (i=0;i<NB_ORDER;i++) |
892 | 0 | st->mem_sw[i]=mem[i]; |
893 | | |
894 | | /* Compute target signal (saturation prevents overflows on clipped input speech) */ |
895 | 0 | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
896 | 0 | target[i]=EXTRACT16(SATURATE(SUB32(sw[i],PSHR32(exc[i],1)),32767)); |
897 | |
|
898 | 0 | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
899 | 0 | exc[i] = inBuf[i]; |
900 | 0 | fir_mem16(exc, interp_qlpc, exc, NB_SUBFRAME_SIZE, NB_ORDER, st->mem_exc2, stack); |
901 | | /* If we have a long-term predictor (otherwise, something's wrong) */ |
902 | 0 | speex_assert (SUBMODE(ltp_quant)); |
903 | 0 | { |
904 | 0 | int pit_min, pit_max; |
905 | | /* Long-term prediction */ |
906 | 0 | if (SUBMODE(lbr_pitch) != -1) |
907 | 0 | { |
908 | | /* Low bit-rate pitch handling */ |
909 | 0 | int margin; |
910 | 0 | margin = SUBMODE(lbr_pitch); |
911 | 0 | if (margin) |
912 | 0 | { |
913 | 0 | if (ol_pitch < NB_PITCH_START+margin-1) |
914 | 0 | ol_pitch=NB_PITCH_START+margin-1; |
915 | 0 | if (ol_pitch > NB_PITCH_END-margin) |
916 | 0 | ol_pitch=NB_PITCH_END-margin; |
917 | 0 | pit_min = ol_pitch-margin+1; |
918 | 0 | pit_max = ol_pitch+margin; |
919 | 0 | } else { |
920 | 0 | pit_min=pit_max=ol_pitch; |
921 | 0 | } |
922 | 0 | } else { |
923 | 0 | pit_min = NB_PITCH_START; |
924 | 0 | pit_max = NB_PITCH_END; |
925 | 0 | } |
926 | | |
927 | | /* Force pitch to use only the current frame if needed */ |
928 | 0 | if (st->bounded_pitch && pit_max>offset) |
929 | 0 | pit_max=offset; |
930 | | |
931 | | /* Perform pitch search */ |
932 | 0 | pitch = SUBMODE(ltp_quant)(target, sw, interp_qlpc, bw_lpc1, bw_lpc2, |
933 | 0 | exc32, SUBMODE(ltp_params), pit_min, pit_max, ol_pitch_coef, |
934 | 0 | NB_ORDER, NB_SUBFRAME_SIZE, bits, stack, |
935 | 0 | exc, syn_resp, st->complexity, 0, st->plc_tuning, &st->cumul_gain); |
936 | |
|
937 | 0 | st->pitch[sub]=pitch; |
938 | 0 | } |
939 | | /* Quantization of innovation */ |
940 | 0 | SPEEX_MEMSET(innov, 0, NB_SUBFRAME_SIZE); |
941 | | |
942 | | /* FIXME: Make sure this is safe from overflows (so far so good) */ |
943 | 0 | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
944 | 0 | exc[i] = EXTRACT16(SUB32(EXTEND32(exc[i]), PSHR32(exc32[i],SIG_SHIFT-1))); |
945 | |
|
946 | 0 | ener = SHL32(EXTEND32(compute_rms16(exc, NB_SUBFRAME_SIZE)),SIG_SHIFT); |
947 | | |
948 | | /*FIXME: Should use DIV32_16 and make sure result fits in 16 bits */ |
949 | | #ifdef FIXED_POINT |
950 | | { |
951 | | spx_word32_t f = PDIV32(ener,PSHR32(ol_gain,SIG_SHIFT)); |
952 | | if (f<=32767) |
953 | | fine_gain = f; |
954 | | else |
955 | | fine_gain = 32767; |
956 | | } |
957 | | #else |
958 | 0 | fine_gain = PDIV32_16(ener,PSHR32(ol_gain,SIG_SHIFT)); |
959 | 0 | #endif |
960 | | /* Calculate gain correction for the sub-frame (if any) */ |
961 | 0 | if (SUBMODE(have_subframe_gain)) |
962 | 0 | { |
963 | 0 | int qe; |
964 | 0 | if (SUBMODE(have_subframe_gain)==3) |
965 | 0 | { |
966 | 0 | qe = scal_quant(fine_gain, exc_gain_quant_scal3_bound, 8); |
967 | 0 | speex_bits_pack(bits, qe, 3); |
968 | 0 | ener=MULT16_32_Q14(exc_gain_quant_scal3[qe],ol_gain); |
969 | 0 | } else { |
970 | 0 | qe = scal_quant(fine_gain, exc_gain_quant_scal1_bound, 2); |
971 | 0 | speex_bits_pack(bits, qe, 1); |
972 | 0 | ener=MULT16_32_Q14(exc_gain_quant_scal1[qe],ol_gain); |
973 | 0 | } |
974 | 0 | } else { |
975 | 0 | ener=ol_gain; |
976 | 0 | } |
977 | | |
978 | | /*printf ("%f %f\n", ener, ol_gain);*/ |
979 | | |
980 | | /* Normalize innovation */ |
981 | 0 | signal_div(target, target, ener, NB_SUBFRAME_SIZE); |
982 | | |
983 | | /* Quantize innovation */ |
984 | 0 | speex_assert (SUBMODE(innovation_quant)); |
985 | 0 | { |
986 | | /* Codebook search */ |
987 | 0 | SUBMODE(innovation_quant)(target, interp_qlpc, bw_lpc1, bw_lpc2, |
988 | 0 | SUBMODE(innovation_params), NB_ORDER, NB_SUBFRAME_SIZE, |
989 | 0 | innov, syn_resp, bits, stack, st->complexity, SUBMODE(double_codebook)); |
990 | | |
991 | | /* De-normalize innovation and update excitation */ |
992 | 0 | signal_mul(innov, innov, ener, NB_SUBFRAME_SIZE); |
993 | | |
994 | | /* In some (rare) modes, we do a second search (more bits) to reduce noise even more */ |
995 | 0 | if (SUBMODE(double_codebook)) { |
996 | 0 | char *tmp_stack=stack; |
997 | 0 | VARDECL(spx_sig_t *innov2); |
998 | 0 | ALLOC(innov2, NB_SUBFRAME_SIZE, spx_sig_t); |
999 | 0 | SPEEX_MEMSET(innov2, 0, NB_SUBFRAME_SIZE); |
1000 | 0 | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1001 | 0 | target[i]=MULT16_16_P13(QCONST16(2.2f,13), target[i]); |
1002 | 0 | SUBMODE(innovation_quant)(target, interp_qlpc, bw_lpc1, bw_lpc2, |
1003 | 0 | SUBMODE(innovation_params), NB_ORDER, NB_SUBFRAME_SIZE, |
1004 | 0 | innov2, syn_resp, bits, stack, st->complexity, 0); |
1005 | 0 | signal_mul(innov2, innov2, MULT16_32_Q15(QCONST16(0.454545f,15),ener), NB_SUBFRAME_SIZE); |
1006 | 0 | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1007 | 0 | innov[i] = ADD32(innov[i],innov2[i]); |
1008 | 0 | stack = tmp_stack; |
1009 | 0 | } |
1010 | 0 | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1011 | 0 | exc[i] = EXTRACT16(SATURATE32(PSHR32(ADD32(SHL32(exc32[i],1),innov[i]),SIG_SHIFT),32767)); |
1012 | 0 | if (st->innov_rms_save) |
1013 | 0 | st->innov_rms_save[sub] = compute_rms(innov, NB_SUBFRAME_SIZE); |
1014 | 0 | } |
1015 | | |
1016 | | /* Final signal synthesis from excitation */ |
1017 | 0 | iir_mem16(exc, interp_qlpc, sw, NB_SUBFRAME_SIZE, NB_ORDER, st->mem_sp, stack); |
1018 | | |
1019 | | /* Compute weighted signal again, from synthesized speech (not sure it's the right thing) */ |
1020 | 0 | if (st->complexity!=0) |
1021 | 0 | filter10(sw, bw_lpc1, bw_lpc2, sw, NB_SUBFRAME_SIZE, st->mem_sw, stack); |
1022 | |
|
1023 | 0 | } |
1024 | | |
1025 | | /* Store the LSPs for interpolation in the next frame */ |
1026 | 0 | if (st->submodeID>=1) |
1027 | 0 | { |
1028 | 0 | for (i=0;i<NB_ORDER;i++) |
1029 | 0 | st->old_lsp[i] = lsp[i]; |
1030 | 0 | for (i=0;i<NB_ORDER;i++) |
1031 | 0 | st->old_qlsp[i] = qlsp[i]; |
1032 | 0 | } |
1033 | |
|
1034 | | #ifdef VORBIS_PSYCHO |
1035 | | if (st->submodeID>=1) |
1036 | | SPEEX_COPY(st->old_curve, st->curve, 128); |
1037 | | #endif |
1038 | |
|
1039 | 0 | if (st->submodeID==1) |
1040 | 0 | { |
1041 | 0 | #ifndef DISABLE_VBR |
1042 | 0 | if (st->dtx_count) |
1043 | 0 | speex_bits_pack(bits, 15, 4); |
1044 | 0 | else |
1045 | 0 | #endif |
1046 | 0 | speex_bits_pack(bits, 0, 4); |
1047 | 0 | } |
1048 | | |
1049 | | /* The next frame will not be the first (Duh!) */ |
1050 | 0 | st->first = 0; |
1051 | 0 | SPEEX_COPY(st->winBuf, in+2*NB_FRAME_SIZE-NB_WINDOW_SIZE, NB_WINDOW_SIZE-NB_FRAME_SIZE); |
1052 | |
|
1053 | 0 | if (SUBMODE(innovation_quant) == noise_codebook_quant || st->submodeID==0) |
1054 | 0 | st->bounded_pitch = 1; |
1055 | 0 | else |
1056 | 0 | st->bounded_pitch = 0; |
1057 | |
|
1058 | 0 | return 1; |
1059 | 0 | } |
1060 | | #endif /* DISABLE_ENCODER */ |
1061 | | |
1062 | | |
1063 | | #ifndef DISABLE_DECODER |
1064 | | void *nb_decoder_init(const SpeexMode *m) |
1065 | 3.07k | { |
1066 | 3.07k | DecState *st; |
1067 | 3.07k | const SpeexNBMode *mode; |
1068 | 3.07k | int i; |
1069 | | |
1070 | 3.07k | mode=(const SpeexNBMode*)m->mode; |
1071 | 3.07k | st = (DecState *)speex_alloc(sizeof(DecState)); |
1072 | 3.07k | if (!st) |
1073 | 0 | return NULL; |
1074 | 3.07k | #if defined(VAR_ARRAYS) || defined (USE_ALLOCA) |
1075 | 3.07k | st->stack = NULL; |
1076 | | #else |
1077 | | st->stack = (char*)speex_alloc_scratch(NB_DEC_STACK); |
1078 | | #endif |
1079 | | |
1080 | 3.07k | st->mode=m; |
1081 | | |
1082 | | |
1083 | 3.07k | st->encode_submode = 1; |
1084 | | |
1085 | 3.07k | st->first=1; |
1086 | | /* Codec parameters, should eventually have several "modes"*/ |
1087 | | |
1088 | 3.07k | st->submodes=mode->submodes; |
1089 | 3.07k | st->submodeID=mode->defaultSubmode; |
1090 | | |
1091 | 3.07k | st->lpc_enh_enabled=1; |
1092 | | |
1093 | 3.07k | SPEEX_MEMSET(st->excBuf, 0, NB_FRAME_SIZE + NB_PITCH_END); |
1094 | | |
1095 | 3.07k | st->last_pitch = 40; |
1096 | 3.07k | st->count_lost=0; |
1097 | 3.07k | st->pitch_gain_buf[0] = st->pitch_gain_buf[1] = st->pitch_gain_buf[2] = 0; |
1098 | 3.07k | st->pitch_gain_buf_idx = 0; |
1099 | 3.07k | st->seed = 1000; |
1100 | | |
1101 | 3.07k | st->sampling_rate=8000; |
1102 | 3.07k | st->last_ol_gain = 0; |
1103 | | |
1104 | 3.07k | st->user_callback.func = &speex_default_user_handler; |
1105 | 3.07k | st->user_callback.data = NULL; |
1106 | 52.1k | for (i=0;i<16;i++) |
1107 | 49.1k | st->speex_callbacks[i].func = NULL; |
1108 | | |
1109 | 3.07k | st->voc_m1=st->voc_m2=st->voc_mean=0; |
1110 | 3.07k | st->voc_offset=0; |
1111 | 3.07k | st->dtx_enabled=0; |
1112 | 3.07k | st->isWideband = 0; |
1113 | 3.07k | st->highpass_enabled = 1; |
1114 | | |
1115 | | #ifdef ENABLE_VALGRIND |
1116 | | VALGRIND_MAKE_MEM_DEFINED(st, NB_DEC_STACK); |
1117 | | #endif |
1118 | 3.07k | return st; |
1119 | 3.07k | } |
1120 | | |
1121 | | void nb_decoder_destroy(void *state) |
1122 | 3.07k | { |
1123 | | #if !(defined(VAR_ARRAYS) || defined (USE_ALLOCA)) |
1124 | | DecState *st; |
1125 | | st=(DecState*)state; |
1126 | | |
1127 | | speex_free_scratch(st->stack); |
1128 | | #endif |
1129 | | |
1130 | 3.07k | speex_free(state); |
1131 | 3.07k | } |
1132 | | |
1133 | | int nb_decoder_ctl(void *state, int request, void *ptr) |
1134 | 134k | { |
1135 | 134k | DecState *st; |
1136 | 134k | st=(DecState*)state; |
1137 | 134k | switch(request) |
1138 | 134k | { |
1139 | 0 | case SPEEX_SET_LOW_MODE: |
1140 | 0 | case SPEEX_SET_MODE: |
1141 | 0 | st->submodeID = (*(spx_int32_t*)ptr); |
1142 | 0 | break; |
1143 | 0 | case SPEEX_GET_LOW_MODE: |
1144 | 0 | case SPEEX_GET_MODE: |
1145 | 0 | (*(spx_int32_t*)ptr) = st->submodeID; |
1146 | 0 | break; |
1147 | 3.07k | case SPEEX_SET_ENH: |
1148 | 3.07k | st->lpc_enh_enabled = *((spx_int32_t*)ptr); |
1149 | 3.07k | break; |
1150 | 0 | case SPEEX_GET_ENH: |
1151 | 0 | *((spx_int32_t*)ptr) = st->lpc_enh_enabled; |
1152 | 0 | break; |
1153 | 1.39k | case SPEEX_GET_FRAME_SIZE: |
1154 | 1.39k | (*(spx_int32_t*)ptr) = NB_FRAME_SIZE; |
1155 | 1.39k | break; |
1156 | 0 | case SPEEX_GET_BITRATE: |
1157 | 0 | if (st->submodes[st->submodeID]) |
1158 | 0 | (*(spx_int32_t*)ptr) = st->sampling_rate*SUBMODE(bits_per_frame)/NB_FRAME_SIZE; |
1159 | 0 | else |
1160 | 0 | (*(spx_int32_t*)ptr) = st->sampling_rate*(NB_SUBMODE_BITS+1)/NB_FRAME_SIZE; |
1161 | 0 | break; |
1162 | 3.07k | case SPEEX_SET_SAMPLING_RATE: |
1163 | 3.07k | st->sampling_rate = (*(spx_int32_t*)ptr); |
1164 | 3.07k | break; |
1165 | 1.67k | case SPEEX_GET_SAMPLING_RATE: |
1166 | 1.67k | (*(spx_int32_t*)ptr)=st->sampling_rate; |
1167 | 1.67k | break; |
1168 | 1.29k | case SPEEX_SET_HANDLER: |
1169 | 1.29k | { |
1170 | 1.29k | SpeexCallback *c = (SpeexCallback*)ptr; |
1171 | 1.29k | st->speex_callbacks[c->callback_id].func=c->func; |
1172 | 1.29k | st->speex_callbacks[c->callback_id].data=c->data; |
1173 | 1.29k | st->speex_callbacks[c->callback_id].callback_id=c->callback_id; |
1174 | 1.29k | } |
1175 | 1.29k | break; |
1176 | 0 | case SPEEX_SET_USER_HANDLER: |
1177 | 0 | { |
1178 | 0 | SpeexCallback *c = (SpeexCallback*)ptr; |
1179 | 0 | st->user_callback.func=c->func; |
1180 | 0 | st->user_callback.data=c->data; |
1181 | 0 | st->user_callback.callback_id=c->callback_id; |
1182 | 0 | } |
1183 | 0 | break; |
1184 | 0 | case SPEEX_RESET_STATE: |
1185 | 0 | { |
1186 | 0 | int i; |
1187 | 0 | for (i=0;i<NB_ORDER;i++) |
1188 | 0 | st->mem_sp[i]=0; |
1189 | 0 | for (i=0;i<NB_FRAME_SIZE + NB_PITCH_END + 1;i++) |
1190 | 0 | st->excBuf[i]=0; |
1191 | 0 | } |
1192 | 0 | break; |
1193 | 0 | case SPEEX_SET_SUBMODE_ENCODING: |
1194 | 0 | st->encode_submode = (*(spx_int32_t*)ptr); |
1195 | 0 | break; |
1196 | 0 | case SPEEX_GET_SUBMODE_ENCODING: |
1197 | 0 | (*(spx_int32_t*)ptr) = st->encode_submode; |
1198 | 0 | break; |
1199 | 2.91k | case SPEEX_GET_LOOKAHEAD: |
1200 | 2.91k | (*(spx_int32_t*)ptr)=NB_SUBFRAME_SIZE; |
1201 | 2.91k | break; |
1202 | 0 | case SPEEX_SET_HIGHPASS: |
1203 | 0 | st->highpass_enabled = (*(spx_int32_t*)ptr); |
1204 | 0 | break; |
1205 | 0 | case SPEEX_GET_HIGHPASS: |
1206 | 0 | (*(spx_int32_t*)ptr) = st->highpass_enabled; |
1207 | 0 | break; |
1208 | | /* FIXME: Convert to fixed-point and re-enable even when float API is disabled */ |
1209 | 0 | #ifndef DISABLE_FLOAT_API |
1210 | 0 | case SPEEX_GET_ACTIVITY: |
1211 | 0 | { |
1212 | 0 | float ret; |
1213 | 0 | ret = log(st->level/st->min_level)/log(st->max_level/st->min_level); |
1214 | 0 | if (ret>1) |
1215 | 0 | ret = 1; |
1216 | | /* Done in a strange way to catch NaNs as well */ |
1217 | 0 | if (!(ret > 0)) |
1218 | 0 | ret = 0; |
1219 | | /*printf ("%f %f %f %f\n", st->level, st->min_level, st->max_level, ret);*/ |
1220 | 0 | (*(spx_int32_t*)ptr) = (int)(100*ret); |
1221 | 0 | } |
1222 | 0 | break; |
1223 | 0 | #endif |
1224 | 3.56k | case SPEEX_GET_PI_GAIN: |
1225 | 3.56k | { |
1226 | 3.56k | int i; |
1227 | 3.56k | spx_word32_t *g = (spx_word32_t*)ptr; |
1228 | 17.8k | for (i=0;i<NB_NB_SUBFRAMES;i++) |
1229 | 14.2k | g[i]=st->pi_gain[i]; |
1230 | 3.56k | } |
1231 | 3.56k | break; |
1232 | 3.56k | case SPEEX_GET_EXC: |
1233 | 3.56k | { |
1234 | 3.56k | int i; |
1235 | 17.8k | for (i=0;i<NB_NB_SUBFRAMES;i++) |
1236 | 14.2k | ((spx_word16_t*)ptr)[i] = compute_rms16(st->exc+i*NB_SUBFRAME_SIZE, NB_SUBFRAME_SIZE); |
1237 | 3.56k | } |
1238 | 3.56k | break; |
1239 | 68.6k | case SPEEX_GET_DTX_STATUS: |
1240 | 68.6k | *((spx_int32_t*)ptr) = st->dtx_enabled; |
1241 | 68.6k | break; |
1242 | 42.4k | case SPEEX_SET_INNOVATION_SAVE: |
1243 | 42.4k | st->innov_save = (spx_word16_t*)ptr; |
1244 | 42.4k | break; |
1245 | 2.65k | case SPEEX_SET_WIDEBAND: |
1246 | 2.65k | st->isWideband = *((spx_int32_t*)ptr); |
1247 | 2.65k | break; |
1248 | 0 | case SPEEX_GET_STACK: |
1249 | 0 | *((char**)ptr) = st->stack; |
1250 | 0 | break; |
1251 | 0 | default: |
1252 | 0 | speex_warning_int("Unknown nb_ctl request: ", request); |
1253 | 0 | return -1; |
1254 | 134k | } |
1255 | 134k | return 0; |
1256 | 134k | } |
1257 | | |
1258 | | |
1259 | 0 | #define median3(a, b, c) ((a) < (b) ? ((b) < (c) ? (b) : ((a) < (c) ? (c) : (a))) : ((c) < (b) ? (b) : ((c) < (a) ? (c) : (a)))) |
1260 | | |
1261 | | #ifdef FIXED_POINT |
1262 | | const spx_word16_t attenuation[10] = {32767, 31483, 27923, 22861, 17278, 12055, 7764, 4616, 2533, 1283}; |
1263 | | #else |
1264 | | 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}; |
1265 | | |
1266 | | #endif |
1267 | | |
1268 | | static void nb_decode_lost(DecState *st, spx_word16_t *out, char *stack) |
1269 | 0 | { |
1270 | 0 | int i; |
1271 | 0 | int pitch_val; |
1272 | 0 | spx_word16_t pitch_gain; |
1273 | 0 | spx_word16_t fact; |
1274 | 0 | spx_word16_t gain_med; |
1275 | 0 | spx_word16_t innov_gain; |
1276 | 0 | spx_word16_t noise_gain; |
1277 | |
|
1278 | 0 | st->exc = st->excBuf + 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 6; |
1279 | |
|
1280 | 0 | if (st->count_lost<10) |
1281 | 0 | fact = attenuation[st->count_lost]; |
1282 | 0 | else |
1283 | 0 | fact = 0; |
1284 | |
|
1285 | 0 | gain_med = median3(st->pitch_gain_buf[0], st->pitch_gain_buf[1], st->pitch_gain_buf[2]); |
1286 | 0 | if (gain_med < st->last_pitch_gain) |
1287 | 0 | st->last_pitch_gain = gain_med; |
1288 | |
|
1289 | | #ifdef FIXED_POINT |
1290 | | pitch_gain = st->last_pitch_gain; |
1291 | | if (pitch_gain>54) |
1292 | | pitch_gain = 54; |
1293 | | pitch_gain = SHL16(pitch_gain, 9); |
1294 | | #else |
1295 | 0 | pitch_gain = GAIN_SCALING_1*st->last_pitch_gain; |
1296 | 0 | if (pitch_gain>.85) |
1297 | 0 | pitch_gain=.85; |
1298 | 0 | #endif |
1299 | 0 | pitch_gain = MULT16_16_Q15(fact,pitch_gain) + VERY_SMALL; |
1300 | | /* FIXME: This was rms of innovation (not exc) */ |
1301 | 0 | innov_gain = compute_rms16(st->exc, NB_FRAME_SIZE); |
1302 | 0 | noise_gain = MULT16_16_Q15(innov_gain, MULT16_16_Q15(fact, SUB16(Q15ONE,MULT16_16_Q15(pitch_gain,pitch_gain)))); |
1303 | | /* Shift all buffers by one frame */ |
1304 | 0 | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 12); |
1305 | | |
1306 | |
|
1307 | 0 | pitch_val = st->last_pitch + SHR32((spx_int32_t)speex_rand(1+st->count_lost, &st->seed),SIG_SHIFT); |
1308 | 0 | if (pitch_val > NB_PITCH_END) |
1309 | 0 | pitch_val = NB_PITCH_END; |
1310 | 0 | if (pitch_val < NB_PITCH_START) |
1311 | 0 | pitch_val = NB_PITCH_START; |
1312 | 0 | for (i=0;i<NB_FRAME_SIZE;i++) |
1313 | 0 | { |
1314 | 0 | st->exc[i]= MULT16_16_Q15(pitch_gain, (st->exc[i-pitch_val]+VERY_SMALL)) + |
1315 | 0 | speex_rand(noise_gain, &st->seed); |
1316 | 0 | } |
1317 | |
|
1318 | 0 | bw_lpc(QCONST16(.98,15), st->interp_qlpc, st->interp_qlpc, NB_ORDER); |
1319 | 0 | iir_mem16(&st->exc[-NB_SUBFRAME_SIZE], st->interp_qlpc, out, NB_FRAME_SIZE, |
1320 | 0 | NB_ORDER, st->mem_sp, stack); |
1321 | 0 | highpass(out, out, NB_FRAME_SIZE, HIGHPASS_NARROWBAND|HIGHPASS_OUTPUT, st->mem_hp); |
1322 | |
|
1323 | 0 | st->first = 0; |
1324 | 0 | st->count_lost++; |
1325 | 0 | st->pitch_gain_buf[st->pitch_gain_buf_idx++] = PSHR16(pitch_gain,9); |
1326 | 0 | if (st->pitch_gain_buf_idx > 2) /* rollover */ |
1327 | 0 | st->pitch_gain_buf_idx = 0; |
1328 | 0 | } |
1329 | | |
1330 | | /* Just so we don't need to carry the complete wideband mode information */ |
1331 | | static const int wb_skip_table[8] = {0, 36, 112, 192, 352, 0, 0, 0}; |
1332 | | |
1333 | | int nb_decode(void *state, SpeexBits *bits, void *vout) |
1334 | 58.2k | { |
1335 | 58.2k | DecState *st; |
1336 | 58.2k | int i, sub; |
1337 | 58.2k | int pitch; |
1338 | 58.2k | spx_word16_t pitch_gain[3]; |
1339 | 58.2k | spx_word32_t ol_gain=0; |
1340 | 58.2k | int ol_pitch=0; |
1341 | 58.2k | spx_word16_t ol_pitch_coef=0; |
1342 | 58.2k | int best_pitch=40; |
1343 | 58.2k | spx_word16_t best_pitch_gain=0; |
1344 | 58.2k | int wideband; |
1345 | 58.2k | int m; |
1346 | 58.2k | char *stack; |
1347 | 58.2k | VARDECL(spx_sig_t *innov); |
1348 | 58.2k | VARDECL(spx_word32_t *exc32); |
1349 | 58.2k | VARDECL(spx_coef_t *ak); |
1350 | 58.2k | VARDECL(spx_lsp_t *qlsp); |
1351 | 58.2k | spx_word16_t pitch_average=0; |
1352 | | |
1353 | 58.2k | spx_word16_t *out = (spx_word16_t*)vout; |
1354 | 58.2k | VARDECL(spx_lsp_t *interp_qlsp); |
1355 | | |
1356 | 58.2k | st=(DecState*)state; |
1357 | 58.2k | stack=st->stack; |
1358 | | |
1359 | 58.2k | st->exc = st->excBuf + 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 6; |
1360 | | |
1361 | | /* Check if we're in DTX mode*/ |
1362 | 58.2k | if (!bits && st->dtx_enabled) |
1363 | 0 | { |
1364 | 0 | st->submodeID=0; |
1365 | 0 | } else |
1366 | 58.2k | { |
1367 | | /* If bits is NULL, consider the packet to be lost (what could we do anyway) */ |
1368 | 58.2k | if (!bits) |
1369 | 0 | { |
1370 | 0 | nb_decode_lost(st, out, stack); |
1371 | 0 | return 0; |
1372 | 0 | } |
1373 | | |
1374 | 58.2k | if (st->encode_submode) |
1375 | 58.2k | { |
1376 | | |
1377 | | /* Search for next narrowband block (handle requests, skip wideband blocks) */ |
1378 | 63.1k | do { |
1379 | 63.1k | if (speex_bits_remaining(bits)<5) |
1380 | 11.3k | return -1; |
1381 | 51.8k | wideband = speex_bits_unpack_unsigned(bits, 1); |
1382 | 51.8k | if (wideband) /* Skip wideband block (for compatibility) */ |
1383 | 3.53k | { |
1384 | 3.53k | int submode; |
1385 | 3.53k | int advance; |
1386 | 3.53k | advance = submode = speex_bits_unpack_unsigned(bits, SB_SUBMODE_BITS); |
1387 | | /*speex_mode_query(&speex_wb_mode, SPEEX_SUBMODE_BITS_PER_FRAME, &advance);*/ |
1388 | 3.53k | advance = wb_skip_table[submode]; |
1389 | 3.53k | if (advance < 0) |
1390 | 0 | { |
1391 | 0 | speex_notify("Invalid mode encountered. The stream is corrupted."); |
1392 | 0 | return -2; |
1393 | 0 | } |
1394 | 3.53k | advance -= (SB_SUBMODE_BITS+1); |
1395 | 3.53k | speex_bits_advance(bits, advance); |
1396 | | |
1397 | 3.53k | if (speex_bits_remaining(bits)<5) |
1398 | 831 | return -1; |
1399 | 2.70k | wideband = speex_bits_unpack_unsigned(bits, 1); |
1400 | 2.70k | if (wideband) |
1401 | 2.02k | { |
1402 | 2.02k | advance = submode = speex_bits_unpack_unsigned(bits, SB_SUBMODE_BITS); |
1403 | | /*speex_mode_query(&speex_wb_mode, SPEEX_SUBMODE_BITS_PER_FRAME, &advance);*/ |
1404 | 2.02k | advance = wb_skip_table[submode]; |
1405 | 2.02k | if (advance < 0) |
1406 | 0 | { |
1407 | 0 | speex_notify("Invalid mode encountered. The stream is corrupted."); |
1408 | 0 | return -2; |
1409 | 0 | } |
1410 | 2.02k | advance -= (SB_SUBMODE_BITS+1); |
1411 | 2.02k | speex_bits_advance(bits, advance); |
1412 | 2.02k | wideband = speex_bits_unpack_unsigned(bits, 1); |
1413 | 2.02k | if (wideband) |
1414 | 1.53k | { |
1415 | 1.53k | speex_notify("More than two wideband layers found. The stream is corrupted."); |
1416 | 1.53k | return -2; |
1417 | 1.53k | } |
1418 | | |
1419 | 2.02k | } |
1420 | 2.70k | } |
1421 | 49.4k | if (speex_bits_remaining(bits)<4) |
1422 | 463 | return -1; |
1423 | | /* FIXME: Check for overflow */ |
1424 | 48.9k | m = speex_bits_unpack_unsigned(bits, 4); |
1425 | 48.9k | if (m==15) /* We found a terminator */ |
1426 | 598 | { |
1427 | 598 | return -1; |
1428 | 48.3k | } else if (m==14) /* Speex in-band request */ |
1429 | 4.08k | { |
1430 | 4.08k | int ret = speex_inband_handler(bits, st->speex_callbacks, state); |
1431 | 4.08k | if (ret) |
1432 | 0 | return ret; |
1433 | 44.2k | } else if (m==13) /* User in-band request */ |
1434 | 819 | { |
1435 | 819 | int ret = st->user_callback.func(bits, state, st->user_callback.data); |
1436 | 819 | if (ret) |
1437 | 0 | return ret; |
1438 | 43.4k | } else if (m>8) /* Invalid mode */ |
1439 | 3.29k | { |
1440 | 3.29k | speex_notify("Invalid mode encountered. The stream is corrupted."); |
1441 | 3.29k | return -2; |
1442 | 3.29k | } |
1443 | | |
1444 | 48.9k | } while (m>8); |
1445 | | |
1446 | | /* Get the sub-mode that was used */ |
1447 | 40.1k | st->submodeID = m; |
1448 | 40.1k | } |
1449 | | |
1450 | 58.2k | } |
1451 | | |
1452 | | /* Shift all buffers by one frame */ |
1453 | 40.1k | SPEEX_MOVE(st->excBuf, st->excBuf+NB_FRAME_SIZE, 2*NB_PITCH_END + NB_SUBFRAME_SIZE + 12); |
1454 | | |
1455 | | /* If null mode (no transmission), just set a couple things to zero*/ |
1456 | 40.1k | if (st->submodes[st->submodeID] == NULL) |
1457 | 19.0k | { |
1458 | 19.0k | VARDECL(spx_coef_t *lpc); |
1459 | 19.0k | ALLOC(lpc, NB_ORDER, spx_coef_t); |
1460 | 19.0k | bw_lpc(QCONST16(0.93f,15), st->interp_qlpc, lpc, NB_ORDER); |
1461 | 19.0k | { |
1462 | 19.0k | spx_word16_t innov_gain=0; |
1463 | | /* FIXME: This was innov, not exc */ |
1464 | 19.0k | innov_gain = compute_rms16(st->exc, NB_FRAME_SIZE); |
1465 | 3.06M | for (i=0;i<NB_FRAME_SIZE;i++) |
1466 | 3.04M | st->exc[i]=speex_rand(innov_gain, &st->seed); |
1467 | 19.0k | } |
1468 | | |
1469 | | |
1470 | 19.0k | st->first=1; |
1471 | | |
1472 | | /* Final signal synthesis from excitation */ |
1473 | 19.0k | iir_mem16(st->exc, lpc, out, NB_FRAME_SIZE, NB_ORDER, st->mem_sp, stack); |
1474 | | |
1475 | 19.0k | st->count_lost=0; |
1476 | 19.0k | return 0; |
1477 | 19.0k | } |
1478 | | |
1479 | 21.1k | ALLOC(qlsp, NB_ORDER, spx_lsp_t); |
1480 | | |
1481 | | /* Unquantize LSPs */ |
1482 | 21.1k | SUBMODE(lsp_unquant)(qlsp, NB_ORDER, bits); |
1483 | | |
1484 | | /*Damp memory if a frame was lost and the LSP changed too much*/ |
1485 | 21.1k | if (st->count_lost) |
1486 | 0 | { |
1487 | 0 | spx_word16_t fact; |
1488 | 0 | spx_word32_t lsp_dist=0; |
1489 | 0 | for (i=0;i<NB_ORDER;i++) |
1490 | 0 | lsp_dist = ADD32(lsp_dist, EXTEND32(ABS(st->old_qlsp[i] - qlsp[i]))); |
1491 | | #ifdef FIXED_POINT |
1492 | | fact = SHR16(19661,SHR32(lsp_dist,LSP_SHIFT+2)); |
1493 | | #else |
1494 | 0 | fact = .6*exp(-.2*lsp_dist); |
1495 | 0 | #endif |
1496 | 0 | for (i=0;i<NB_ORDER;i++) |
1497 | 0 | st->mem_sp[i] = MULT16_32_Q15(fact,st->mem_sp[i]); |
1498 | 0 | } |
1499 | | |
1500 | | |
1501 | | /* Handle first frame and lost-packet case */ |
1502 | 21.1k | if (st->first || st->count_lost) |
1503 | 3.85k | { |
1504 | 42.4k | for (i=0;i<NB_ORDER;i++) |
1505 | 38.5k | st->old_qlsp[i] = qlsp[i]; |
1506 | 3.85k | } |
1507 | | |
1508 | | /* Get open-loop pitch estimation for low bit-rate pitch coding */ |
1509 | 21.1k | if (SUBMODE(lbr_pitch)!=-1) |
1510 | 8.15k | { |
1511 | 8.15k | ol_pitch = NB_PITCH_START+speex_bits_unpack_unsigned(bits, 7); |
1512 | 8.15k | } |
1513 | | |
1514 | 21.1k | if (SUBMODE(forced_pitch_gain)) |
1515 | 6.31k | { |
1516 | 6.31k | int quant; |
1517 | 6.31k | quant = speex_bits_unpack_unsigned(bits, 4); |
1518 | 6.31k | ol_pitch_coef=MULT16_16_P15(QCONST16(0.066667,15),SHL16(quant,GAIN_SHIFT)); |
1519 | 6.31k | } |
1520 | | |
1521 | | /* Get global excitation gain */ |
1522 | 21.1k | { |
1523 | 21.1k | int qe; |
1524 | 21.1k | qe = speex_bits_unpack_unsigned(bits, 5); |
1525 | | #ifdef FIXED_POINT |
1526 | | /* FIXME: Perhaps we could slightly lower the gain here when the output is going to saturate? */ |
1527 | | ol_gain = MULT16_32_Q15(28406,ol_gain_table[qe]); |
1528 | | #else |
1529 | 21.1k | ol_gain = SIG_SCALING*exp(qe/3.5); |
1530 | 21.1k | #endif |
1531 | 21.1k | } |
1532 | | |
1533 | 21.1k | ALLOC(ak, NB_ORDER, spx_coef_t); |
1534 | 21.1k | ALLOC(innov, NB_SUBFRAME_SIZE, spx_sig_t); |
1535 | 21.1k | ALLOC(exc32, NB_SUBFRAME_SIZE, spx_word32_t); |
1536 | | |
1537 | 21.1k | if (st->submodeID==1) |
1538 | 3.71k | { |
1539 | 3.71k | int extra; |
1540 | 3.71k | extra = speex_bits_unpack_unsigned(bits, 4); |
1541 | | |
1542 | 3.71k | if (extra==15) |
1543 | 650 | st->dtx_enabled=1; |
1544 | 3.06k | else |
1545 | 3.06k | st->dtx_enabled=0; |
1546 | 3.71k | } |
1547 | 21.1k | if (st->submodeID>1) |
1548 | 17.4k | st->dtx_enabled=0; |
1549 | | |
1550 | | /*Loop on subframes */ |
1551 | 105k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) |
1552 | 84.6k | { |
1553 | 84.6k | int offset; |
1554 | 84.6k | spx_word16_t *exc; |
1555 | 84.6k | spx_word16_t *innov_save = NULL; |
1556 | 84.6k | spx_word16_t tmp; |
1557 | | |
1558 | | /* Offset relative to start of frame */ |
1559 | 84.6k | offset = NB_SUBFRAME_SIZE*sub; |
1560 | | /* Excitation */ |
1561 | 84.6k | exc=st->exc+offset; |
1562 | | /* Original signal */ |
1563 | 84.6k | if (st->innov_save) |
1564 | 50.9k | innov_save = st->innov_save+offset; |
1565 | | |
1566 | | |
1567 | | /* Reset excitation */ |
1568 | 84.6k | SPEEX_MEMSET(exc, 0, NB_SUBFRAME_SIZE); |
1569 | | |
1570 | | /*Adaptive codebook contribution*/ |
1571 | 84.6k | speex_assert (SUBMODE(ltp_unquant)); |
1572 | 84.6k | { |
1573 | 84.6k | int pit_min, pit_max; |
1574 | | /* Handle pitch constraints if any */ |
1575 | 84.6k | if (SUBMODE(lbr_pitch) != -1) |
1576 | 32.6k | { |
1577 | 32.6k | int margin; |
1578 | 32.6k | margin = SUBMODE(lbr_pitch); |
1579 | 32.6k | if (margin) |
1580 | 0 | { |
1581 | | /* GT - need optimization? |
1582 | | if (ol_pitch < NB_PITCH_START+margin-1) |
1583 | | ol_pitch=NB_PITCH_START+margin-1; |
1584 | | if (ol_pitch > NB_PITCH_END-margin) |
1585 | | ol_pitch=NB_PITCH_END-margin; |
1586 | | pit_min = ol_pitch-margin+1; |
1587 | | pit_max = ol_pitch+margin; |
1588 | | */ |
1589 | 0 | pit_min = ol_pitch-margin+1; |
1590 | 0 | if (pit_min < NB_PITCH_START) |
1591 | 0 | pit_min = NB_PITCH_START; |
1592 | 0 | pit_max = ol_pitch+margin; |
1593 | 0 | if (pit_max > NB_PITCH_END) |
1594 | 0 | pit_max = NB_PITCH_END; |
1595 | 32.6k | } else { |
1596 | 32.6k | pit_min = pit_max = ol_pitch; |
1597 | 32.6k | } |
1598 | 52.0k | } else { |
1599 | 52.0k | pit_min = NB_PITCH_START; |
1600 | 52.0k | pit_max = NB_PITCH_END; |
1601 | 52.0k | } |
1602 | | |
1603 | | |
1604 | | |
1605 | 84.6k | SUBMODE(ltp_unquant)(exc, exc32, pit_min, pit_max, ol_pitch_coef, SUBMODE(ltp_params), |
1606 | 84.6k | NB_SUBFRAME_SIZE, &pitch, &pitch_gain[0], bits, stack, |
1607 | 84.6k | st->count_lost, offset, st->last_pitch_gain, 0); |
1608 | | |
1609 | | /* Ensuring that things aren't blowing up as would happen if e.g. an encoder is |
1610 | | crafting packets to make us produce NaNs and slow down the decoder (vague DoS threat). |
1611 | | We can probably be even more aggressive and limit to 15000 or so. */ |
1612 | 84.6k | sanitize_values32(exc32, NEG32(QCONST32(32000,SIG_SHIFT-1)), QCONST32(32000,SIG_SHIFT-1), NB_SUBFRAME_SIZE); |
1613 | | |
1614 | 84.6k | tmp = gain_3tap_to_1tap(pitch_gain); |
1615 | | |
1616 | 84.6k | pitch_average += tmp; |
1617 | 84.6k | if ((tmp>best_pitch_gain&&ABS(2*best_pitch-pitch)>=3&&ABS(3*best_pitch-pitch)>=4&&ABS(4*best_pitch-pitch)>=5) |
1618 | 84.6k | || (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)) |
1619 | 84.6k | || (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)) ) |
1620 | 12.0k | { |
1621 | 12.0k | best_pitch = pitch; |
1622 | 12.0k | if (tmp > best_pitch_gain) |
1623 | 10.9k | best_pitch_gain = tmp; |
1624 | 12.0k | } |
1625 | 84.6k | } |
1626 | | |
1627 | | /* Unquantize the innovation */ |
1628 | 84.6k | { |
1629 | 84.6k | int q_energy; |
1630 | 84.6k | spx_word32_t ener; |
1631 | | |
1632 | 84.6k | SPEEX_MEMSET(innov, 0, NB_SUBFRAME_SIZE); |
1633 | | |
1634 | | /* Decode sub-frame gain correction */ |
1635 | 84.6k | if (SUBMODE(have_subframe_gain)==3) |
1636 | 23.8k | { |
1637 | 23.8k | q_energy = speex_bits_unpack_unsigned(bits, 3); |
1638 | 23.8k | ener = MULT16_32_Q14(exc_gain_quant_scal3[q_energy],ol_gain); |
1639 | 60.8k | } else if (SUBMODE(have_subframe_gain)==1) |
1640 | 28.2k | { |
1641 | 28.2k | q_energy = speex_bits_unpack_unsigned(bits, 1); |
1642 | 28.2k | ener = MULT16_32_Q14(exc_gain_quant_scal1[q_energy],ol_gain); |
1643 | 32.6k | } else { |
1644 | 32.6k | ener = ol_gain; |
1645 | 32.6k | } |
1646 | | |
1647 | 84.6k | speex_assert (SUBMODE(innovation_unquant)); |
1648 | 84.6k | { |
1649 | | /*Fixed codebook contribution*/ |
1650 | 84.6k | SUBMODE(innovation_unquant)(innov, SUBMODE(innovation_params), NB_SUBFRAME_SIZE, bits, stack, &st->seed); |
1651 | | /* De-normalize innovation and update excitation */ |
1652 | | |
1653 | 84.6k | signal_mul(innov, innov, ener, NB_SUBFRAME_SIZE); |
1654 | | |
1655 | | /* Decode second codebook (only for some modes) */ |
1656 | 84.6k | if (SUBMODE(double_codebook)) |
1657 | 13.9k | { |
1658 | 13.9k | char *tmp_stack=stack; |
1659 | 13.9k | VARDECL(spx_sig_t *innov2); |
1660 | 13.9k | ALLOC(innov2, NB_SUBFRAME_SIZE, spx_sig_t); |
1661 | 13.9k | SPEEX_MEMSET(innov2, 0, NB_SUBFRAME_SIZE); |
1662 | 13.9k | SUBMODE(innovation_unquant)(innov2, SUBMODE(innovation_params), NB_SUBFRAME_SIZE, bits, stack, &st->seed); |
1663 | 13.9k | signal_mul(innov2, innov2, MULT16_32_Q15(QCONST16(0.454545f,15),ener), NB_SUBFRAME_SIZE); |
1664 | 570k | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1665 | 556k | innov[i] = ADD32(innov[i], innov2[i]); |
1666 | 13.9k | stack = tmp_stack; |
1667 | 13.9k | } |
1668 | 3.47M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1669 | 3.38M | exc[i]=EXTRACT16(SATURATE32(PSHR32(ADD32(SHL32(exc32[i],1),innov[i]),SIG_SHIFT),32767)); |
1670 | | /*print_vec(exc, 40, "innov");*/ |
1671 | 84.6k | if (innov_save) |
1672 | 50.9k | { |
1673 | 2.08M | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1674 | 2.03M | innov_save[i] = EXTRACT16(PSHR32(innov[i], SIG_SHIFT)); |
1675 | 50.9k | } |
1676 | 84.6k | } |
1677 | | |
1678 | | /*Vocoder mode*/ |
1679 | 84.6k | if (st->submodeID==1) |
1680 | 14.8k | { |
1681 | 14.8k | spx_word16_t g=ol_pitch_coef; |
1682 | 14.8k | g=MULT16_16_P14(QCONST16(1.5f,14),(g-QCONST16(.2f,6))); |
1683 | 14.8k | if (g<0) |
1684 | 9.64k | g=0; |
1685 | 14.8k | if (g>GAIN_SCALING) |
1686 | 2.37k | g=GAIN_SCALING; |
1687 | | |
1688 | 14.8k | SPEEX_MEMSET(exc, 0, NB_SUBFRAME_SIZE); |
1689 | 37.9k | while (st->voc_offset<NB_SUBFRAME_SIZE) |
1690 | 23.0k | { |
1691 | | /* exc[st->voc_offset]= g*sqrt(2*ol_pitch)*ol_gain; |
1692 | | Not quite sure why we need the factor of two in the sqrt */ |
1693 | 23.0k | if (st->voc_offset>=0) |
1694 | 23.0k | 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))); |
1695 | 23.0k | st->voc_offset+=ol_pitch; |
1696 | 23.0k | } |
1697 | 14.8k | st->voc_offset -= NB_SUBFRAME_SIZE; |
1698 | | |
1699 | 609k | for (i=0;i<NB_SUBFRAME_SIZE;i++) |
1700 | 594k | { |
1701 | 594k | spx_word16_t exci=exc[i]; |
1702 | 594k | exc[i]= ADD16(ADD16(MULT16_16_Q15(QCONST16(.7f,15),exc[i]) , MULT16_16_Q15(QCONST16(.3f,15),st->voc_m1)), |
1703 | 594k | SUB16(MULT16_16_Q15(Q15_ONE-MULT16_16_16(QCONST16(.85f,9),g),EXTRACT16(PSHR32(innov[i],SIG_SHIFT))), |
1704 | 594k | MULT16_16_Q15(MULT16_16_16(QCONST16(.15f,9),g),EXTRACT16(PSHR32(st->voc_m2,SIG_SHIFT))) |
1705 | 594k | )); |
1706 | 594k | st->voc_m1 = exci; |
1707 | 594k | st->voc_m2=innov[i]; |
1708 | 594k | st->voc_mean = EXTRACT16(PSHR32(ADD32(MULT16_16(QCONST16(.8f,15),st->voc_mean), MULT16_16(QCONST16(.2f,15),exc[i])), 15)); |
1709 | 594k | exc[i]-=st->voc_mean; |
1710 | 594k | } |
1711 | 14.8k | } |
1712 | | |
1713 | 84.6k | } |
1714 | 84.6k | } |
1715 | | |
1716 | 21.1k | ALLOC(interp_qlsp, NB_ORDER, spx_lsp_t); |
1717 | | |
1718 | 21.1k | if (st->lpc_enh_enabled && SUBMODE(comb_gain)>0 && !st->count_lost) |
1719 | 17.4k | { |
1720 | 17.4k | multicomb(st->exc-NB_SUBFRAME_SIZE, out, st->interp_qlpc, NB_ORDER, 2*NB_SUBFRAME_SIZE, best_pitch, 40, SUBMODE(comb_gain), stack); |
1721 | 17.4k | 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); |
1722 | 17.4k | } else { |
1723 | 3.71k | SPEEX_COPY(out, &st->exc[-NB_SUBFRAME_SIZE], NB_FRAME_SIZE); |
1724 | 3.71k | } |
1725 | | |
1726 | | /* If the last packet was lost, re-scale the excitation to obtain the same energy as encoded in ol_gain */ |
1727 | 21.1k | if (st->count_lost) |
1728 | 0 | { |
1729 | 0 | spx_word16_t exc_ener; |
1730 | 0 | spx_word32_t gain32; |
1731 | 0 | spx_word16_t gain; |
1732 | 0 | exc_ener = compute_rms16 (st->exc, NB_FRAME_SIZE); |
1733 | 0 | gain32 = PDIV32(ol_gain, ADD16(exc_ener,1)); |
1734 | | #ifdef FIXED_POINT |
1735 | | if (gain32 > 32767) |
1736 | | gain32 = 32767; |
1737 | | gain = EXTRACT16(gain32); |
1738 | | #else |
1739 | 0 | if (gain32 > 2) |
1740 | 0 | gain32=2; |
1741 | 0 | gain = gain32; |
1742 | 0 | #endif |
1743 | 0 | for (i=0;i<NB_FRAME_SIZE;i++) |
1744 | 0 | { |
1745 | 0 | st->exc[i] = MULT16_16_Q14(gain, st->exc[i]); |
1746 | 0 | out[i]=st->exc[i-NB_SUBFRAME_SIZE]; |
1747 | 0 | } |
1748 | 0 | } |
1749 | | |
1750 | | /*Loop on subframes */ |
1751 | 105k | for (sub=0;sub<NB_NB_SUBFRAMES;sub++) |
1752 | 84.6k | { |
1753 | 84.6k | int offset; |
1754 | 84.6k | spx_word16_t *sp; |
1755 | | |
1756 | | /* Offset relative to start of frame */ |
1757 | 84.6k | offset = NB_SUBFRAME_SIZE*sub; |
1758 | | /* Original signal */ |
1759 | 84.6k | sp=out+offset; |
1760 | | |
1761 | | /* LSP interpolation (quantized and unquantized) */ |
1762 | 84.6k | lsp_interpolate(st->old_qlsp, qlsp, interp_qlsp, NB_ORDER, sub, NB_NB_SUBFRAMES, LSP_MARGIN); |
1763 | | |
1764 | | /* Compute interpolated LPCs (unquantized) */ |
1765 | 84.6k | lsp_to_lpc(interp_qlsp, ak, NB_ORDER, stack); |
1766 | | |
1767 | | /* Compute analysis filter at w=pi */ |
1768 | 84.6k | { |
1769 | 84.6k | spx_word32_t pi_g=LPC_SCALING; |
1770 | 508k | for (i=0;i<NB_ORDER;i+=2) |
1771 | 423k | { |
1772 | | /*pi_g += -st->interp_qlpc[i] + st->interp_qlpc[i+1];*/ |
1773 | 423k | pi_g = ADD32(pi_g, SUB32(EXTEND32(ak[i+1]),EXTEND32(ak[i]))); |
1774 | 423k | } |
1775 | 84.6k | st->pi_gain[sub] = pi_g; |
1776 | 84.6k | } |
1777 | | |
1778 | 84.6k | iir_mem16(sp, st->interp_qlpc, sp, NB_SUBFRAME_SIZE, NB_ORDER, |
1779 | 84.6k | st->mem_sp, stack); |
1780 | | |
1781 | 931k | for (i=0;i<NB_ORDER;i++) |
1782 | 846k | st->interp_qlpc[i] = ak[i]; |
1783 | | |
1784 | 84.6k | } |
1785 | | |
1786 | 21.1k | if (st->highpass_enabled) |
1787 | 21.1k | highpass(out, out, NB_FRAME_SIZE, (st->isWideband?HIGHPASS_WIDEBAND:HIGHPASS_NARROWBAND)|HIGHPASS_OUTPUT, st->mem_hp); |
1788 | | /*for (i=0;i<NB_FRAME_SIZE;i++) |
1789 | | printf ("%d\n", (int)st->frame[i]);*/ |
1790 | | |
1791 | | /* Tracking output level */ |
1792 | 21.1k | st->level = 1+PSHR32(ol_gain,SIG_SHIFT); |
1793 | 21.1k | st->max_level = MAX16(MULT16_16_Q15(QCONST16(.99f,15), st->max_level), st->level); |
1794 | 21.1k | st->min_level = MIN16(ADD16(1,MULT16_16_Q14(QCONST16(1.01f,14), st->min_level)), st->level); |
1795 | 21.1k | if (st->max_level < st->min_level+1) |
1796 | 7.36k | st->max_level = st->min_level+1; |
1797 | | /*printf ("%f %f %f %d\n", og, st->min_level, st->max_level, update);*/ |
1798 | | |
1799 | | /* Store the LSPs for interpolation in the next frame */ |
1800 | 232k | for (i=0;i<NB_ORDER;i++) |
1801 | 211k | st->old_qlsp[i] = qlsp[i]; |
1802 | | |
1803 | | /* The next frame will not be the first (Duh!) */ |
1804 | 21.1k | st->first = 0; |
1805 | 21.1k | st->count_lost=0; |
1806 | 21.1k | st->last_pitch = best_pitch; |
1807 | | #ifdef FIXED_POINT |
1808 | | st->last_pitch_gain = PSHR16(pitch_average,2); |
1809 | | #else |
1810 | 21.1k | st->last_pitch_gain = .25*pitch_average; |
1811 | 21.1k | #endif |
1812 | 21.1k | st->pitch_gain_buf[st->pitch_gain_buf_idx++] = st->last_pitch_gain; |
1813 | 21.1k | if (st->pitch_gain_buf_idx > 2) /* rollover */ |
1814 | 6.56k | st->pitch_gain_buf_idx = 0; |
1815 | | |
1816 | 21.1k | st->last_ol_gain = ol_gain; |
1817 | | |
1818 | 21.1k | return 0; |
1819 | 40.1k | } |
1820 | | #endif /* DISABLE_DECODER */ |
1821 | | |