/src/opus/silk/float/encode_frame_FLP.c
Line | Count | Source (jump to first uncovered line) |
1 | | /*********************************************************************** |
2 | | Copyright (c) 2006-2011, Skype Limited. All rights reserved. |
3 | | Redistribution and use in source and binary forms, with or without |
4 | | modification, are permitted provided that the following conditions |
5 | | are met: |
6 | | - Redistributions of source code must retain the above copyright notice, |
7 | | this list of conditions and the following disclaimer. |
8 | | - Redistributions in binary form must reproduce the above copyright |
9 | | notice, this list of conditions and the following disclaimer in the |
10 | | documentation and/or other materials provided with the distribution. |
11 | | - Neither the name of Internet Society, IETF or IETF Trust, nor the |
12 | | names of specific contributors, may be used to endorse or promote |
13 | | products derived from this software without specific prior written |
14 | | permission. |
15 | | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" |
16 | | AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE |
17 | | IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE |
18 | | ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE |
19 | | LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR |
20 | | CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF |
21 | | SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS |
22 | | INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN |
23 | | CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) |
24 | | ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE |
25 | | POSSIBILITY OF SUCH DAMAGE. |
26 | | ***********************************************************************/ |
27 | | |
28 | | #ifdef HAVE_CONFIG_H |
29 | | #include "config.h" |
30 | | #endif |
31 | | |
32 | | #include <stdlib.h> |
33 | | #include "main_FLP.h" |
34 | | #include "tuning_parameters.h" |
35 | | |
36 | | /* Low Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode with lower bitrate */ |
37 | | static OPUS_INLINE void silk_LBRR_encode_FLP( |
38 | | silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ |
39 | | silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ |
40 | | const silk_float xfw[], /* I Input signal */ |
41 | | opus_int condCoding /* I The type of conditional coding used so far for this frame */ |
42 | | ); |
43 | | |
44 | | void silk_encode_do_VAD_FLP( |
45 | | silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ |
46 | | opus_int activity /* I Decision of Opus voice activity detector */ |
47 | | ) |
48 | 0 | { |
49 | 0 | const opus_int activity_threshold = SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THRES, 8 ); |
50 | | |
51 | | /****************************/ |
52 | | /* Voice Activity Detection */ |
53 | | /****************************/ |
54 | 0 | silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.arch ); |
55 | | /* If Opus VAD is inactive and Silk VAD is active: lower Silk VAD to just under the threshold */ |
56 | 0 | if( activity == VAD_NO_ACTIVITY && psEnc->sCmn.speech_activity_Q8 >= activity_threshold ) { |
57 | 0 | psEnc->sCmn.speech_activity_Q8 = activity_threshold - 1; |
58 | 0 | } |
59 | | |
60 | | /**************************************************/ |
61 | | /* Convert speech activity into VAD and DTX flags */ |
62 | | /**************************************************/ |
63 | 0 | if( psEnc->sCmn.speech_activity_Q8 < activity_threshold ) { |
64 | 0 | psEnc->sCmn.indices.signalType = TYPE_NO_VOICE_ACTIVITY; |
65 | 0 | psEnc->sCmn.noSpeechCounter++; |
66 | 0 | if( psEnc->sCmn.noSpeechCounter <= NB_SPEECH_FRAMES_BEFORE_DTX ) { |
67 | 0 | psEnc->sCmn.inDTX = 0; |
68 | 0 | } else if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NB_SPEECH_FRAMES_BEFORE_DTX ) { |
69 | 0 | psEnc->sCmn.noSpeechCounter = NB_SPEECH_FRAMES_BEFORE_DTX; |
70 | 0 | psEnc->sCmn.inDTX = 0; |
71 | 0 | } |
72 | 0 | psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 0; |
73 | 0 | } else { |
74 | 0 | psEnc->sCmn.noSpeechCounter = 0; |
75 | 0 | psEnc->sCmn.inDTX = 0; |
76 | 0 | psEnc->sCmn.indices.signalType = TYPE_UNVOICED; |
77 | 0 | psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 1; |
78 | 0 | } |
79 | 0 | } |
80 | | |
81 | | /****************/ |
82 | | /* Encode frame */ |
83 | | /****************/ |
84 | | opus_int silk_encode_frame_FLP( |
85 | | silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ |
86 | | opus_int32 *pnBytesOut, /* O Number of payload bytes; */ |
87 | | ec_enc *psRangeEnc, /* I/O compressor data structure */ |
88 | | opus_int condCoding, /* I The type of conditional coding to use */ |
89 | | opus_int maxBits, /* I If > 0: maximum number of output bits */ |
90 | | opus_int useCBR /* I Flag to force constant-bitrate operation */ |
91 | | ) |
92 | 0 | { |
93 | 0 | silk_encoder_control_FLP sEncCtrl; |
94 | 0 | opus_int i, iter, maxIter, found_upper, found_lower, ret = 0; |
95 | 0 | silk_float *x_frame, *res_pitch_frame; |
96 | 0 | silk_float res_pitch[ 2 * MAX_FRAME_LENGTH + LA_PITCH_MAX ]; |
97 | 0 | ec_enc sRangeEnc_copy, sRangeEnc_copy2; |
98 | 0 | silk_nsq_state sNSQ_copy, sNSQ_copy2; |
99 | 0 | opus_int32 seed_copy, nBits, nBits_lower, nBits_upper, gainMult_lower, gainMult_upper; |
100 | 0 | opus_int32 gainsID, gainsID_lower, gainsID_upper; |
101 | 0 | opus_int16 gainMult_Q8; |
102 | 0 | opus_int16 ec_prevLagIndex_copy; |
103 | 0 | opus_int ec_prevSignalType_copy; |
104 | 0 | opus_int8 LastGainIndex_copy2; |
105 | 0 | opus_int32 pGains_Q16[ MAX_NB_SUBFR ]; |
106 | 0 | opus_uint8 ec_buf_copy[ 1275 ]; |
107 | 0 | opus_int gain_lock[ MAX_NB_SUBFR ] = {0}; |
108 | 0 | opus_int16 best_gain_mult[ MAX_NB_SUBFR ]; |
109 | 0 | opus_int best_sum[ MAX_NB_SUBFR ]; |
110 | 0 | opus_int bits_margin; |
111 | | |
112 | | /* For CBR, 5 bits below budget is close enough. For VBR, allow up to 25% below the cap if we initially busted the budget. */ |
113 | 0 | bits_margin = useCBR ? 5 : maxBits/4; |
114 | | /* This is totally unnecessary but many compilers (including gcc) are too dumb to realise it */ |
115 | 0 | LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_upper = 0; |
116 | |
|
117 | 0 | psEnc->sCmn.indices.Seed = psEnc->sCmn.frameCounter++ & 3; |
118 | | |
119 | | /**************************************************************/ |
120 | | /* Set up Input Pointers, and insert frame in input buffer */ |
121 | | /**************************************************************/ |
122 | | /* pointers aligned with start of frame to encode */ |
123 | 0 | x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length; /* start of frame to encode */ |
124 | 0 | res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length; /* start of pitch LPC residual frame */ |
125 | | |
126 | | /***************************************/ |
127 | | /* Ensure smooth bandwidth transitions */ |
128 | | /***************************************/ |
129 | 0 | silk_LP_variable_cutoff( &psEnc->sCmn.sLP, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length ); |
130 | | |
131 | | /*******************************************/ |
132 | | /* Copy new frame to front of input buffer */ |
133 | | /*******************************************/ |
134 | 0 | silk_short2float_array( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, psEnc->sCmn.inputBuf + 1, psEnc->sCmn.frame_length ); |
135 | | |
136 | | /* Add tiny signal to avoid high CPU load from denormalized floating point numbers */ |
137 | 0 | for( i = 0; i < 8; i++ ) { |
138 | 0 | x_frame[ LA_SHAPE_MS * psEnc->sCmn.fs_kHz + i * ( psEnc->sCmn.frame_length >> 3 ) ] += ( 1 - ( i & 2 ) ) * 1e-6f; |
139 | 0 | } |
140 | |
|
141 | 0 | if( !psEnc->sCmn.prefillFlag ) { |
142 | | /*****************************************/ |
143 | | /* Find pitch lags, initial LPC analysis */ |
144 | | /*****************************************/ |
145 | 0 | silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame, psEnc->sCmn.arch ); |
146 | | |
147 | | /************************/ |
148 | | /* Noise shape analysis */ |
149 | | /************************/ |
150 | 0 | silk_noise_shape_analysis_FLP( psEnc, &sEncCtrl, res_pitch_frame, x_frame ); |
151 | | |
152 | | /***************************************************/ |
153 | | /* Find linear prediction coefficients (LPC + LTP) */ |
154 | | /***************************************************/ |
155 | 0 | silk_find_pred_coefs_FLP( psEnc, &sEncCtrl, res_pitch_frame, x_frame, condCoding ); |
156 | | |
157 | | /****************************************/ |
158 | | /* Process gains */ |
159 | | /****************************************/ |
160 | 0 | silk_process_gains_FLP( psEnc, &sEncCtrl, condCoding ); |
161 | | |
162 | | /****************************************/ |
163 | | /* Low Bitrate Redundant Encoding */ |
164 | | /****************************************/ |
165 | 0 | silk_LBRR_encode_FLP( psEnc, &sEncCtrl, x_frame, condCoding ); |
166 | | |
167 | | /* Loop over quantizer and entroy coding to control bitrate */ |
168 | 0 | maxIter = 6; |
169 | 0 | gainMult_Q8 = SILK_FIX_CONST( 1, 8 ); |
170 | 0 | found_lower = 0; |
171 | 0 | found_upper = 0; |
172 | 0 | gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr ); |
173 | 0 | gainsID_lower = -1; |
174 | 0 | gainsID_upper = -1; |
175 | | /* Copy part of the input state */ |
176 | 0 | silk_memcpy( &sRangeEnc_copy, psRangeEnc, sizeof( ec_enc ) ); |
177 | 0 | silk_memcpy( &sNSQ_copy, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); |
178 | 0 | seed_copy = psEnc->sCmn.indices.Seed; |
179 | 0 | ec_prevLagIndex_copy = psEnc->sCmn.ec_prevLagIndex; |
180 | 0 | ec_prevSignalType_copy = psEnc->sCmn.ec_prevSignalType; |
181 | 0 | for( iter = 0; ; iter++ ) { |
182 | 0 | if( gainsID == gainsID_lower ) { |
183 | 0 | nBits = nBits_lower; |
184 | 0 | } else if( gainsID == gainsID_upper ) { |
185 | 0 | nBits = nBits_upper; |
186 | 0 | } else { |
187 | | /* Restore part of the input state */ |
188 | 0 | if( iter > 0 ) { |
189 | 0 | silk_memcpy( psRangeEnc, &sRangeEnc_copy, sizeof( ec_enc ) ); |
190 | 0 | silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy, sizeof( silk_nsq_state ) ); |
191 | 0 | psEnc->sCmn.indices.Seed = seed_copy; |
192 | 0 | psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; |
193 | 0 | psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; |
194 | 0 | } |
195 | | |
196 | | /*****************************************/ |
197 | | /* Noise shaping quantization */ |
198 | | /*****************************************/ |
199 | 0 | silk_NSQ_wrapper_FLP( psEnc, &sEncCtrl, &psEnc->sCmn.indices, &psEnc->sCmn.sNSQ, psEnc->sCmn.pulses, x_frame ); |
200 | |
|
201 | 0 | if ( iter == maxIter && !found_lower ) { |
202 | 0 | silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); |
203 | 0 | } |
204 | | |
205 | | /****************************************/ |
206 | | /* Encode Parameters */ |
207 | | /****************************************/ |
208 | 0 | silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding ); |
209 | | |
210 | | /****************************************/ |
211 | | /* Encode Excitation Signal */ |
212 | | /****************************************/ |
213 | 0 | silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType, |
214 | 0 | psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); |
215 | |
|
216 | 0 | nBits = ec_tell( psRangeEnc ); |
217 | | |
218 | | /* If we still bust after the last iteration, do some damage control. */ |
219 | 0 | if ( iter == maxIter && !found_lower && nBits > maxBits ) { |
220 | 0 | silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); |
221 | | |
222 | | /* Keep gains the same as the last frame. */ |
223 | 0 | psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; |
224 | 0 | for ( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { |
225 | 0 | psEnc->sCmn.indices.GainsIndices[ i ] = 4; |
226 | 0 | } |
227 | 0 | if (condCoding != CODE_CONDITIONALLY) { |
228 | 0 | psEnc->sCmn.indices.GainsIndices[ 0 ] = sEncCtrl.lastGainIndexPrev; |
229 | 0 | } |
230 | 0 | psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; |
231 | 0 | psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; |
232 | | /* Clear all pulses. */ |
233 | 0 | for ( i = 0; i < psEnc->sCmn.frame_length; i++ ) { |
234 | 0 | psEnc->sCmn.pulses[ i ] = 0; |
235 | 0 | } |
236 | |
|
237 | 0 | silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFramesEncoded, 0, condCoding ); |
238 | |
|
239 | 0 | silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType, psEnc->sCmn.indices.quantOffsetType, |
240 | 0 | psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); |
241 | |
|
242 | 0 | nBits = ec_tell( psRangeEnc ); |
243 | 0 | } |
244 | |
|
245 | 0 | if( useCBR == 0 && iter == 0 && nBits <= maxBits ) { |
246 | 0 | break; |
247 | 0 | } |
248 | 0 | } |
249 | | |
250 | 0 | if( iter == maxIter ) { |
251 | 0 | if( found_lower && ( gainsID == gainsID_lower || nBits > maxBits ) ) { |
252 | | /* Restore output state from earlier iteration that did meet the bitrate budget */ |
253 | 0 | silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc ) ); |
254 | 0 | celt_assert( sRangeEnc_copy2.offs <= 1275 ); |
255 | 0 | silk_memcpy( psRangeEnc->buf, ec_buf_copy, sRangeEnc_copy2.offs ); |
256 | 0 | silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy2, sizeof( silk_nsq_state ) ); |
257 | 0 | psEnc->sShape.LastGainIndex = LastGainIndex_copy2; |
258 | 0 | } |
259 | 0 | break; |
260 | 0 | } |
261 | | |
262 | 0 | if( nBits > maxBits ) { |
263 | 0 | if( found_lower == 0 && iter >= 2 ) { |
264 | | /* Adjust the quantizer's rate/distortion tradeoff and discard previous "upper" results */ |
265 | 0 | sEncCtrl.Lambda = silk_max_float(sEncCtrl.Lambda*1.5f, 1.5f); |
266 | | /* Reducing dithering can help us hit the target. */ |
267 | 0 | psEnc->sCmn.indices.quantOffsetType = 0; |
268 | 0 | found_upper = 0; |
269 | 0 | gainsID_upper = -1; |
270 | 0 | } else { |
271 | 0 | found_upper = 1; |
272 | 0 | nBits_upper = nBits; |
273 | 0 | gainMult_upper = gainMult_Q8; |
274 | 0 | gainsID_upper = gainsID; |
275 | 0 | } |
276 | 0 | } else if( nBits < maxBits - bits_margin ) { |
277 | 0 | found_lower = 1; |
278 | 0 | nBits_lower = nBits; |
279 | 0 | gainMult_lower = gainMult_Q8; |
280 | 0 | if( gainsID != gainsID_lower ) { |
281 | 0 | gainsID_lower = gainsID; |
282 | | /* Copy part of the output state */ |
283 | 0 | silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc ) ); |
284 | 0 | celt_assert( psRangeEnc->offs <= 1275 ); |
285 | 0 | silk_memcpy( ec_buf_copy, psRangeEnc->buf, psRangeEnc->offs ); |
286 | 0 | silk_memcpy( &sNSQ_copy2, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); |
287 | 0 | LastGainIndex_copy2 = psEnc->sShape.LastGainIndex; |
288 | 0 | } |
289 | 0 | } else { |
290 | | /* Close enough */ |
291 | 0 | break; |
292 | 0 | } |
293 | | |
294 | 0 | if ( !found_lower && nBits > maxBits ) { |
295 | 0 | int j; |
296 | 0 | for ( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { |
297 | 0 | int sum=0; |
298 | 0 | for ( j = i*psEnc->sCmn.subfr_length; j < (i+1)*psEnc->sCmn.subfr_length; j++ ) { |
299 | 0 | sum += abs( psEnc->sCmn.pulses[j] ); |
300 | 0 | } |
301 | 0 | if ( iter == 0 || (sum < best_sum[i] && !gain_lock[i]) ) { |
302 | 0 | best_sum[i] = sum; |
303 | 0 | best_gain_mult[i] = gainMult_Q8; |
304 | 0 | } else { |
305 | 0 | gain_lock[i] = 1; |
306 | 0 | } |
307 | 0 | } |
308 | 0 | } |
309 | 0 | if( ( found_lower & found_upper ) == 0 ) { |
310 | | /* Adjust gain according to high-rate rate/distortion curve */ |
311 | 0 | if( nBits > maxBits ) { |
312 | 0 | gainMult_Q8 = silk_min_32( 1024, gainMult_Q8*3/2 ); |
313 | 0 | } else { |
314 | 0 | gainMult_Q8 = silk_max_32( 64, gainMult_Q8*4/5 ); |
315 | 0 | } |
316 | 0 | } else { |
317 | | /* Adjust gain by interpolating */ |
318 | 0 | gainMult_Q8 = gainMult_lower + ( ( gainMult_upper - gainMult_lower ) * ( maxBits - nBits_lower ) ) / ( nBits_upper - nBits_lower ); |
319 | | /* New gain multplier must be between 25% and 75% of old range (note that gainMult_upper < gainMult_lower) */ |
320 | 0 | if( gainMult_Q8 > silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ) ) { |
321 | 0 | gainMult_Q8 = silk_ADD_RSHIFT32( gainMult_lower, gainMult_upper - gainMult_lower, 2 ); |
322 | 0 | } else |
323 | 0 | if( gainMult_Q8 < silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ) ) { |
324 | 0 | gainMult_Q8 = silk_SUB_RSHIFT32( gainMult_upper, gainMult_upper - gainMult_lower, 2 ); |
325 | 0 | } |
326 | 0 | } |
327 | |
|
328 | 0 | for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { |
329 | 0 | opus_int16 tmp; |
330 | 0 | if ( gain_lock[i] ) { |
331 | 0 | tmp = best_gain_mult[i]; |
332 | 0 | } else { |
333 | 0 | tmp = gainMult_Q8; |
334 | 0 | } |
335 | 0 | pGains_Q16[ i ] = silk_LSHIFT_SAT32( silk_SMULWB( sEncCtrl.GainsUnq_Q16[ i ], tmp ), 8 ); |
336 | 0 | } |
337 | | |
338 | | /* Quantize gains */ |
339 | 0 | psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; |
340 | 0 | silk_gains_quant( psEnc->sCmn.indices.GainsIndices, pGains_Q16, |
341 | 0 | &psEnc->sShape.LastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); |
342 | | |
343 | | /* Unique identifier of gains vector */ |
344 | 0 | gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.nb_subfr ); |
345 | | |
346 | | /* Overwrite unquantized gains with quantized gains and convert back to Q0 from Q16 */ |
347 | 0 | for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { |
348 | 0 | sEncCtrl.Gains[ i ] = pGains_Q16[ i ] / 65536.0f; |
349 | 0 | } |
350 | 0 | } |
351 | 0 | } |
352 | | |
353 | | /* Update input buffer */ |
354 | 0 | silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ], |
355 | 0 | ( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * sizeof( silk_float ) ); |
356 | | |
357 | | /* Exit without entropy coding */ |
358 | 0 | if( psEnc->sCmn.prefillFlag ) { |
359 | | /* No payload */ |
360 | 0 | *pnBytesOut = 0; |
361 | 0 | return ret; |
362 | 0 | } |
363 | | |
364 | | /* Parameters needed for next frame */ |
365 | 0 | psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; |
366 | 0 | psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; |
367 | | |
368 | | /****************************************/ |
369 | | /* Finalize payload */ |
370 | | /****************************************/ |
371 | 0 | psEnc->sCmn.first_frame_after_reset = 0; |
372 | | /* Payload size */ |
373 | 0 | *pnBytesOut = silk_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 ); |
374 | |
|
375 | 0 | return ret; |
376 | 0 | } |
377 | | |
378 | | /* Low-Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode excitation at lower bitrate */ |
379 | | static OPUS_INLINE void silk_LBRR_encode_FLP( |
380 | | silk_encoder_state_FLP *psEnc, /* I/O Encoder state FLP */ |
381 | | silk_encoder_control_FLP *psEncCtrl, /* I/O Encoder control FLP */ |
382 | | const silk_float xfw[], /* I Input signal */ |
383 | | opus_int condCoding /* I The type of conditional coding used so far for this frame */ |
384 | | ) |
385 | 0 | { |
386 | 0 | opus_int k; |
387 | 0 | opus_int32 Gains_Q16[ MAX_NB_SUBFR ]; |
388 | 0 | silk_float TempGains[ MAX_NB_SUBFR ]; |
389 | 0 | SideInfoIndices *psIndices_LBRR = &psEnc->sCmn.indices_LBRR[ psEnc->sCmn.nFramesEncoded ]; |
390 | 0 | silk_nsq_state sNSQ_LBRR; |
391 | | |
392 | | /*******************************************/ |
393 | | /* Control use of inband LBRR */ |
394 | | /*******************************************/ |
395 | 0 | if( psEnc->sCmn.LBRR_enabled && psEnc->sCmn.speech_activity_Q8 > SILK_FIX_CONST( LBRR_SPEECH_ACTIVITY_THRES, 8 ) ) { |
396 | 0 | psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded ] = 1; |
397 | | |
398 | | /* Copy noise shaping quantizer state and quantization indices from regular encoding */ |
399 | 0 | silk_memcpy( &sNSQ_LBRR, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); |
400 | 0 | silk_memcpy( psIndices_LBRR, &psEnc->sCmn.indices, sizeof( SideInfoIndices ) ); |
401 | | |
402 | | /* Save original gains */ |
403 | 0 | silk_memcpy( TempGains, psEncCtrl->Gains, psEnc->sCmn.nb_subfr * sizeof( silk_float ) ); |
404 | |
|
405 | 0 | if( psEnc->sCmn.nFramesEncoded == 0 || psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded - 1 ] == 0 ) { |
406 | | /* First frame in packet or previous frame not LBRR coded */ |
407 | 0 | psEnc->sCmn.LBRRprevLastGainIndex = psEnc->sShape.LastGainIndex; |
408 | | |
409 | | /* Increase Gains to get target LBRR rate */ |
410 | 0 | psIndices_LBRR->GainsIndices[ 0 ] += psEnc->sCmn.LBRR_GainIncreases; |
411 | 0 | psIndices_LBRR->GainsIndices[ 0 ] = silk_min_int( psIndices_LBRR->GainsIndices[ 0 ], N_LEVELS_QGAIN - 1 ); |
412 | 0 | } |
413 | | |
414 | | /* Decode to get gains in sync with decoder */ |
415 | 0 | silk_gains_dequant( Gains_Q16, psIndices_LBRR->GainsIndices, |
416 | 0 | &psEnc->sCmn.LBRRprevLastGainIndex, condCoding == CODE_CONDITIONALLY, psEnc->sCmn.nb_subfr ); |
417 | | |
418 | | /* Overwrite unquantized gains with quantized gains and convert back to Q0 from Q16 */ |
419 | 0 | for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { |
420 | 0 | psEncCtrl->Gains[ k ] = Gains_Q16[ k ] * ( 1.0f / 65536.0f ); |
421 | 0 | } |
422 | | |
423 | | /*****************************************/ |
424 | | /* Noise shaping quantization */ |
425 | | /*****************************************/ |
426 | 0 | silk_NSQ_wrapper_FLP( psEnc, psEncCtrl, psIndices_LBRR, &sNSQ_LBRR, |
427 | 0 | psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], xfw ); |
428 | | |
429 | | /* Restore original gains */ |
430 | 0 | silk_memcpy( psEncCtrl->Gains, TempGains, psEnc->sCmn.nb_subfr * sizeof( silk_float ) ); |
431 | 0 | } |
432 | 0 | } |