| OLD | NEW |
| (Empty) | |
| 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 "main_FLP.h" |
| 33 #include "tuning_parameters.h" |
| 34 |
| 35 /* Low Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode with
lower bitrate */ |
| 36 static inline void silk_LBRR_encode_FLP( |
| 37 silk_encoder_state_FLP *psEnc, /* I/O
Encoder state FLP */ |
| 38 silk_encoder_control_FLP *psEncCtrl, /* I/O
Encoder control FLP */ |
| 39 const silk_float xfw[], /* I
Input signal */ |
| 40 opus_int condCoding /* I
The type of conditional coding used so far for this frame */ |
| 41 ); |
| 42 |
| 43 void silk_encode_do_VAD_FLP( |
| 44 silk_encoder_state_FLP *psEnc /* I/O
Encoder state FLP */ |
| 45 ) |
| 46 { |
| 47 /****************************/ |
| 48 /* Voice Activity Detection */ |
| 49 /****************************/ |
| 50 silk_VAD_GetSA_Q8( &psEnc->sCmn, psEnc->sCmn.inputBuf + 1 ); |
| 51 |
| 52 /**************************************************/ |
| 53 /* Convert speech activity into VAD and DTX flags */ |
| 54 /**************************************************/ |
| 55 if( psEnc->sCmn.speech_activity_Q8 < SILK_FIX_CONST( SPEECH_ACTIVITY_DTX_THR
ES, 8 ) ) { |
| 56 psEnc->sCmn.indices.signalType = TYPE_NO_VOICE_ACTIVITY; |
| 57 psEnc->sCmn.noSpeechCounter++; |
| 58 if( psEnc->sCmn.noSpeechCounter < NB_SPEECH_FRAMES_BEFORE_DTX ) { |
| 59 psEnc->sCmn.inDTX = 0; |
| 60 } else if( psEnc->sCmn.noSpeechCounter > MAX_CONSECUTIVE_DTX + NB_SPEECH
_FRAMES_BEFORE_DTX ) { |
| 61 psEnc->sCmn.noSpeechCounter = NB_SPEECH_FRAMES_BEFORE_DTX; |
| 62 psEnc->sCmn.inDTX = 0; |
| 63 } |
| 64 psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 0; |
| 65 } else { |
| 66 psEnc->sCmn.noSpeechCounter = 0; |
| 67 psEnc->sCmn.inDTX = 0; |
| 68 psEnc->sCmn.indices.signalType = TYPE_UNVOICED; |
| 69 psEnc->sCmn.VAD_flags[ psEnc->sCmn.nFramesEncoded ] = 1; |
| 70 } |
| 71 } |
| 72 |
| 73 /****************/ |
| 74 /* Encode frame */ |
| 75 /****************/ |
| 76 opus_int silk_encode_frame_FLP( |
| 77 silk_encoder_state_FLP *psEnc, /* I/O
Encoder state FLP */ |
| 78 opus_int32 *pnBytesOut, /* O
Number of payload bytes; */ |
| 79 ec_enc *psRangeEnc, /* I/O
compressor data structure */ |
| 80 opus_int condCoding, /* I
The type of conditional coding to use */ |
| 81 opus_int maxBits, /* I
If > 0: maximum number of output bits */ |
| 82 opus_int useCBR /* I
Flag to force constant-bitrate operation */ |
| 83 ) |
| 84 { |
| 85 silk_encoder_control_FLP sEncCtrl; |
| 86 opus_int i, iter, maxIter, found_upper, found_lower, ret = 0; |
| 87 silk_float *x_frame, *res_pitch_frame; |
| 88 silk_float xfw[ MAX_FRAME_LENGTH ]; |
| 89 silk_float res_pitch[ 2 * MAX_FRAME_LENGTH + LA_PITCH_MAX ]; |
| 90 ec_enc sRangeEnc_copy, sRangeEnc_copy2; |
| 91 silk_nsq_state sNSQ_copy, sNSQ_copy2; |
| 92 opus_int32 seed_copy, nBits, nBits_lower, nBits_upper, gainMult_lower, gai
nMult_upper; |
| 93 opus_int32 gainsID, gainsID_lower, gainsID_upper; |
| 94 opus_int16 gainMult_Q8; |
| 95 opus_int16 ec_prevLagIndex_copy; |
| 96 opus_int ec_prevSignalType_copy; |
| 97 opus_int8 LastGainIndex_copy2; |
| 98 opus_int32 pGains_Q16[ MAX_NB_SUBFR ]; |
| 99 opus_uint8 ec_buf_copy[ 1275 ]; |
| 100 |
| 101 /* This is totally unnecessary but many compilers (including gcc) are too du
mb to realise it */ |
| 102 LastGainIndex_copy2 = nBits_lower = nBits_upper = gainMult_lower = gainMult_
upper = 0; |
| 103 |
| 104 psEnc->sCmn.indices.Seed = psEnc->sCmn.frameCounter++ & 3; |
| 105 |
| 106 /**************************************************************/ |
| 107 /* Set up Input Pointers, and insert frame in input buffer */ |
| 108 /**************************************************************/ |
| 109 /* pointers aligned with start of frame to encode */ |
| 110 x_frame = psEnc->x_buf + psEnc->sCmn.ltp_mem_length; /* start of
frame to encode */ |
| 111 res_pitch_frame = res_pitch + psEnc->sCmn.ltp_mem_length; /* start of
pitch LPC residual frame */ |
| 112 |
| 113 /***************************************/ |
| 114 /* Ensure smooth bandwidth transitions */ |
| 115 /***************************************/ |
| 116 silk_LP_variable_cutoff( &psEnc->sCmn.sLP, psEnc->sCmn.inputBuf + 1, psEnc->
sCmn.frame_length ); |
| 117 |
| 118 /*******************************************/ |
| 119 /* Copy new frame to front of input buffer */ |
| 120 /*******************************************/ |
| 121 silk_short2float_array( x_frame + LA_SHAPE_MS * psEnc->sCmn.fs_kHz, psEnc->s
Cmn.inputBuf + 1, psEnc->sCmn.frame_length ); |
| 122 |
| 123 /* Add tiny signal to avoid high CPU load from denormalized floating point n
umbers */ |
| 124 for( i = 0; i < 8; i++ ) { |
| 125 x_frame[ LA_SHAPE_MS * psEnc->sCmn.fs_kHz + i * ( psEnc->sCmn.frame_leng
th >> 3 ) ] += ( 1 - ( i & 2 ) ) * 1e-6f; |
| 126 } |
| 127 |
| 128 if( !psEnc->sCmn.prefillFlag ) { |
| 129 /*****************************************/ |
| 130 /* Find pitch lags, initial LPC analysis */ |
| 131 /*****************************************/ |
| 132 silk_find_pitch_lags_FLP( psEnc, &sEncCtrl, res_pitch, x_frame ); |
| 133 |
| 134 /************************/ |
| 135 /* Noise shape analysis */ |
| 136 /************************/ |
| 137 silk_noise_shape_analysis_FLP( psEnc, &sEncCtrl, res_pitch_frame, x_fram
e ); |
| 138 |
| 139 /***************************************************/ |
| 140 /* Find linear prediction coefficients (LPC + LTP) */ |
| 141 /***************************************************/ |
| 142 silk_find_pred_coefs_FLP( psEnc, &sEncCtrl, res_pitch, x_frame, condCodi
ng ); |
| 143 |
| 144 /****************************************/ |
| 145 /* Process gains */ |
| 146 /****************************************/ |
| 147 silk_process_gains_FLP( psEnc, &sEncCtrl, condCoding ); |
| 148 |
| 149 /*****************************************/ |
| 150 /* Prefiltering for noise shaper */ |
| 151 /*****************************************/ |
| 152 silk_prefilter_FLP( psEnc, &sEncCtrl, xfw, x_frame ); |
| 153 |
| 154 /****************************************/ |
| 155 /* Low Bitrate Redundant Encoding */ |
| 156 /****************************************/ |
| 157 silk_LBRR_encode_FLP( psEnc, &sEncCtrl, xfw, condCoding ); |
| 158 |
| 159 /* Loop over quantizer and entroy coding to control bitrate */ |
| 160 maxIter = 6; |
| 161 gainMult_Q8 = SILK_FIX_CONST( 1, 8 ); |
| 162 found_lower = 0; |
| 163 found_upper = 0; |
| 164 gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sCmn.n
b_subfr ); |
| 165 gainsID_lower = -1; |
| 166 gainsID_upper = -1; |
| 167 /* Copy part of the input state */ |
| 168 silk_memcpy( &sRangeEnc_copy, psRangeEnc, sizeof( ec_enc ) ); |
| 169 silk_memcpy( &sNSQ_copy, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); |
| 170 seed_copy = psEnc->sCmn.indices.Seed; |
| 171 ec_prevLagIndex_copy = psEnc->sCmn.ec_prevLagIndex; |
| 172 ec_prevSignalType_copy = psEnc->sCmn.ec_prevSignalType; |
| 173 for( iter = 0; ; iter++ ) { |
| 174 if( gainsID == gainsID_lower ) { |
| 175 nBits = nBits_lower; |
| 176 } else if( gainsID == gainsID_upper ) { |
| 177 nBits = nBits_upper; |
| 178 } else { |
| 179 /* Restore part of the input state */ |
| 180 if( iter > 0 ) { |
| 181 silk_memcpy( psRangeEnc, &sRangeEnc_copy, sizeof( ec_enc ) )
; |
| 182 silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy, sizeof( silk_nsq
_state ) ); |
| 183 psEnc->sCmn.indices.Seed = seed_copy; |
| 184 psEnc->sCmn.ec_prevLagIndex = ec_prevLagIndex_copy; |
| 185 psEnc->sCmn.ec_prevSignalType = ec_prevSignalType_copy; |
| 186 } |
| 187 |
| 188 /*****************************************/ |
| 189 /* Noise shaping quantization */ |
| 190 /*****************************************/ |
| 191 silk_NSQ_wrapper_FLP( psEnc, &sEncCtrl, &psEnc->sCmn.indices, &p
sEnc->sCmn.sNSQ, psEnc->sCmn.pulses, xfw ); |
| 192 |
| 193 /****************************************/ |
| 194 /* Encode Parameters */ |
| 195 /****************************************/ |
| 196 silk_encode_indices( &psEnc->sCmn, psRangeEnc, psEnc->sCmn.nFram
esEncoded, 0, condCoding ); |
| 197 |
| 198 /****************************************/ |
| 199 /* Encode Excitation Signal */ |
| 200 /****************************************/ |
| 201 silk_encode_pulses( psRangeEnc, psEnc->sCmn.indices.signalType,
psEnc->sCmn.indices.quantOffsetType, |
| 202 psEnc->sCmn.pulses, psEnc->sCmn.frame_length ); |
| 203 |
| 204 nBits = ec_tell( psRangeEnc ); |
| 205 |
| 206 if( useCBR == 0 && iter == 0 && nBits <= maxBits ) { |
| 207 break; |
| 208 } |
| 209 } |
| 210 |
| 211 if( iter == maxIter ) { |
| 212 if( found_lower && ( gainsID == gainsID_lower || nBits > maxBits
) ) { |
| 213 /* Restore output state from earlier iteration that did meet
the bitrate budget */ |
| 214 silk_memcpy( psRangeEnc, &sRangeEnc_copy2, sizeof( ec_enc )
); |
| 215 silk_assert( sRangeEnc_copy2.offs <= 1275 ); |
| 216 silk_memcpy( psRangeEnc->buf, ec_buf_copy, sRangeEnc_copy2.o
ffs ); |
| 217 silk_memcpy( &psEnc->sCmn.sNSQ, &sNSQ_copy2, sizeof( silk_ns
q_state ) ); |
| 218 psEnc->sShape.LastGainIndex = LastGainIndex_copy2; |
| 219 } |
| 220 break; |
| 221 } |
| 222 |
| 223 if( nBits > maxBits ) { |
| 224 if( found_lower == 0 && iter >= 2 ) { |
| 225 /* Adjust the quantizer's rate/distortion tradeoff and disca
rd previous "upper" results */ |
| 226 sEncCtrl.Lambda *= 1.5f; |
| 227 found_upper = 0; |
| 228 gainsID_upper = -1; |
| 229 } else { |
| 230 found_upper = 1; |
| 231 nBits_upper = nBits; |
| 232 gainMult_upper = gainMult_Q8; |
| 233 gainsID_upper = gainsID; |
| 234 } |
| 235 } else if( nBits < maxBits - 5 ) { |
| 236 found_lower = 1; |
| 237 nBits_lower = nBits; |
| 238 gainMult_lower = gainMult_Q8; |
| 239 if( gainsID != gainsID_lower ) { |
| 240 gainsID_lower = gainsID; |
| 241 /* Copy part of the output state */ |
| 242 silk_memcpy( &sRangeEnc_copy2, psRangeEnc, sizeof( ec_enc )
); |
| 243 silk_assert( psRangeEnc->offs <= 1275 ); |
| 244 silk_memcpy( ec_buf_copy, psRangeEnc->buf, psRangeEnc->offs
); |
| 245 silk_memcpy( &sNSQ_copy2, &psEnc->sCmn.sNSQ, sizeof( silk_ns
q_state ) ); |
| 246 LastGainIndex_copy2 = psEnc->sShape.LastGainIndex; |
| 247 } |
| 248 } else { |
| 249 /* Within 5 bits of budget: close enough */ |
| 250 break; |
| 251 } |
| 252 |
| 253 if( ( found_lower & found_upper ) == 0 ) { |
| 254 /* Adjust gain according to high-rate rate/distortion curve */ |
| 255 opus_int32 gain_factor_Q16; |
| 256 gain_factor_Q16 = silk_log2lin( silk_LSHIFT( nBits - maxBits, 7
) / psEnc->sCmn.frame_length + SILK_FIX_CONST( 16, 7 ) ); |
| 257 gain_factor_Q16 = silk_min_32( gain_factor_Q16, SILK_FIX_CONST(
2, 16 ) ); |
| 258 if( nBits > maxBits ) { |
| 259 gain_factor_Q16 = silk_max_32( gain_factor_Q16, SILK_FIX_CON
ST( 1.3, 16 ) ); |
| 260 } |
| 261 gainMult_Q8 = silk_SMULWB( gain_factor_Q16, gainMult_Q8 ); |
| 262 } else { |
| 263 /* Adjust gain by interpolating */ |
| 264 gainMult_Q8 = gainMult_lower + ( ( gainMult_upper - gainMult_low
er ) * ( maxBits - nBits_lower ) ) / ( nBits_upper - nBits_lower ); |
| 265 /* New gain multplier must be between 25% and 75% of old range (
note that gainMult_upper < gainMult_lower) */ |
| 266 if( gainMult_Q8 > silk_ADD_RSHIFT32( gainMult_lower, gainMult_up
per - gainMult_lower, 2 ) ) { |
| 267 gainMult_Q8 = silk_ADD_RSHIFT32( gainMult_lower, gainMult_up
per - gainMult_lower, 2 ); |
| 268 } else |
| 269 if( gainMult_Q8 < silk_SUB_RSHIFT32( gainMult_upper, gainMult_up
per - gainMult_lower, 2 ) ) { |
| 270 gainMult_Q8 = silk_SUB_RSHIFT32( gainMult_upper, gainMult_up
per - gainMult_lower, 2 ); |
| 271 } |
| 272 } |
| 273 |
| 274 for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { |
| 275 pGains_Q16[ i ] = silk_LSHIFT_SAT32( silk_SMULWB( sEncCtrl.Gains
Unq_Q16[ i ], gainMult_Q8 ), 8 ); |
| 276 } |
| 277 |
| 278 /* Quantize gains */ |
| 279 psEnc->sShape.LastGainIndex = sEncCtrl.lastGainIndexPrev; |
| 280 silk_gains_quant( psEnc->sCmn.indices.GainsIndices, pGains_Q16, |
| 281 &psEnc->sShape.LastGainIndex, condCoding == CODE_CONDITIONALLY
, psEnc->sCmn.nb_subfr ); |
| 282 |
| 283 /* Unique identifier of gains vector */ |
| 284 gainsID = silk_gains_ID( psEnc->sCmn.indices.GainsIndices, psEnc->sC
mn.nb_subfr ); |
| 285 |
| 286 /* Overwrite unquantized gains with quantized gains and convert back
to Q0 from Q16 */ |
| 287 for( i = 0; i < psEnc->sCmn.nb_subfr; i++ ) { |
| 288 sEncCtrl.Gains[ i ] = pGains_Q16[ i ] / 65536.0f; |
| 289 } |
| 290 } |
| 291 } |
| 292 |
| 293 /* Update input buffer */ |
| 294 silk_memmove( psEnc->x_buf, &psEnc->x_buf[ psEnc->sCmn.frame_length ], |
| 295 ( psEnc->sCmn.ltp_mem_length + LA_SHAPE_MS * psEnc->sCmn.fs_kHz ) * size
of( silk_float ) ); |
| 296 |
| 297 /* Parameters needed for next frame */ |
| 298 psEnc->sCmn.prevLag = sEncCtrl.pitchL[ psEnc->sCmn.nb_subfr - 1 ]; |
| 299 psEnc->sCmn.prevSignalType = psEnc->sCmn.indices.signalType; |
| 300 |
| 301 /* Exit without entropy coding */ |
| 302 if( psEnc->sCmn.prefillFlag ) { |
| 303 /* No payload */ |
| 304 *pnBytesOut = 0; |
| 305 return ret; |
| 306 } |
| 307 |
| 308 /****************************************/ |
| 309 /* Finalize payload */ |
| 310 /****************************************/ |
| 311 psEnc->sCmn.first_frame_after_reset = 0; |
| 312 /* Payload size */ |
| 313 *pnBytesOut = silk_RSHIFT( ec_tell( psRangeEnc ) + 7, 3 ); |
| 314 |
| 315 return ret; |
| 316 } |
| 317 |
| 318 /* Low-Bitrate Redundancy (LBRR) encoding. Reuse all parameters but encode excit
ation at lower bitrate */ |
| 319 static inline void silk_LBRR_encode_FLP( |
| 320 silk_encoder_state_FLP *psEnc, /* I/O
Encoder state FLP */ |
| 321 silk_encoder_control_FLP *psEncCtrl, /* I/O
Encoder control FLP */ |
| 322 const silk_float xfw[], /* I
Input signal */ |
| 323 opus_int condCoding /* I
The type of conditional coding used so far for this frame */ |
| 324 ) |
| 325 { |
| 326 opus_int k; |
| 327 opus_int32 Gains_Q16[ MAX_NB_SUBFR ]; |
| 328 silk_float TempGains[ MAX_NB_SUBFR ]; |
| 329 SideInfoIndices *psIndices_LBRR = &psEnc->sCmn.indices_LBRR[ psEnc->sCmn.nFr
amesEncoded ]; |
| 330 silk_nsq_state sNSQ_LBRR; |
| 331 |
| 332 /*******************************************/ |
| 333 /* Control use of inband LBRR */ |
| 334 /*******************************************/ |
| 335 if( psEnc->sCmn.LBRR_enabled && psEnc->sCmn.speech_activity_Q8 > SILK_FIX_CO
NST( LBRR_SPEECH_ACTIVITY_THRES, 8 ) ) { |
| 336 psEnc->sCmn.LBRR_flags[ psEnc->sCmn.nFramesEncoded ] = 1; |
| 337 |
| 338 /* Copy noise shaping quantizer state and quantization indices from regu
lar encoding */ |
| 339 silk_memcpy( &sNSQ_LBRR, &psEnc->sCmn.sNSQ, sizeof( silk_nsq_state ) ); |
| 340 silk_memcpy( psIndices_LBRR, &psEnc->sCmn.indices, sizeof( SideInfoIndic
es ) ); |
| 341 |
| 342 /* Save original gains */ |
| 343 silk_memcpy( TempGains, psEncCtrl->Gains, psEnc->sCmn.nb_subfr * sizeof(
silk_float ) ); |
| 344 |
| 345 if( psEnc->sCmn.nFramesEncoded == 0 || psEnc->sCmn.LBRR_flags[ psEnc->sC
mn.nFramesEncoded - 1 ] == 0 ) { |
| 346 /* First frame in packet or previous frame not LBRR coded */ |
| 347 psEnc->sCmn.LBRRprevLastGainIndex = psEnc->sShape.LastGainIndex; |
| 348 |
| 349 /* Increase Gains to get target LBRR rate */ |
| 350 psIndices_LBRR->GainsIndices[ 0 ] += psEnc->sCmn.LBRR_GainIncreases; |
| 351 psIndices_LBRR->GainsIndices[ 0 ] = silk_min_int( psIndices_LBRR->Ga
insIndices[ 0 ], N_LEVELS_QGAIN - 1 ); |
| 352 } |
| 353 |
| 354 /* Decode to get gains in sync with decoder */ |
| 355 silk_gains_dequant( Gains_Q16, psIndices_LBRR->GainsIndices, |
| 356 &psEnc->sCmn.LBRRprevLastGainIndex, condCoding == CODE_CONDITIONALLY
, psEnc->sCmn.nb_subfr ); |
| 357 |
| 358 /* Overwrite unquantized gains with quantized gains and convert back to
Q0 from Q16 */ |
| 359 for( k = 0; k < psEnc->sCmn.nb_subfr; k++ ) { |
| 360 psEncCtrl->Gains[ k ] = Gains_Q16[ k ] * ( 1.0f / 65536.0f ); |
| 361 } |
| 362 |
| 363 /*****************************************/ |
| 364 /* Noise shaping quantization */ |
| 365 /*****************************************/ |
| 366 silk_NSQ_wrapper_FLP( psEnc, psEncCtrl, psIndices_LBRR, &sNSQ_LBRR, |
| 367 psEnc->sCmn.pulses_LBRR[ psEnc->sCmn.nFramesEncoded ], xfw ); |
| 368 |
| 369 /* Restore original gains */ |
| 370 silk_memcpy( psEncCtrl->Gains, TempGains, psEnc->sCmn.nb_subfr * sizeof(
silk_float ) ); |
| 371 } |
| 372 } |
| OLD | NEW |