media/libopus/src/opus_encoder.c

Wed, 31 Dec 2014 06:09:35 +0100

author
Michael Schloh von Bennewitz <michael@schloh.com>
date
Wed, 31 Dec 2014 06:09:35 +0100
changeset 0
6474c204b198
permissions
-rw-r--r--

Cloned upstream origin tor-browser at tor-browser-31.3.0esr-4.5-1-build1
revision ID fc1c9ff7c1b2defdbc039f12214767608f46423f for hacking purpose.

michael@0 1 /* Copyright (c) 2010-2011 Xiph.Org Foundation, Skype Limited
michael@0 2 Written by Jean-Marc Valin and Koen Vos */
michael@0 3 /*
michael@0 4 Redistribution and use in source and binary forms, with or without
michael@0 5 modification, are permitted provided that the following conditions
michael@0 6 are met:
michael@0 7
michael@0 8 - Redistributions of source code must retain the above copyright
michael@0 9 notice, this list of conditions and the following disclaimer.
michael@0 10
michael@0 11 - Redistributions in binary form must reproduce the above copyright
michael@0 12 notice, this list of conditions and the following disclaimer in the
michael@0 13 documentation and/or other materials provided with the distribution.
michael@0 14
michael@0 15 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
michael@0 16 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
michael@0 17 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
michael@0 18 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER
michael@0 19 OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
michael@0 20 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
michael@0 21 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
michael@0 22 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
michael@0 23 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
michael@0 24 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
michael@0 25 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
michael@0 26 */
michael@0 27
michael@0 28 #ifdef HAVE_CONFIG_H
michael@0 29 #include "config.h"
michael@0 30 #endif
michael@0 31
michael@0 32 #include <stdarg.h>
michael@0 33 #include "celt.h"
michael@0 34 #include "entenc.h"
michael@0 35 #include "modes.h"
michael@0 36 #include "API.h"
michael@0 37 #include "stack_alloc.h"
michael@0 38 #include "float_cast.h"
michael@0 39 #include "opus.h"
michael@0 40 #include "arch.h"
michael@0 41 #include "opus_private.h"
michael@0 42 #include "os_support.h"
michael@0 43 #include "cpu_support.h"
michael@0 44 #include "analysis.h"
michael@0 45 #include "mathops.h"
michael@0 46 #include "tuning_parameters.h"
michael@0 47 #ifdef FIXED_POINT
michael@0 48 #include "fixed/structs_FIX.h"
michael@0 49 #else
michael@0 50 #include "float/structs_FLP.h"
michael@0 51 #endif
michael@0 52
michael@0 53 #define MAX_ENCODER_BUFFER 480
michael@0 54
michael@0 55 typedef struct {
michael@0 56 opus_val32 XX, XY, YY;
michael@0 57 opus_val16 smoothed_width;
michael@0 58 opus_val16 max_follower;
michael@0 59 } StereoWidthState;
michael@0 60
michael@0 61 struct OpusEncoder {
michael@0 62 int celt_enc_offset;
michael@0 63 int silk_enc_offset;
michael@0 64 silk_EncControlStruct silk_mode;
michael@0 65 int application;
michael@0 66 int channels;
michael@0 67 int delay_compensation;
michael@0 68 int force_channels;
michael@0 69 int signal_type;
michael@0 70 int user_bandwidth;
michael@0 71 int max_bandwidth;
michael@0 72 int user_forced_mode;
michael@0 73 int voice_ratio;
michael@0 74 opus_int32 Fs;
michael@0 75 int use_vbr;
michael@0 76 int vbr_constraint;
michael@0 77 int variable_duration;
michael@0 78 opus_int32 bitrate_bps;
michael@0 79 opus_int32 user_bitrate_bps;
michael@0 80 int lsb_depth;
michael@0 81 int encoder_buffer;
michael@0 82 int lfe;
michael@0 83
michael@0 84 #define OPUS_ENCODER_RESET_START stream_channels
michael@0 85 int stream_channels;
michael@0 86 opus_int16 hybrid_stereo_width_Q14;
michael@0 87 opus_int32 variable_HP_smth2_Q15;
michael@0 88 opus_val16 prev_HB_gain;
michael@0 89 opus_val32 hp_mem[4];
michael@0 90 int mode;
michael@0 91 int prev_mode;
michael@0 92 int prev_channels;
michael@0 93 int prev_framesize;
michael@0 94 int bandwidth;
michael@0 95 int silk_bw_switch;
michael@0 96 /* Sampling rate (at the API level) */
michael@0 97 int first;
michael@0 98 opus_val16 * energy_masking;
michael@0 99 StereoWidthState width_mem;
michael@0 100 opus_val16 delay_buffer[MAX_ENCODER_BUFFER*2];
michael@0 101 #ifndef DISABLE_FLOAT_API
michael@0 102 TonalityAnalysisState analysis;
michael@0 103 int detected_bandwidth;
michael@0 104 int analysis_offset;
michael@0 105 #endif
michael@0 106 opus_uint32 rangeFinal;
michael@0 107 int arch;
michael@0 108 };
michael@0 109
michael@0 110 /* Transition tables for the voice and music. First column is the
michael@0 111 middle (memoriless) threshold. The second column is the hysteresis
michael@0 112 (difference with the middle) */
michael@0 113 static const opus_int32 mono_voice_bandwidth_thresholds[8] = {
michael@0 114 11000, 1000, /* NB<->MB */
michael@0 115 14000, 1000, /* MB<->WB */
michael@0 116 17000, 1000, /* WB<->SWB */
michael@0 117 21000, 2000, /* SWB<->FB */
michael@0 118 };
michael@0 119 static const opus_int32 mono_music_bandwidth_thresholds[8] = {
michael@0 120 12000, 1000, /* NB<->MB */
michael@0 121 15000, 1000, /* MB<->WB */
michael@0 122 18000, 2000, /* WB<->SWB */
michael@0 123 22000, 2000, /* SWB<->FB */
michael@0 124 };
michael@0 125 static const opus_int32 stereo_voice_bandwidth_thresholds[8] = {
michael@0 126 11000, 1000, /* NB<->MB */
michael@0 127 14000, 1000, /* MB<->WB */
michael@0 128 21000, 2000, /* WB<->SWB */
michael@0 129 28000, 2000, /* SWB<->FB */
michael@0 130 };
michael@0 131 static const opus_int32 stereo_music_bandwidth_thresholds[8] = {
michael@0 132 12000, 1000, /* NB<->MB */
michael@0 133 18000, 2000, /* MB<->WB */
michael@0 134 21000, 2000, /* WB<->SWB */
michael@0 135 30000, 2000, /* SWB<->FB */
michael@0 136 };
michael@0 137 /* Threshold bit-rates for switching between mono and stereo */
michael@0 138 static const opus_int32 stereo_voice_threshold = 30000;
michael@0 139 static const opus_int32 stereo_music_threshold = 30000;
michael@0 140
michael@0 141 /* Threshold bit-rate for switching between SILK/hybrid and CELT-only */
michael@0 142 static const opus_int32 mode_thresholds[2][2] = {
michael@0 143 /* voice */ /* music */
michael@0 144 { 64000, 16000}, /* mono */
michael@0 145 { 36000, 16000}, /* stereo */
michael@0 146 };
michael@0 147
michael@0 148 int opus_encoder_get_size(int channels)
michael@0 149 {
michael@0 150 int silkEncSizeBytes, celtEncSizeBytes;
michael@0 151 int ret;
michael@0 152 if (channels<1 || channels > 2)
michael@0 153 return 0;
michael@0 154 ret = silk_Get_Encoder_Size( &silkEncSizeBytes );
michael@0 155 if (ret)
michael@0 156 return 0;
michael@0 157 silkEncSizeBytes = align(silkEncSizeBytes);
michael@0 158 celtEncSizeBytes = celt_encoder_get_size(channels);
michael@0 159 return align(sizeof(OpusEncoder))+silkEncSizeBytes+celtEncSizeBytes;
michael@0 160 }
michael@0 161
michael@0 162 int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int application)
michael@0 163 {
michael@0 164 void *silk_enc;
michael@0 165 CELTEncoder *celt_enc;
michael@0 166 int err;
michael@0 167 int ret, silkEncSizeBytes;
michael@0 168
michael@0 169 if((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000)||(channels!=1&&channels!=2)||
michael@0 170 (application != OPUS_APPLICATION_VOIP && application != OPUS_APPLICATION_AUDIO
michael@0 171 && application != OPUS_APPLICATION_RESTRICTED_LOWDELAY))
michael@0 172 return OPUS_BAD_ARG;
michael@0 173
michael@0 174 OPUS_CLEAR((char*)st, opus_encoder_get_size(channels));
michael@0 175 /* Create SILK encoder */
michael@0 176 ret = silk_Get_Encoder_Size( &silkEncSizeBytes );
michael@0 177 if (ret)
michael@0 178 return OPUS_BAD_ARG;
michael@0 179 silkEncSizeBytes = align(silkEncSizeBytes);
michael@0 180 st->silk_enc_offset = align(sizeof(OpusEncoder));
michael@0 181 st->celt_enc_offset = st->silk_enc_offset+silkEncSizeBytes;
michael@0 182 silk_enc = (char*)st+st->silk_enc_offset;
michael@0 183 celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset);
michael@0 184
michael@0 185 st->stream_channels = st->channels = channels;
michael@0 186
michael@0 187 st->Fs = Fs;
michael@0 188
michael@0 189 st->arch = opus_select_arch();
michael@0 190
michael@0 191 ret = silk_InitEncoder( silk_enc, st->arch, &st->silk_mode );
michael@0 192 if(ret)return OPUS_INTERNAL_ERROR;
michael@0 193
michael@0 194 /* default SILK parameters */
michael@0 195 st->silk_mode.nChannelsAPI = channels;
michael@0 196 st->silk_mode.nChannelsInternal = channels;
michael@0 197 st->silk_mode.API_sampleRate = st->Fs;
michael@0 198 st->silk_mode.maxInternalSampleRate = 16000;
michael@0 199 st->silk_mode.minInternalSampleRate = 8000;
michael@0 200 st->silk_mode.desiredInternalSampleRate = 16000;
michael@0 201 st->silk_mode.payloadSize_ms = 20;
michael@0 202 st->silk_mode.bitRate = 25000;
michael@0 203 st->silk_mode.packetLossPercentage = 0;
michael@0 204 st->silk_mode.complexity = 9;
michael@0 205 st->silk_mode.useInBandFEC = 0;
michael@0 206 st->silk_mode.useDTX = 0;
michael@0 207 st->silk_mode.useCBR = 0;
michael@0 208 st->silk_mode.reducedDependency = 0;
michael@0 209
michael@0 210 /* Create CELT encoder */
michael@0 211 /* Initialize CELT encoder */
michael@0 212 err = celt_encoder_init(celt_enc, Fs, channels, st->arch);
michael@0 213 if(err!=OPUS_OK)return OPUS_INTERNAL_ERROR;
michael@0 214
michael@0 215 celt_encoder_ctl(celt_enc, CELT_SET_SIGNALLING(0));
michael@0 216 celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(st->silk_mode.complexity));
michael@0 217
michael@0 218 st->use_vbr = 1;
michael@0 219 /* Makes constrained VBR the default (safer for real-time use) */
michael@0 220 st->vbr_constraint = 1;
michael@0 221 st->user_bitrate_bps = OPUS_AUTO;
michael@0 222 st->bitrate_bps = 3000+Fs*channels;
michael@0 223 st->application = application;
michael@0 224 st->signal_type = OPUS_AUTO;
michael@0 225 st->user_bandwidth = OPUS_AUTO;
michael@0 226 st->max_bandwidth = OPUS_BANDWIDTH_FULLBAND;
michael@0 227 st->force_channels = OPUS_AUTO;
michael@0 228 st->user_forced_mode = OPUS_AUTO;
michael@0 229 st->voice_ratio = -1;
michael@0 230 st->encoder_buffer = st->Fs/100;
michael@0 231 st->lsb_depth = 24;
michael@0 232 st->variable_duration = OPUS_FRAMESIZE_ARG;
michael@0 233
michael@0 234 /* Delay compensation of 4 ms (2.5 ms for SILK's extra look-ahead
michael@0 235 + 1.5 ms for SILK resamplers and stereo prediction) */
michael@0 236 st->delay_compensation = st->Fs/250;
michael@0 237
michael@0 238 st->hybrid_stereo_width_Q14 = 1 << 14;
michael@0 239 st->prev_HB_gain = Q15ONE;
michael@0 240 st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 );
michael@0 241 st->first = 1;
michael@0 242 st->mode = MODE_HYBRID;
michael@0 243 st->bandwidth = OPUS_BANDWIDTH_FULLBAND;
michael@0 244
michael@0 245 return OPUS_OK;
michael@0 246 }
michael@0 247
michael@0 248 static unsigned char gen_toc(int mode, int framerate, int bandwidth, int channels)
michael@0 249 {
michael@0 250 int period;
michael@0 251 unsigned char toc;
michael@0 252 period = 0;
michael@0 253 while (framerate < 400)
michael@0 254 {
michael@0 255 framerate <<= 1;
michael@0 256 period++;
michael@0 257 }
michael@0 258 if (mode == MODE_SILK_ONLY)
michael@0 259 {
michael@0 260 toc = (bandwidth-OPUS_BANDWIDTH_NARROWBAND)<<5;
michael@0 261 toc |= (period-2)<<3;
michael@0 262 } else if (mode == MODE_CELT_ONLY)
michael@0 263 {
michael@0 264 int tmp = bandwidth-OPUS_BANDWIDTH_MEDIUMBAND;
michael@0 265 if (tmp < 0)
michael@0 266 tmp = 0;
michael@0 267 toc = 0x80;
michael@0 268 toc |= tmp << 5;
michael@0 269 toc |= period<<3;
michael@0 270 } else /* Hybrid */
michael@0 271 {
michael@0 272 toc = 0x60;
michael@0 273 toc |= (bandwidth-OPUS_BANDWIDTH_SUPERWIDEBAND)<<4;
michael@0 274 toc |= (period-2)<<3;
michael@0 275 }
michael@0 276 toc |= (channels==2)<<2;
michael@0 277 return toc;
michael@0 278 }
michael@0 279
michael@0 280 #ifndef FIXED_POINT
michael@0 281 static void silk_biquad_float(
michael@0 282 const opus_val16 *in, /* I: Input signal */
michael@0 283 const opus_int32 *B_Q28, /* I: MA coefficients [3] */
michael@0 284 const opus_int32 *A_Q28, /* I: AR coefficients [2] */
michael@0 285 opus_val32 *S, /* I/O: State vector [2] */
michael@0 286 opus_val16 *out, /* O: Output signal */
michael@0 287 const opus_int32 len, /* I: Signal length (must be even) */
michael@0 288 int stride
michael@0 289 )
michael@0 290 {
michael@0 291 /* DIRECT FORM II TRANSPOSED (uses 2 element state vector) */
michael@0 292 opus_int k;
michael@0 293 opus_val32 vout;
michael@0 294 opus_val32 inval;
michael@0 295 opus_val32 A[2], B[3];
michael@0 296
michael@0 297 A[0] = (opus_val32)(A_Q28[0] * (1.f/((opus_int32)1<<28)));
michael@0 298 A[1] = (opus_val32)(A_Q28[1] * (1.f/((opus_int32)1<<28)));
michael@0 299 B[0] = (opus_val32)(B_Q28[0] * (1.f/((opus_int32)1<<28)));
michael@0 300 B[1] = (opus_val32)(B_Q28[1] * (1.f/((opus_int32)1<<28)));
michael@0 301 B[2] = (opus_val32)(B_Q28[2] * (1.f/((opus_int32)1<<28)));
michael@0 302
michael@0 303 /* Negate A_Q28 values and split in two parts */
michael@0 304
michael@0 305 for( k = 0; k < len; k++ ) {
michael@0 306 /* S[ 0 ], S[ 1 ]: Q12 */
michael@0 307 inval = in[ k*stride ];
michael@0 308 vout = S[ 0 ] + B[0]*inval;
michael@0 309
michael@0 310 S[ 0 ] = S[1] - vout*A[0] + B[1]*inval;
michael@0 311
michael@0 312 S[ 1 ] = - vout*A[1] + B[2]*inval + VERY_SMALL;
michael@0 313
michael@0 314 /* Scale back to Q0 and saturate */
michael@0 315 out[ k*stride ] = vout;
michael@0 316 }
michael@0 317 }
michael@0 318 #endif
michael@0 319
michael@0 320 static void hp_cutoff(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs)
michael@0 321 {
michael@0 322 opus_int32 B_Q28[ 3 ], A_Q28[ 2 ];
michael@0 323 opus_int32 Fc_Q19, r_Q28, r_Q22;
michael@0 324
michael@0 325 silk_assert( cutoff_Hz <= silk_int32_MAX / SILK_FIX_CONST( 1.5 * 3.14159 / 1000, 19 ) );
michael@0 326 Fc_Q19 = silk_DIV32_16( silk_SMULBB( SILK_FIX_CONST( 1.5 * 3.14159 / 1000, 19 ), cutoff_Hz ), Fs/1000 );
michael@0 327 silk_assert( Fc_Q19 > 0 && Fc_Q19 < 32768 );
michael@0 328
michael@0 329 r_Q28 = SILK_FIX_CONST( 1.0, 28 ) - silk_MUL( SILK_FIX_CONST( 0.92, 9 ), Fc_Q19 );
michael@0 330
michael@0 331 /* b = r * [ 1; -2; 1 ]; */
michael@0 332 /* a = [ 1; -2 * r * ( 1 - 0.5 * Fc^2 ); r^2 ]; */
michael@0 333 B_Q28[ 0 ] = r_Q28;
michael@0 334 B_Q28[ 1 ] = silk_LSHIFT( -r_Q28, 1 );
michael@0 335 B_Q28[ 2 ] = r_Q28;
michael@0 336
michael@0 337 /* -r * ( 2 - Fc * Fc ); */
michael@0 338 r_Q22 = silk_RSHIFT( r_Q28, 6 );
michael@0 339 A_Q28[ 0 ] = silk_SMULWW( r_Q22, silk_SMULWW( Fc_Q19, Fc_Q19 ) - SILK_FIX_CONST( 2.0, 22 ) );
michael@0 340 A_Q28[ 1 ] = silk_SMULWW( r_Q22, r_Q22 );
michael@0 341
michael@0 342 #ifdef FIXED_POINT
michael@0 343 silk_biquad_alt( in, B_Q28, A_Q28, hp_mem, out, len, channels );
michael@0 344 if( channels == 2 ) {
michael@0 345 silk_biquad_alt( in+1, B_Q28, A_Q28, hp_mem+2, out+1, len, channels );
michael@0 346 }
michael@0 347 #else
michael@0 348 silk_biquad_float( in, B_Q28, A_Q28, hp_mem, out, len, channels );
michael@0 349 if( channels == 2 ) {
michael@0 350 silk_biquad_float( in+1, B_Q28, A_Q28, hp_mem+2, out+1, len, channels );
michael@0 351 }
michael@0 352 #endif
michael@0 353 }
michael@0 354
michael@0 355 #ifdef FIXED_POINT
michael@0 356 static void dc_reject(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs)
michael@0 357 {
michael@0 358 int c, i;
michael@0 359 int shift;
michael@0 360
michael@0 361 /* Approximates -round(log2(4.*cutoff_Hz/Fs)) */
michael@0 362 shift=celt_ilog2(Fs/(cutoff_Hz*3));
michael@0 363 for (c=0;c<channels;c++)
michael@0 364 {
michael@0 365 for (i=0;i<len;i++)
michael@0 366 {
michael@0 367 opus_val32 x, tmp, y;
michael@0 368 x = SHL32(EXTEND32(in[channels*i+c]), 15);
michael@0 369 /* First stage */
michael@0 370 tmp = x-hp_mem[2*c];
michael@0 371 hp_mem[2*c] = hp_mem[2*c] + PSHR32(x - hp_mem[2*c], shift);
michael@0 372 /* Second stage */
michael@0 373 y = tmp - hp_mem[2*c+1];
michael@0 374 hp_mem[2*c+1] = hp_mem[2*c+1] + PSHR32(tmp - hp_mem[2*c+1], shift);
michael@0 375 out[channels*i+c] = EXTRACT16(SATURATE(PSHR32(y, 15), 32767));
michael@0 376 }
michael@0 377 }
michael@0 378 }
michael@0 379
michael@0 380 #else
michael@0 381 static void dc_reject(const opus_val16 *in, opus_int32 cutoff_Hz, opus_val16 *out, opus_val32 *hp_mem, int len, int channels, opus_int32 Fs)
michael@0 382 {
michael@0 383 int c, i;
michael@0 384 float coef;
michael@0 385
michael@0 386 coef = 4.0f*cutoff_Hz/Fs;
michael@0 387 for (c=0;c<channels;c++)
michael@0 388 {
michael@0 389 for (i=0;i<len;i++)
michael@0 390 {
michael@0 391 opus_val32 x, tmp, y;
michael@0 392 x = in[channels*i+c];
michael@0 393 /* First stage */
michael@0 394 tmp = x-hp_mem[2*c];
michael@0 395 hp_mem[2*c] = hp_mem[2*c] + coef*(x - hp_mem[2*c]) + VERY_SMALL;
michael@0 396 /* Second stage */
michael@0 397 y = tmp - hp_mem[2*c+1];
michael@0 398 hp_mem[2*c+1] = hp_mem[2*c+1] + coef*(tmp - hp_mem[2*c+1]) + VERY_SMALL;
michael@0 399 out[channels*i+c] = y;
michael@0 400 }
michael@0 401 }
michael@0 402 }
michael@0 403 #endif
michael@0 404
michael@0 405 static void stereo_fade(const opus_val16 *in, opus_val16 *out, opus_val16 g1, opus_val16 g2,
michael@0 406 int overlap48, int frame_size, int channels, const opus_val16 *window, opus_int32 Fs)
michael@0 407 {
michael@0 408 int i;
michael@0 409 int overlap;
michael@0 410 int inc;
michael@0 411 inc = 48000/Fs;
michael@0 412 overlap=overlap48/inc;
michael@0 413 g1 = Q15ONE-g1;
michael@0 414 g2 = Q15ONE-g2;
michael@0 415 for (i=0;i<overlap;i++)
michael@0 416 {
michael@0 417 opus_val32 diff;
michael@0 418 opus_val16 g, w;
michael@0 419 w = MULT16_16_Q15(window[i*inc], window[i*inc]);
michael@0 420 g = SHR32(MAC16_16(MULT16_16(w,g2),
michael@0 421 Q15ONE-w, g1), 15);
michael@0 422 diff = EXTRACT16(HALF32((opus_val32)in[i*channels] - (opus_val32)in[i*channels+1]));
michael@0 423 diff = MULT16_16_Q15(g, diff);
michael@0 424 out[i*channels] = out[i*channels] - diff;
michael@0 425 out[i*channels+1] = out[i*channels+1] + diff;
michael@0 426 }
michael@0 427 for (;i<frame_size;i++)
michael@0 428 {
michael@0 429 opus_val32 diff;
michael@0 430 diff = EXTRACT16(HALF32((opus_val32)in[i*channels] - (opus_val32)in[i*channels+1]));
michael@0 431 diff = MULT16_16_Q15(g2, diff);
michael@0 432 out[i*channels] = out[i*channels] - diff;
michael@0 433 out[i*channels+1] = out[i*channels+1] + diff;
michael@0 434 }
michael@0 435 }
michael@0 436
michael@0 437 static void gain_fade(const opus_val16 *in, opus_val16 *out, opus_val16 g1, opus_val16 g2,
michael@0 438 int overlap48, int frame_size, int channels, const opus_val16 *window, opus_int32 Fs)
michael@0 439 {
michael@0 440 int i;
michael@0 441 int inc;
michael@0 442 int overlap;
michael@0 443 int c;
michael@0 444 inc = 48000/Fs;
michael@0 445 overlap=overlap48/inc;
michael@0 446 if (channels==1)
michael@0 447 {
michael@0 448 for (i=0;i<overlap;i++)
michael@0 449 {
michael@0 450 opus_val16 g, w;
michael@0 451 w = MULT16_16_Q15(window[i*inc], window[i*inc]);
michael@0 452 g = SHR32(MAC16_16(MULT16_16(w,g2),
michael@0 453 Q15ONE-w, g1), 15);
michael@0 454 out[i] = MULT16_16_Q15(g, in[i]);
michael@0 455 }
michael@0 456 } else {
michael@0 457 for (i=0;i<overlap;i++)
michael@0 458 {
michael@0 459 opus_val16 g, w;
michael@0 460 w = MULT16_16_Q15(window[i*inc], window[i*inc]);
michael@0 461 g = SHR32(MAC16_16(MULT16_16(w,g2),
michael@0 462 Q15ONE-w, g1), 15);
michael@0 463 out[i*2] = MULT16_16_Q15(g, in[i*2]);
michael@0 464 out[i*2+1] = MULT16_16_Q15(g, in[i*2+1]);
michael@0 465 }
michael@0 466 }
michael@0 467 c=0;do {
michael@0 468 for (i=overlap;i<frame_size;i++)
michael@0 469 {
michael@0 470 out[i*channels+c] = MULT16_16_Q15(g2, in[i*channels+c]);
michael@0 471 }
michael@0 472 }
michael@0 473 while (++c<channels);
michael@0 474 }
michael@0 475
michael@0 476 OpusEncoder *opus_encoder_create(opus_int32 Fs, int channels, int application, int *error)
michael@0 477 {
michael@0 478 int ret;
michael@0 479 OpusEncoder *st;
michael@0 480 if((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000)||(channels!=1&&channels!=2)||
michael@0 481 (application != OPUS_APPLICATION_VOIP && application != OPUS_APPLICATION_AUDIO
michael@0 482 && application != OPUS_APPLICATION_RESTRICTED_LOWDELAY))
michael@0 483 {
michael@0 484 if (error)
michael@0 485 *error = OPUS_BAD_ARG;
michael@0 486 return NULL;
michael@0 487 }
michael@0 488 st = (OpusEncoder *)opus_alloc(opus_encoder_get_size(channels));
michael@0 489 if (st == NULL)
michael@0 490 {
michael@0 491 if (error)
michael@0 492 *error = OPUS_ALLOC_FAIL;
michael@0 493 return NULL;
michael@0 494 }
michael@0 495 ret = opus_encoder_init(st, Fs, channels, application);
michael@0 496 if (error)
michael@0 497 *error = ret;
michael@0 498 if (ret != OPUS_OK)
michael@0 499 {
michael@0 500 opus_free(st);
michael@0 501 st = NULL;
michael@0 502 }
michael@0 503 return st;
michael@0 504 }
michael@0 505
michael@0 506 static opus_int32 user_bitrate_to_bitrate(OpusEncoder *st, int frame_size, int max_data_bytes)
michael@0 507 {
michael@0 508 if(!frame_size)frame_size=st->Fs/400;
michael@0 509 if (st->user_bitrate_bps==OPUS_AUTO)
michael@0 510 return 60*st->Fs/frame_size + st->Fs*st->channels;
michael@0 511 else if (st->user_bitrate_bps==OPUS_BITRATE_MAX)
michael@0 512 return max_data_bytes*8*st->Fs/frame_size;
michael@0 513 else
michael@0 514 return st->user_bitrate_bps;
michael@0 515 }
michael@0 516
michael@0 517 #ifndef DISABLE_FLOAT_API
michael@0 518 /* Don't use more than 60 ms for the frame size analysis */
michael@0 519 #define MAX_DYNAMIC_FRAMESIZE 24
michael@0 520 /* Estimates how much the bitrate will be boosted based on the sub-frame energy */
michael@0 521 static float transient_boost(const float *E, const float *E_1, int LM, int maxM)
michael@0 522 {
michael@0 523 int i;
michael@0 524 int M;
michael@0 525 float sumE=0, sumE_1=0;
michael@0 526 float metric;
michael@0 527
michael@0 528 M = IMIN(maxM, (1<<LM)+1);
michael@0 529 for (i=0;i<M;i++)
michael@0 530 {
michael@0 531 sumE += E[i];
michael@0 532 sumE_1 += E_1[i];
michael@0 533 }
michael@0 534 metric = sumE*sumE_1/(M*M);
michael@0 535 /*if (LM==3)
michael@0 536 printf("%f\n", metric);*/
michael@0 537 /*return metric>10 ? 1 : 0;*/
michael@0 538 /*return MAX16(0,1-exp(-.25*(metric-2.)));*/
michael@0 539 return MIN16(1,(float)sqrt(MAX16(0,.05f*(metric-2))));
michael@0 540 }
michael@0 541
michael@0 542 /* Viterbi decoding trying to find the best frame size combination using look-ahead
michael@0 543
michael@0 544 State numbering:
michael@0 545 0: unused
michael@0 546 1: 2.5 ms
michael@0 547 2: 5 ms (#1)
michael@0 548 3: 5 ms (#2)
michael@0 549 4: 10 ms (#1)
michael@0 550 5: 10 ms (#2)
michael@0 551 6: 10 ms (#3)
michael@0 552 7: 10 ms (#4)
michael@0 553 8: 20 ms (#1)
michael@0 554 9: 20 ms (#2)
michael@0 555 10: 20 ms (#3)
michael@0 556 11: 20 ms (#4)
michael@0 557 12: 20 ms (#5)
michael@0 558 13: 20 ms (#6)
michael@0 559 14: 20 ms (#7)
michael@0 560 15: 20 ms (#8)
michael@0 561 */
michael@0 562 static int transient_viterbi(const float *E, const float *E_1, int N, int frame_cost, int rate)
michael@0 563 {
michael@0 564 int i;
michael@0 565 float cost[MAX_DYNAMIC_FRAMESIZE][16];
michael@0 566 int states[MAX_DYNAMIC_FRAMESIZE][16];
michael@0 567 float best_cost;
michael@0 568 int best_state;
michael@0 569 float factor;
michael@0 570 /* Take into account that we damp VBR in the 32 kb/s to 64 kb/s range. */
michael@0 571 if (rate<80)
michael@0 572 factor=0;
michael@0 573 else if (rate>160)
michael@0 574 factor=1;
michael@0 575 else
michael@0 576 factor = (rate-80.f)/80.f;
michael@0 577 /* Makes variable framesize less aggressive at lower bitrates, but I can't
michael@0 578 find any valid theoretical justification for this (other than it seems
michael@0 579 to help) */
michael@0 580 for (i=0;i<16;i++)
michael@0 581 {
michael@0 582 /* Impossible state */
michael@0 583 states[0][i] = -1;
michael@0 584 cost[0][i] = 1e10;
michael@0 585 }
michael@0 586 for (i=0;i<4;i++)
michael@0 587 {
michael@0 588 cost[0][1<<i] = (frame_cost + rate*(1<<i))*(1+factor*transient_boost(E, E_1, i, N+1));
michael@0 589 states[0][1<<i] = i;
michael@0 590 }
michael@0 591 for (i=1;i<N;i++)
michael@0 592 {
michael@0 593 int j;
michael@0 594
michael@0 595 /* Follow continuations */
michael@0 596 for (j=2;j<16;j++)
michael@0 597 {
michael@0 598 cost[i][j] = cost[i-1][j-1];
michael@0 599 states[i][j] = j-1;
michael@0 600 }
michael@0 601
michael@0 602 /* New frames */
michael@0 603 for(j=0;j<4;j++)
michael@0 604 {
michael@0 605 int k;
michael@0 606 float min_cost;
michael@0 607 float curr_cost;
michael@0 608 states[i][1<<j] = 1;
michael@0 609 min_cost = cost[i-1][1];
michael@0 610 for(k=1;k<4;k++)
michael@0 611 {
michael@0 612 float tmp = cost[i-1][(1<<(k+1))-1];
michael@0 613 if (tmp < min_cost)
michael@0 614 {
michael@0 615 states[i][1<<j] = (1<<(k+1))-1;
michael@0 616 min_cost = tmp;
michael@0 617 }
michael@0 618 }
michael@0 619 curr_cost = (frame_cost + rate*(1<<j))*(1+factor*transient_boost(E+i, E_1+i, j, N-i+1));
michael@0 620 cost[i][1<<j] = min_cost;
michael@0 621 /* If part of the frame is outside the analysis window, only count part of the cost */
michael@0 622 if (N-i < (1<<j))
michael@0 623 cost[i][1<<j] += curr_cost*(float)(N-i)/(1<<j);
michael@0 624 else
michael@0 625 cost[i][1<<j] += curr_cost;
michael@0 626 }
michael@0 627 }
michael@0 628
michael@0 629 best_state=1;
michael@0 630 best_cost = cost[N-1][1];
michael@0 631 /* Find best end state (doesn't force a frame to end at N-1) */
michael@0 632 for (i=2;i<16;i++)
michael@0 633 {
michael@0 634 if (cost[N-1][i]<best_cost)
michael@0 635 {
michael@0 636 best_cost = cost[N-1][i];
michael@0 637 best_state = i;
michael@0 638 }
michael@0 639 }
michael@0 640
michael@0 641 /* Follow transitions back */
michael@0 642 for (i=N-1;i>=0;i--)
michael@0 643 {
michael@0 644 /*printf("%d ", best_state);*/
michael@0 645 best_state = states[i][best_state];
michael@0 646 }
michael@0 647 /*printf("%d\n", best_state);*/
michael@0 648 return best_state;
michael@0 649 }
michael@0 650
michael@0 651 int optimize_framesize(const opus_val16 *x, int len, int C, opus_int32 Fs,
michael@0 652 int bitrate, opus_val16 tonality, float *mem, int buffering,
michael@0 653 downmix_func downmix)
michael@0 654 {
michael@0 655 int N;
michael@0 656 int i;
michael@0 657 float e[MAX_DYNAMIC_FRAMESIZE+4];
michael@0 658 float e_1[MAX_DYNAMIC_FRAMESIZE+3];
michael@0 659 opus_val32 memx;
michael@0 660 int bestLM=0;
michael@0 661 int subframe;
michael@0 662 int pos;
michael@0 663 VARDECL(opus_val32, sub);
michael@0 664
michael@0 665 subframe = Fs/400;
michael@0 666 ALLOC(sub, subframe, opus_val32);
michael@0 667 e[0]=mem[0];
michael@0 668 e_1[0]=1.f/(EPSILON+mem[0]);
michael@0 669 if (buffering)
michael@0 670 {
michael@0 671 /* Consider the CELT delay when not in restricted-lowdelay */
michael@0 672 /* We assume the buffering is between 2.5 and 5 ms */
michael@0 673 int offset = 2*subframe - buffering;
michael@0 674 celt_assert(offset>=0 && offset <= subframe);
michael@0 675 x += C*offset;
michael@0 676 len -= offset;
michael@0 677 e[1]=mem[1];
michael@0 678 e_1[1]=1.f/(EPSILON+mem[1]);
michael@0 679 e[2]=mem[2];
michael@0 680 e_1[2]=1.f/(EPSILON+mem[2]);
michael@0 681 pos = 3;
michael@0 682 } else {
michael@0 683 pos=1;
michael@0 684 }
michael@0 685 N=IMIN(len/subframe, MAX_DYNAMIC_FRAMESIZE);
michael@0 686 /* Just silencing a warning, it's really initialized later */
michael@0 687 memx = 0;
michael@0 688 for (i=0;i<N;i++)
michael@0 689 {
michael@0 690 float tmp;
michael@0 691 opus_val32 tmpx;
michael@0 692 int j;
michael@0 693 tmp=EPSILON;
michael@0 694
michael@0 695 downmix(x, sub, subframe, i*subframe, 0, -2, C);
michael@0 696 if (i==0)
michael@0 697 memx = sub[0];
michael@0 698 for (j=0;j<subframe;j++)
michael@0 699 {
michael@0 700 tmpx = sub[j];
michael@0 701 tmp += (tmpx-memx)*(float)(tmpx-memx);
michael@0 702 memx = tmpx;
michael@0 703 }
michael@0 704 e[i+pos] = tmp;
michael@0 705 e_1[i+pos] = 1.f/tmp;
michael@0 706 }
michael@0 707 /* Hack to get 20 ms working with APPLICATION_AUDIO
michael@0 708 The real problem is that the corresponding memory needs to use 1.5 ms
michael@0 709 from this frame and 1 ms from the next frame */
michael@0 710 e[i+pos] = e[i+pos-1];
michael@0 711 if (buffering)
michael@0 712 N=IMIN(MAX_DYNAMIC_FRAMESIZE, N+2);
michael@0 713 bestLM = transient_viterbi(e, e_1, N, (int)((1.f+.5f*tonality)*(60*C+40)), bitrate/400);
michael@0 714 mem[0] = e[1<<bestLM];
michael@0 715 if (buffering)
michael@0 716 {
michael@0 717 mem[1] = e[(1<<bestLM)+1];
michael@0 718 mem[2] = e[(1<<bestLM)+2];
michael@0 719 }
michael@0 720 return bestLM;
michael@0 721 }
michael@0 722
michael@0 723 #endif
michael@0 724
michael@0 725 #ifndef DISABLE_FLOAT_API
michael@0 726 #ifdef FIXED_POINT
michael@0 727 #define PCM2VAL(x) FLOAT2INT16(x)
michael@0 728 #else
michael@0 729 #define PCM2VAL(x) SCALEIN(x)
michael@0 730 #endif
michael@0 731 void downmix_float(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C)
michael@0 732 {
michael@0 733 const float *x;
michael@0 734 opus_val32 scale;
michael@0 735 int j;
michael@0 736 x = (const float *)_x;
michael@0 737 for (j=0;j<subframe;j++)
michael@0 738 sub[j] = PCM2VAL(x[(j+offset)*C+c1]);
michael@0 739 if (c2>-1)
michael@0 740 {
michael@0 741 for (j=0;j<subframe;j++)
michael@0 742 sub[j] += PCM2VAL(x[(j+offset)*C+c2]);
michael@0 743 } else if (c2==-2)
michael@0 744 {
michael@0 745 int c;
michael@0 746 for (c=1;c<C;c++)
michael@0 747 {
michael@0 748 for (j=0;j<subframe;j++)
michael@0 749 sub[j] += PCM2VAL(x[(j+offset)*C+c]);
michael@0 750 }
michael@0 751 }
michael@0 752 #ifdef FIXED_POINT
michael@0 753 scale = (1<<SIG_SHIFT);
michael@0 754 #else
michael@0 755 scale = 1.f;
michael@0 756 #endif
michael@0 757 if (C==-2)
michael@0 758 scale /= C;
michael@0 759 else
michael@0 760 scale /= 2;
michael@0 761 for (j=0;j<subframe;j++)
michael@0 762 sub[j] *= scale;
michael@0 763 }
michael@0 764 #endif
michael@0 765
michael@0 766 void downmix_int(const void *_x, opus_val32 *sub, int subframe, int offset, int c1, int c2, int C)
michael@0 767 {
michael@0 768 const opus_int16 *x;
michael@0 769 opus_val32 scale;
michael@0 770 int j;
michael@0 771 x = (const opus_int16 *)_x;
michael@0 772 for (j=0;j<subframe;j++)
michael@0 773 sub[j] = x[(j+offset)*C+c1];
michael@0 774 if (c2>-1)
michael@0 775 {
michael@0 776 for (j=0;j<subframe;j++)
michael@0 777 sub[j] += x[(j+offset)*C+c2];
michael@0 778 } else if (c2==-2)
michael@0 779 {
michael@0 780 int c;
michael@0 781 for (c=1;c<C;c++)
michael@0 782 {
michael@0 783 for (j=0;j<subframe;j++)
michael@0 784 sub[j] += x[(j+offset)*C+c];
michael@0 785 }
michael@0 786 }
michael@0 787 #ifdef FIXED_POINT
michael@0 788 scale = (1<<SIG_SHIFT);
michael@0 789 #else
michael@0 790 scale = 1.f/32768;
michael@0 791 #endif
michael@0 792 if (C==-2)
michael@0 793 scale /= C;
michael@0 794 else
michael@0 795 scale /= 2;
michael@0 796 for (j=0;j<subframe;j++)
michael@0 797 sub[j] *= scale;
michael@0 798 }
michael@0 799
michael@0 800 opus_int32 frame_size_select(opus_int32 frame_size, int variable_duration, opus_int32 Fs)
michael@0 801 {
michael@0 802 int new_size;
michael@0 803 if (frame_size<Fs/400)
michael@0 804 return -1;
michael@0 805 if (variable_duration == OPUS_FRAMESIZE_ARG)
michael@0 806 new_size = frame_size;
michael@0 807 else if (variable_duration == OPUS_FRAMESIZE_VARIABLE)
michael@0 808 new_size = Fs/50;
michael@0 809 else if (variable_duration >= OPUS_FRAMESIZE_2_5_MS && variable_duration <= OPUS_FRAMESIZE_60_MS)
michael@0 810 new_size = IMIN(3*Fs/50, (Fs/400)<<(variable_duration-OPUS_FRAMESIZE_2_5_MS));
michael@0 811 else
michael@0 812 return -1;
michael@0 813 if (new_size>frame_size)
michael@0 814 return -1;
michael@0 815 if (400*new_size!=Fs && 200*new_size!=Fs && 100*new_size!=Fs &&
michael@0 816 50*new_size!=Fs && 25*new_size!=Fs && 50*new_size!=3*Fs)
michael@0 817 return -1;
michael@0 818 return new_size;
michael@0 819 }
michael@0 820
michael@0 821 opus_int32 compute_frame_size(const void *analysis_pcm, int frame_size,
michael@0 822 int variable_duration, int C, opus_int32 Fs, int bitrate_bps,
michael@0 823 int delay_compensation, downmix_func downmix
michael@0 824 #ifndef DISABLE_FLOAT_API
michael@0 825 , float *subframe_mem
michael@0 826 #endif
michael@0 827 )
michael@0 828 {
michael@0 829 #ifndef DISABLE_FLOAT_API
michael@0 830 if (variable_duration == OPUS_FRAMESIZE_VARIABLE && frame_size >= Fs/200)
michael@0 831 {
michael@0 832 int LM = 3;
michael@0 833 LM = optimize_framesize(analysis_pcm, frame_size, C, Fs, bitrate_bps,
michael@0 834 0, subframe_mem, delay_compensation, downmix);
michael@0 835 while ((Fs/400<<LM)>frame_size)
michael@0 836 LM--;
michael@0 837 frame_size = (Fs/400<<LM);
michael@0 838 } else
michael@0 839 #endif
michael@0 840 {
michael@0 841 frame_size = frame_size_select(frame_size, variable_duration, Fs);
michael@0 842 }
michael@0 843 if (frame_size<0)
michael@0 844 return -1;
michael@0 845 return frame_size;
michael@0 846 }
michael@0 847
michael@0 848 opus_val16 compute_stereo_width(const opus_val16 *pcm, int frame_size, opus_int32 Fs, StereoWidthState *mem)
michael@0 849 {
michael@0 850 opus_val16 corr;
michael@0 851 opus_val16 ldiff;
michael@0 852 opus_val16 width;
michael@0 853 opus_val32 xx, xy, yy;
michael@0 854 opus_val16 sqrt_xx, sqrt_yy;
michael@0 855 opus_val16 qrrt_xx, qrrt_yy;
michael@0 856 int frame_rate;
michael@0 857 int i;
michael@0 858 opus_val16 short_alpha;
michael@0 859
michael@0 860 frame_rate = Fs/frame_size;
michael@0 861 short_alpha = Q15ONE - 25*Q15ONE/IMAX(50,frame_rate);
michael@0 862 xx=xy=yy=0;
michael@0 863 for (i=0;i<frame_size;i+=4)
michael@0 864 {
michael@0 865 opus_val32 pxx=0;
michael@0 866 opus_val32 pxy=0;
michael@0 867 opus_val32 pyy=0;
michael@0 868 opus_val16 x, y;
michael@0 869 x = pcm[2*i];
michael@0 870 y = pcm[2*i+1];
michael@0 871 pxx = SHR32(MULT16_16(x,x),2);
michael@0 872 pxy = SHR32(MULT16_16(x,y),2);
michael@0 873 pyy = SHR32(MULT16_16(y,y),2);
michael@0 874 x = pcm[2*i+2];
michael@0 875 y = pcm[2*i+3];
michael@0 876 pxx += SHR32(MULT16_16(x,x),2);
michael@0 877 pxy += SHR32(MULT16_16(x,y),2);
michael@0 878 pyy += SHR32(MULT16_16(y,y),2);
michael@0 879 x = pcm[2*i+4];
michael@0 880 y = pcm[2*i+5];
michael@0 881 pxx += SHR32(MULT16_16(x,x),2);
michael@0 882 pxy += SHR32(MULT16_16(x,y),2);
michael@0 883 pyy += SHR32(MULT16_16(y,y),2);
michael@0 884 x = pcm[2*i+6];
michael@0 885 y = pcm[2*i+7];
michael@0 886 pxx += SHR32(MULT16_16(x,x),2);
michael@0 887 pxy += SHR32(MULT16_16(x,y),2);
michael@0 888 pyy += SHR32(MULT16_16(y,y),2);
michael@0 889
michael@0 890 xx += SHR32(pxx, 10);
michael@0 891 xy += SHR32(pxy, 10);
michael@0 892 yy += SHR32(pyy, 10);
michael@0 893 }
michael@0 894 mem->XX += MULT16_32_Q15(short_alpha, xx-mem->XX);
michael@0 895 mem->XY += MULT16_32_Q15(short_alpha, xy-mem->XY);
michael@0 896 mem->YY += MULT16_32_Q15(short_alpha, yy-mem->YY);
michael@0 897 mem->XX = MAX32(0, mem->XX);
michael@0 898 mem->XY = MAX32(0, mem->XY);
michael@0 899 mem->YY = MAX32(0, mem->YY);
michael@0 900 if (MAX32(mem->XX, mem->YY)>QCONST16(8e-4f, 18))
michael@0 901 {
michael@0 902 sqrt_xx = celt_sqrt(mem->XX);
michael@0 903 sqrt_yy = celt_sqrt(mem->YY);
michael@0 904 qrrt_xx = celt_sqrt(sqrt_xx);
michael@0 905 qrrt_yy = celt_sqrt(sqrt_yy);
michael@0 906 /* Inter-channel correlation */
michael@0 907 mem->XY = MIN32(mem->XY, sqrt_xx*sqrt_yy);
michael@0 908 corr = SHR32(frac_div32(mem->XY,EPSILON+MULT16_16(sqrt_xx,sqrt_yy)),16);
michael@0 909 /* Approximate loudness difference */
michael@0 910 ldiff = Q15ONE*ABS16(qrrt_xx-qrrt_yy)/(EPSILON+qrrt_xx+qrrt_yy);
michael@0 911 width = MULT16_16_Q15(celt_sqrt(QCONST32(1.f,30)-MULT16_16(corr,corr)), ldiff);
michael@0 912 /* Smoothing over one second */
michael@0 913 mem->smoothed_width += (width-mem->smoothed_width)/frame_rate;
michael@0 914 /* Peak follower */
michael@0 915 mem->max_follower = MAX16(mem->max_follower-QCONST16(.02f,15)/frame_rate, mem->smoothed_width);
michael@0 916 } else {
michael@0 917 width = 0;
michael@0 918 corr=Q15ONE;
michael@0 919 ldiff=0;
michael@0 920 }
michael@0 921 /*printf("%f %f %f %f %f ", corr/(float)Q15ONE, ldiff/(float)Q15ONE, width/(float)Q15ONE, mem->smoothed_width/(float)Q15ONE, mem->max_follower/(float)Q15ONE);*/
michael@0 922 return EXTRACT16(MIN32(Q15ONE,20*mem->max_follower));
michael@0 923 }
michael@0 924
michael@0 925 opus_int32 opus_encode_native(OpusEncoder *st, const opus_val16 *pcm, int frame_size,
michael@0 926 unsigned char *data, opus_int32 out_data_bytes, int lsb_depth,
michael@0 927 const void *analysis_pcm, opus_int32 analysis_size, int c1, int c2, int analysis_channels, downmix_func downmix)
michael@0 928 {
michael@0 929 void *silk_enc;
michael@0 930 CELTEncoder *celt_enc;
michael@0 931 int i;
michael@0 932 int ret=0;
michael@0 933 opus_int32 nBytes;
michael@0 934 ec_enc enc;
michael@0 935 int bytes_target;
michael@0 936 int prefill=0;
michael@0 937 int start_band = 0;
michael@0 938 int redundancy = 0;
michael@0 939 int redundancy_bytes = 0; /* Number of bytes to use for redundancy frame */
michael@0 940 int celt_to_silk = 0;
michael@0 941 VARDECL(opus_val16, pcm_buf);
michael@0 942 int nb_compr_bytes;
michael@0 943 int to_celt = 0;
michael@0 944 opus_uint32 redundant_rng = 0;
michael@0 945 int cutoff_Hz, hp_freq_smth1;
michael@0 946 int voice_est; /* Probability of voice in Q7 */
michael@0 947 opus_int32 equiv_rate;
michael@0 948 int delay_compensation;
michael@0 949 int frame_rate;
michael@0 950 opus_int32 max_rate; /* Max bitrate we're allowed to use */
michael@0 951 int curr_bandwidth;
michael@0 952 opus_val16 HB_gain;
michael@0 953 opus_int32 max_data_bytes; /* Max number of bytes we're allowed to use */
michael@0 954 int total_buffer;
michael@0 955 opus_val16 stereo_width;
michael@0 956 const CELTMode *celt_mode;
michael@0 957 AnalysisInfo analysis_info;
michael@0 958 int analysis_read_pos_bak=-1;
michael@0 959 int analysis_read_subframe_bak=-1;
michael@0 960 VARDECL(opus_val16, tmp_prefill);
michael@0 961
michael@0 962 ALLOC_STACK;
michael@0 963
michael@0 964 max_data_bytes = IMIN(1276, out_data_bytes);
michael@0 965
michael@0 966 st->rangeFinal = 0;
michael@0 967 if ((!st->variable_duration && 400*frame_size != st->Fs && 200*frame_size != st->Fs && 100*frame_size != st->Fs &&
michael@0 968 50*frame_size != st->Fs && 25*frame_size != st->Fs && 50*frame_size != 3*st->Fs)
michael@0 969 || (400*frame_size < st->Fs)
michael@0 970 || max_data_bytes<=0
michael@0 971 )
michael@0 972 {
michael@0 973 RESTORE_STACK;
michael@0 974 return OPUS_BAD_ARG;
michael@0 975 }
michael@0 976 silk_enc = (char*)st+st->silk_enc_offset;
michael@0 977 celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset);
michael@0 978 if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY)
michael@0 979 delay_compensation = 0;
michael@0 980 else
michael@0 981 delay_compensation = st->delay_compensation;
michael@0 982
michael@0 983 lsb_depth = IMIN(lsb_depth, st->lsb_depth);
michael@0 984
michael@0 985 analysis_info.valid = 0;
michael@0 986 celt_encoder_ctl(celt_enc, CELT_GET_MODE(&celt_mode));
michael@0 987 #ifndef DISABLE_FLOAT_API
michael@0 988 #ifdef FIXED_POINT
michael@0 989 if (st->silk_mode.complexity >= 10 && st->Fs==48000)
michael@0 990 #else
michael@0 991 if (st->silk_mode.complexity >= 7 && st->Fs==48000)
michael@0 992 #endif
michael@0 993 {
michael@0 994 analysis_read_pos_bak = st->analysis.read_pos;
michael@0 995 analysis_read_subframe_bak = st->analysis.read_subframe;
michael@0 996 run_analysis(&st->analysis, celt_mode, analysis_pcm, analysis_size, frame_size,
michael@0 997 c1, c2, analysis_channels, st->Fs,
michael@0 998 lsb_depth, downmix, &analysis_info);
michael@0 999 }
michael@0 1000 #endif
michael@0 1001
michael@0 1002 st->voice_ratio = -1;
michael@0 1003
michael@0 1004 #ifndef DISABLE_FLOAT_API
michael@0 1005 st->detected_bandwidth = 0;
michael@0 1006 if (analysis_info.valid)
michael@0 1007 {
michael@0 1008 int analysis_bandwidth;
michael@0 1009 if (st->signal_type == OPUS_AUTO)
michael@0 1010 st->voice_ratio = (int)floor(.5+100*(1-analysis_info.music_prob));
michael@0 1011
michael@0 1012 analysis_bandwidth = analysis_info.bandwidth;
michael@0 1013 if (analysis_bandwidth<=12)
michael@0 1014 st->detected_bandwidth = OPUS_BANDWIDTH_NARROWBAND;
michael@0 1015 else if (analysis_bandwidth<=14)
michael@0 1016 st->detected_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND;
michael@0 1017 else if (analysis_bandwidth<=16)
michael@0 1018 st->detected_bandwidth = OPUS_BANDWIDTH_WIDEBAND;
michael@0 1019 else if (analysis_bandwidth<=18)
michael@0 1020 st->detected_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND;
michael@0 1021 else
michael@0 1022 st->detected_bandwidth = OPUS_BANDWIDTH_FULLBAND;
michael@0 1023 }
michael@0 1024 #endif
michael@0 1025
michael@0 1026 if (st->channels==2 && st->force_channels!=1)
michael@0 1027 stereo_width = compute_stereo_width(pcm, frame_size, st->Fs, &st->width_mem);
michael@0 1028 else
michael@0 1029 stereo_width = 0;
michael@0 1030 total_buffer = delay_compensation;
michael@0 1031 st->bitrate_bps = user_bitrate_to_bitrate(st, frame_size, max_data_bytes);
michael@0 1032
michael@0 1033 frame_rate = st->Fs/frame_size;
michael@0 1034 if (max_data_bytes<3 || st->bitrate_bps < 3*frame_rate*8
michael@0 1035 || (frame_rate<50 && (max_data_bytes*frame_rate<300 || st->bitrate_bps < 2400)))
michael@0 1036 {
michael@0 1037 /*If the space is too low to do something useful, emit 'PLC' frames.*/
michael@0 1038 int tocmode = st->mode;
michael@0 1039 int bw = st->bandwidth == 0 ? OPUS_BANDWIDTH_NARROWBAND : st->bandwidth;
michael@0 1040 if (tocmode==0)
michael@0 1041 tocmode = MODE_SILK_ONLY;
michael@0 1042 if (frame_rate>100)
michael@0 1043 tocmode = MODE_CELT_ONLY;
michael@0 1044 if (frame_rate < 50)
michael@0 1045 tocmode = MODE_SILK_ONLY;
michael@0 1046 if(tocmode==MODE_SILK_ONLY&&bw>OPUS_BANDWIDTH_WIDEBAND)
michael@0 1047 bw=OPUS_BANDWIDTH_WIDEBAND;
michael@0 1048 else if (tocmode==MODE_CELT_ONLY&&bw==OPUS_BANDWIDTH_MEDIUMBAND)
michael@0 1049 bw=OPUS_BANDWIDTH_NARROWBAND;
michael@0 1050 else if (bw<=OPUS_BANDWIDTH_SUPERWIDEBAND)
michael@0 1051 bw=OPUS_BANDWIDTH_SUPERWIDEBAND;
michael@0 1052 data[0] = gen_toc(tocmode, frame_rate, bw, st->stream_channels);
michael@0 1053 RESTORE_STACK;
michael@0 1054 return 1;
michael@0 1055 }
michael@0 1056 if (!st->use_vbr)
michael@0 1057 {
michael@0 1058 int cbrBytes;
michael@0 1059 cbrBytes = IMIN( (st->bitrate_bps + 4*frame_rate)/(8*frame_rate) , max_data_bytes);
michael@0 1060 st->bitrate_bps = cbrBytes * (8*frame_rate);
michael@0 1061 max_data_bytes = cbrBytes;
michael@0 1062 }
michael@0 1063 max_rate = frame_rate*max_data_bytes*8;
michael@0 1064
michael@0 1065 /* Equivalent 20-ms rate for mode/channel/bandwidth decisions */
michael@0 1066 equiv_rate = st->bitrate_bps - (40*st->channels+20)*(st->Fs/frame_size - 50);
michael@0 1067
michael@0 1068 if (st->signal_type == OPUS_SIGNAL_VOICE)
michael@0 1069 voice_est = 127;
michael@0 1070 else if (st->signal_type == OPUS_SIGNAL_MUSIC)
michael@0 1071 voice_est = 0;
michael@0 1072 else if (st->voice_ratio >= 0)
michael@0 1073 {
michael@0 1074 voice_est = st->voice_ratio*327>>8;
michael@0 1075 /* For AUDIO, never be more than 90% confident of having speech */
michael@0 1076 if (st->application == OPUS_APPLICATION_AUDIO)
michael@0 1077 voice_est = IMIN(voice_est, 115);
michael@0 1078 } else if (st->application == OPUS_APPLICATION_VOIP)
michael@0 1079 voice_est = 115;
michael@0 1080 else
michael@0 1081 voice_est = 48;
michael@0 1082
michael@0 1083 if (st->force_channels!=OPUS_AUTO && st->channels == 2)
michael@0 1084 {
michael@0 1085 st->stream_channels = st->force_channels;
michael@0 1086 } else {
michael@0 1087 #ifdef FUZZING
michael@0 1088 /* Random mono/stereo decision */
michael@0 1089 if (st->channels == 2 && (rand()&0x1F)==0)
michael@0 1090 st->stream_channels = 3-st->stream_channels;
michael@0 1091 #else
michael@0 1092 /* Rate-dependent mono-stereo decision */
michael@0 1093 if (st->channels == 2)
michael@0 1094 {
michael@0 1095 opus_int32 stereo_threshold;
michael@0 1096 stereo_threshold = stereo_music_threshold + ((voice_est*voice_est*(stereo_voice_threshold-stereo_music_threshold))>>14);
michael@0 1097 if (st->stream_channels == 2)
michael@0 1098 stereo_threshold -= 1000;
michael@0 1099 else
michael@0 1100 stereo_threshold += 1000;
michael@0 1101 st->stream_channels = (equiv_rate > stereo_threshold) ? 2 : 1;
michael@0 1102 } else {
michael@0 1103 st->stream_channels = st->channels;
michael@0 1104 }
michael@0 1105 #endif
michael@0 1106 }
michael@0 1107 equiv_rate = st->bitrate_bps - (40*st->stream_channels+20)*(st->Fs/frame_size - 50);
michael@0 1108
michael@0 1109 /* Mode selection depending on application and signal type */
michael@0 1110 if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY)
michael@0 1111 {
michael@0 1112 st->mode = MODE_CELT_ONLY;
michael@0 1113 } else if (st->user_forced_mode == OPUS_AUTO)
michael@0 1114 {
michael@0 1115 #ifdef FUZZING
michael@0 1116 /* Random mode switching */
michael@0 1117 if ((rand()&0xF)==0)
michael@0 1118 {
michael@0 1119 if ((rand()&0x1)==0)
michael@0 1120 st->mode = MODE_CELT_ONLY;
michael@0 1121 else
michael@0 1122 st->mode = MODE_SILK_ONLY;
michael@0 1123 } else {
michael@0 1124 if (st->prev_mode==MODE_CELT_ONLY)
michael@0 1125 st->mode = MODE_CELT_ONLY;
michael@0 1126 else
michael@0 1127 st->mode = MODE_SILK_ONLY;
michael@0 1128 }
michael@0 1129 #else
michael@0 1130 opus_int32 mode_voice, mode_music;
michael@0 1131 opus_int32 threshold;
michael@0 1132
michael@0 1133 /* Interpolate based on stereo width */
michael@0 1134 mode_voice = (opus_int32)(MULT16_32_Q15(Q15ONE-stereo_width,mode_thresholds[0][0])
michael@0 1135 + MULT16_32_Q15(stereo_width,mode_thresholds[1][0]));
michael@0 1136 mode_music = (opus_int32)(MULT16_32_Q15(Q15ONE-stereo_width,mode_thresholds[1][1])
michael@0 1137 + MULT16_32_Q15(stereo_width,mode_thresholds[1][1]));
michael@0 1138 /* Interpolate based on speech/music probability */
michael@0 1139 threshold = mode_music + ((voice_est*voice_est*(mode_voice-mode_music))>>14);
michael@0 1140 /* Bias towards SILK for VoIP because of some useful features */
michael@0 1141 if (st->application == OPUS_APPLICATION_VOIP)
michael@0 1142 threshold += 8000;
michael@0 1143
michael@0 1144 /*printf("%f %d\n", stereo_width/(float)Q15ONE, threshold);*/
michael@0 1145 /* Hysteresis */
michael@0 1146 if (st->prev_mode == MODE_CELT_ONLY)
michael@0 1147 threshold -= 4000;
michael@0 1148 else if (st->prev_mode>0)
michael@0 1149 threshold += 4000;
michael@0 1150
michael@0 1151 st->mode = (equiv_rate >= threshold) ? MODE_CELT_ONLY: MODE_SILK_ONLY;
michael@0 1152
michael@0 1153 /* When FEC is enabled and there's enough packet loss, use SILK */
michael@0 1154 if (st->silk_mode.useInBandFEC && st->silk_mode.packetLossPercentage > (128-voice_est)>>4)
michael@0 1155 st->mode = MODE_SILK_ONLY;
michael@0 1156 /* When encoding voice and DTX is enabled, set the encoder to SILK mode (at least for now) */
michael@0 1157 if (st->silk_mode.useDTX && voice_est > 100)
michael@0 1158 st->mode = MODE_SILK_ONLY;
michael@0 1159 #endif
michael@0 1160 } else {
michael@0 1161 st->mode = st->user_forced_mode;
michael@0 1162 }
michael@0 1163
michael@0 1164 /* Override the chosen mode to make sure we meet the requested frame size */
michael@0 1165 if (st->mode != MODE_CELT_ONLY && frame_size < st->Fs/100)
michael@0 1166 st->mode = MODE_CELT_ONLY;
michael@0 1167 if (st->lfe)
michael@0 1168 st->mode = MODE_CELT_ONLY;
michael@0 1169 /* If max_data_bytes represents less than 8 kb/s, switch to CELT-only mode */
michael@0 1170 if (max_data_bytes < (frame_rate > 50 ? 12000 : 8000)*frame_size / (st->Fs * 8))
michael@0 1171 st->mode = MODE_CELT_ONLY;
michael@0 1172
michael@0 1173 if (st->stream_channels == 1 && st->prev_channels ==2 && st->silk_mode.toMono==0
michael@0 1174 && st->mode != MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY)
michael@0 1175 {
michael@0 1176 /* Delay stereo->mono transition by two frames so that SILK can do a smooth downmix */
michael@0 1177 st->silk_mode.toMono = 1;
michael@0 1178 st->stream_channels = 2;
michael@0 1179 } else {
michael@0 1180 st->silk_mode.toMono = 0;
michael@0 1181 }
michael@0 1182
michael@0 1183 if (st->prev_mode > 0 &&
michael@0 1184 ((st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) ||
michael@0 1185 (st->mode == MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY)))
michael@0 1186 {
michael@0 1187 redundancy = 1;
michael@0 1188 celt_to_silk = (st->mode != MODE_CELT_ONLY);
michael@0 1189 if (!celt_to_silk)
michael@0 1190 {
michael@0 1191 /* Switch to SILK/hybrid if frame size is 10 ms or more*/
michael@0 1192 if (frame_size >= st->Fs/100)
michael@0 1193 {
michael@0 1194 st->mode = st->prev_mode;
michael@0 1195 to_celt = 1;
michael@0 1196 } else {
michael@0 1197 redundancy=0;
michael@0 1198 }
michael@0 1199 }
michael@0 1200 }
michael@0 1201 /* For the first frame at a new SILK bandwidth */
michael@0 1202 if (st->silk_bw_switch)
michael@0 1203 {
michael@0 1204 redundancy = 1;
michael@0 1205 celt_to_silk = 1;
michael@0 1206 st->silk_bw_switch = 0;
michael@0 1207 prefill=1;
michael@0 1208 }
michael@0 1209
michael@0 1210 if (redundancy)
michael@0 1211 {
michael@0 1212 /* Fair share of the max size allowed */
michael@0 1213 redundancy_bytes = IMIN(257, max_data_bytes*(opus_int32)(st->Fs/200)/(frame_size+st->Fs/200));
michael@0 1214 /* For VBR, target the actual bitrate (subject to the limit above) */
michael@0 1215 if (st->use_vbr)
michael@0 1216 redundancy_bytes = IMIN(redundancy_bytes, st->bitrate_bps/1600);
michael@0 1217 }
michael@0 1218
michael@0 1219 if (st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY)
michael@0 1220 {
michael@0 1221 silk_EncControlStruct dummy;
michael@0 1222 silk_InitEncoder( silk_enc, st->arch, &dummy);
michael@0 1223 prefill=1;
michael@0 1224 }
michael@0 1225
michael@0 1226 /* Automatic (rate-dependent) bandwidth selection */
michael@0 1227 if (st->mode == MODE_CELT_ONLY || st->first || st->silk_mode.allowBandwidthSwitch)
michael@0 1228 {
michael@0 1229 const opus_int32 *voice_bandwidth_thresholds, *music_bandwidth_thresholds;
michael@0 1230 opus_int32 bandwidth_thresholds[8];
michael@0 1231 int bandwidth = OPUS_BANDWIDTH_FULLBAND;
michael@0 1232 opus_int32 equiv_rate2;
michael@0 1233
michael@0 1234 equiv_rate2 = equiv_rate;
michael@0 1235 if (st->mode != MODE_CELT_ONLY)
michael@0 1236 {
michael@0 1237 /* Adjust the threshold +/- 10% depending on complexity */
michael@0 1238 equiv_rate2 = equiv_rate2 * (45+st->silk_mode.complexity)/50;
michael@0 1239 /* CBR is less efficient by ~1 kb/s */
michael@0 1240 if (!st->use_vbr)
michael@0 1241 equiv_rate2 -= 1000;
michael@0 1242 }
michael@0 1243 if (st->channels==2 && st->force_channels!=1)
michael@0 1244 {
michael@0 1245 voice_bandwidth_thresholds = stereo_voice_bandwidth_thresholds;
michael@0 1246 music_bandwidth_thresholds = stereo_music_bandwidth_thresholds;
michael@0 1247 } else {
michael@0 1248 voice_bandwidth_thresholds = mono_voice_bandwidth_thresholds;
michael@0 1249 music_bandwidth_thresholds = mono_music_bandwidth_thresholds;
michael@0 1250 }
michael@0 1251 /* Interpolate bandwidth thresholds depending on voice estimation */
michael@0 1252 for (i=0;i<8;i++)
michael@0 1253 {
michael@0 1254 bandwidth_thresholds[i] = music_bandwidth_thresholds[i]
michael@0 1255 + ((voice_est*voice_est*(voice_bandwidth_thresholds[i]-music_bandwidth_thresholds[i]))>>14);
michael@0 1256 }
michael@0 1257 do {
michael@0 1258 int threshold, hysteresis;
michael@0 1259 threshold = bandwidth_thresholds[2*(bandwidth-OPUS_BANDWIDTH_MEDIUMBAND)];
michael@0 1260 hysteresis = bandwidth_thresholds[2*(bandwidth-OPUS_BANDWIDTH_MEDIUMBAND)+1];
michael@0 1261 if (!st->first)
michael@0 1262 {
michael@0 1263 if (st->bandwidth >= bandwidth)
michael@0 1264 threshold -= hysteresis;
michael@0 1265 else
michael@0 1266 threshold += hysteresis;
michael@0 1267 }
michael@0 1268 if (equiv_rate2 >= threshold)
michael@0 1269 break;
michael@0 1270 } while (--bandwidth>OPUS_BANDWIDTH_NARROWBAND);
michael@0 1271 st->bandwidth = bandwidth;
michael@0 1272 /* Prevents any transition to SWB/FB until the SILK layer has fully
michael@0 1273 switched to WB mode and turned the variable LP filter off */
michael@0 1274 if (!st->first && st->mode != MODE_CELT_ONLY && !st->silk_mode.inWBmodeWithoutVariableLP && st->bandwidth > OPUS_BANDWIDTH_WIDEBAND)
michael@0 1275 st->bandwidth = OPUS_BANDWIDTH_WIDEBAND;
michael@0 1276 }
michael@0 1277
michael@0 1278 if (st->bandwidth>st->max_bandwidth)
michael@0 1279 st->bandwidth = st->max_bandwidth;
michael@0 1280
michael@0 1281 if (st->user_bandwidth != OPUS_AUTO)
michael@0 1282 st->bandwidth = st->user_bandwidth;
michael@0 1283
michael@0 1284 /* This prevents us from using hybrid at unsafe CBR/max rates */
michael@0 1285 if (st->mode != MODE_CELT_ONLY && max_rate < 15000)
michael@0 1286 {
michael@0 1287 st->bandwidth = IMIN(st->bandwidth, OPUS_BANDWIDTH_WIDEBAND);
michael@0 1288 }
michael@0 1289
michael@0 1290 /* Prevents Opus from wasting bits on frequencies that are above
michael@0 1291 the Nyquist rate of the input signal */
michael@0 1292 if (st->Fs <= 24000 && st->bandwidth > OPUS_BANDWIDTH_SUPERWIDEBAND)
michael@0 1293 st->bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND;
michael@0 1294 if (st->Fs <= 16000 && st->bandwidth > OPUS_BANDWIDTH_WIDEBAND)
michael@0 1295 st->bandwidth = OPUS_BANDWIDTH_WIDEBAND;
michael@0 1296 if (st->Fs <= 12000 && st->bandwidth > OPUS_BANDWIDTH_MEDIUMBAND)
michael@0 1297 st->bandwidth = OPUS_BANDWIDTH_MEDIUMBAND;
michael@0 1298 if (st->Fs <= 8000 && st->bandwidth > OPUS_BANDWIDTH_NARROWBAND)
michael@0 1299 st->bandwidth = OPUS_BANDWIDTH_NARROWBAND;
michael@0 1300 #ifndef DISABLE_FLOAT_API
michael@0 1301 /* Use detected bandwidth to reduce the encoded bandwidth. */
michael@0 1302 if (st->detected_bandwidth && st->user_bandwidth == OPUS_AUTO)
michael@0 1303 {
michael@0 1304 int min_detected_bandwidth;
michael@0 1305 /* Makes bandwidth detection more conservative just in case the detector
michael@0 1306 gets it wrong when we could have coded a high bandwidth transparently.
michael@0 1307 When operating in SILK/hybrid mode, we don't go below wideband to avoid
michael@0 1308 more complicated switches that require redundancy. */
michael@0 1309 if (equiv_rate <= 18000*st->stream_channels && st->mode == MODE_CELT_ONLY)
michael@0 1310 min_detected_bandwidth = OPUS_BANDWIDTH_NARROWBAND;
michael@0 1311 else if (equiv_rate <= 24000*st->stream_channels && st->mode == MODE_CELT_ONLY)
michael@0 1312 min_detected_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND;
michael@0 1313 else if (equiv_rate <= 30000*st->stream_channels)
michael@0 1314 min_detected_bandwidth = OPUS_BANDWIDTH_WIDEBAND;
michael@0 1315 else if (equiv_rate <= 44000*st->stream_channels)
michael@0 1316 min_detected_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND;
michael@0 1317 else
michael@0 1318 min_detected_bandwidth = OPUS_BANDWIDTH_FULLBAND;
michael@0 1319
michael@0 1320 st->detected_bandwidth = IMAX(st->detected_bandwidth, min_detected_bandwidth);
michael@0 1321 st->bandwidth = IMIN(st->bandwidth, st->detected_bandwidth);
michael@0 1322 }
michael@0 1323 #endif
michael@0 1324 celt_encoder_ctl(celt_enc, OPUS_SET_LSB_DEPTH(lsb_depth));
michael@0 1325
michael@0 1326 /* CELT mode doesn't support mediumband, use wideband instead */
michael@0 1327 if (st->mode == MODE_CELT_ONLY && st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND)
michael@0 1328 st->bandwidth = OPUS_BANDWIDTH_WIDEBAND;
michael@0 1329 if (st->lfe)
michael@0 1330 st->bandwidth = OPUS_BANDWIDTH_NARROWBAND;
michael@0 1331
michael@0 1332 /* Can't support higher than wideband for >20 ms frames */
michael@0 1333 if (frame_size > st->Fs/50 && (st->mode == MODE_CELT_ONLY || st->bandwidth > OPUS_BANDWIDTH_WIDEBAND))
michael@0 1334 {
michael@0 1335 VARDECL(unsigned char, tmp_data);
michael@0 1336 int nb_frames;
michael@0 1337 int bak_mode, bak_bandwidth, bak_channels, bak_to_mono;
michael@0 1338 VARDECL(OpusRepacketizer, rp);
michael@0 1339 opus_int32 bytes_per_frame;
michael@0 1340 opus_int32 repacketize_len;
michael@0 1341
michael@0 1342 #ifndef DISABLE_FLOAT_API
michael@0 1343 if (analysis_read_pos_bak!= -1)
michael@0 1344 {
michael@0 1345 st->analysis.read_pos = analysis_read_pos_bak;
michael@0 1346 st->analysis.read_subframe = analysis_read_subframe_bak;
michael@0 1347 }
michael@0 1348 #endif
michael@0 1349
michael@0 1350 nb_frames = frame_size > st->Fs/25 ? 3 : 2;
michael@0 1351 bytes_per_frame = IMIN(1276,(out_data_bytes-3)/nb_frames);
michael@0 1352
michael@0 1353 ALLOC(tmp_data, nb_frames*bytes_per_frame, unsigned char);
michael@0 1354
michael@0 1355 ALLOC(rp, 1, OpusRepacketizer);
michael@0 1356 opus_repacketizer_init(rp);
michael@0 1357
michael@0 1358 bak_mode = st->user_forced_mode;
michael@0 1359 bak_bandwidth = st->user_bandwidth;
michael@0 1360 bak_channels = st->force_channels;
michael@0 1361
michael@0 1362 st->user_forced_mode = st->mode;
michael@0 1363 st->user_bandwidth = st->bandwidth;
michael@0 1364 st->force_channels = st->stream_channels;
michael@0 1365 bak_to_mono = st->silk_mode.toMono;
michael@0 1366
michael@0 1367 if (bak_to_mono)
michael@0 1368 st->force_channels = 1;
michael@0 1369 else
michael@0 1370 st->prev_channels = st->stream_channels;
michael@0 1371 for (i=0;i<nb_frames;i++)
michael@0 1372 {
michael@0 1373 int tmp_len;
michael@0 1374 st->silk_mode.toMono = 0;
michael@0 1375 /* When switching from SILK/Hybrid to CELT, only ask for a switch at the last frame */
michael@0 1376 if (to_celt && i==nb_frames-1)
michael@0 1377 st->user_forced_mode = MODE_CELT_ONLY;
michael@0 1378 tmp_len = opus_encode_native(st, pcm+i*(st->channels*st->Fs/50), st->Fs/50,
michael@0 1379 tmp_data+i*bytes_per_frame, bytes_per_frame, lsb_depth,
michael@0 1380 NULL, 0, c1, c2, analysis_channels, downmix);
michael@0 1381 if (tmp_len<0)
michael@0 1382 {
michael@0 1383 RESTORE_STACK;
michael@0 1384 return OPUS_INTERNAL_ERROR;
michael@0 1385 }
michael@0 1386 ret = opus_repacketizer_cat(rp, tmp_data+i*bytes_per_frame, tmp_len);
michael@0 1387 if (ret<0)
michael@0 1388 {
michael@0 1389 RESTORE_STACK;
michael@0 1390 return OPUS_INTERNAL_ERROR;
michael@0 1391 }
michael@0 1392 }
michael@0 1393 if (st->use_vbr)
michael@0 1394 repacketize_len = out_data_bytes;
michael@0 1395 else
michael@0 1396 repacketize_len = IMIN(3*st->bitrate_bps/(3*8*50/nb_frames), out_data_bytes);
michael@0 1397 ret = opus_repacketizer_out_range_impl(rp, 0, nb_frames, data, repacketize_len, 0, !st->use_vbr);
michael@0 1398 if (ret<0)
michael@0 1399 {
michael@0 1400 RESTORE_STACK;
michael@0 1401 return OPUS_INTERNAL_ERROR;
michael@0 1402 }
michael@0 1403 st->user_forced_mode = bak_mode;
michael@0 1404 st->user_bandwidth = bak_bandwidth;
michael@0 1405 st->force_channels = bak_channels;
michael@0 1406 st->silk_mode.toMono = bak_to_mono;
michael@0 1407 RESTORE_STACK;
michael@0 1408 return ret;
michael@0 1409 }
michael@0 1410 curr_bandwidth = st->bandwidth;
michael@0 1411
michael@0 1412 /* Chooses the appropriate mode for speech
michael@0 1413 *NEVER* switch to/from CELT-only mode here as this will invalidate some assumptions */
michael@0 1414 if (st->mode == MODE_SILK_ONLY && curr_bandwidth > OPUS_BANDWIDTH_WIDEBAND)
michael@0 1415 st->mode = MODE_HYBRID;
michael@0 1416 if (st->mode == MODE_HYBRID && curr_bandwidth <= OPUS_BANDWIDTH_WIDEBAND)
michael@0 1417 st->mode = MODE_SILK_ONLY;
michael@0 1418
michael@0 1419 /* printf("%d %d %d %d\n", st->bitrate_bps, st->stream_channels, st->mode, curr_bandwidth); */
michael@0 1420 bytes_target = IMIN(max_data_bytes-redundancy_bytes, st->bitrate_bps * frame_size / (st->Fs * 8)) - 1;
michael@0 1421
michael@0 1422 data += 1;
michael@0 1423
michael@0 1424 ec_enc_init(&enc, data, max_data_bytes-1);
michael@0 1425
michael@0 1426 ALLOC(pcm_buf, (total_buffer+frame_size)*st->channels, opus_val16);
michael@0 1427 for (i=0;i<total_buffer*st->channels;i++)
michael@0 1428 pcm_buf[i] = st->delay_buffer[(st->encoder_buffer-total_buffer)*st->channels+i];
michael@0 1429
michael@0 1430 if (st->mode == MODE_CELT_ONLY)
michael@0 1431 hp_freq_smth1 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 );
michael@0 1432 else
michael@0 1433 hp_freq_smth1 = ((silk_encoder*)silk_enc)->state_Fxx[0].sCmn.variable_HP_smth1_Q15;
michael@0 1434
michael@0 1435 st->variable_HP_smth2_Q15 = silk_SMLAWB( st->variable_HP_smth2_Q15,
michael@0 1436 hp_freq_smth1 - st->variable_HP_smth2_Q15, SILK_FIX_CONST( VARIABLE_HP_SMTH_COEF2, 16 ) );
michael@0 1437
michael@0 1438 /* convert from log scale to Hertz */
michael@0 1439 cutoff_Hz = silk_log2lin( silk_RSHIFT( st->variable_HP_smth2_Q15, 8 ) );
michael@0 1440
michael@0 1441 if (st->application == OPUS_APPLICATION_VOIP)
michael@0 1442 {
michael@0 1443 hp_cutoff(pcm, cutoff_Hz, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs);
michael@0 1444 } else {
michael@0 1445 dc_reject(pcm, 3, &pcm_buf[total_buffer*st->channels], st->hp_mem, frame_size, st->channels, st->Fs);
michael@0 1446 }
michael@0 1447
michael@0 1448
michael@0 1449
michael@0 1450 /* SILK processing */
michael@0 1451 HB_gain = Q15ONE;
michael@0 1452 if (st->mode != MODE_CELT_ONLY)
michael@0 1453 {
michael@0 1454 opus_int32 total_bitRate, celt_rate;
michael@0 1455 #ifdef FIXED_POINT
michael@0 1456 const opus_int16 *pcm_silk;
michael@0 1457 #else
michael@0 1458 VARDECL(opus_int16, pcm_silk);
michael@0 1459 ALLOC(pcm_silk, st->channels*frame_size, opus_int16);
michael@0 1460 #endif
michael@0 1461
michael@0 1462 /* Distribute bits between SILK and CELT */
michael@0 1463 total_bitRate = 8 * bytes_target * frame_rate;
michael@0 1464 if( st->mode == MODE_HYBRID ) {
michael@0 1465 int HB_gain_ref;
michael@0 1466 /* Base rate for SILK */
michael@0 1467 st->silk_mode.bitRate = st->stream_channels * ( 5000 + 1000 * ( st->Fs == 100 * frame_size ) );
michael@0 1468 if( curr_bandwidth == OPUS_BANDWIDTH_SUPERWIDEBAND ) {
michael@0 1469 /* SILK gets 2/3 of the remaining bits */
michael@0 1470 st->silk_mode.bitRate += ( total_bitRate - st->silk_mode.bitRate ) * 2 / 3;
michael@0 1471 } else { /* FULLBAND */
michael@0 1472 /* SILK gets 3/5 of the remaining bits */
michael@0 1473 st->silk_mode.bitRate += ( total_bitRate - st->silk_mode.bitRate ) * 3 / 5;
michael@0 1474 }
michael@0 1475 /* Don't let SILK use more than 80% */
michael@0 1476 if( st->silk_mode.bitRate > total_bitRate * 4/5 ) {
michael@0 1477 st->silk_mode.bitRate = total_bitRate * 4/5;
michael@0 1478 }
michael@0 1479 if (!st->energy_masking)
michael@0 1480 {
michael@0 1481 /* Increasingly attenuate high band when it gets allocated fewer bits */
michael@0 1482 celt_rate = total_bitRate - st->silk_mode.bitRate;
michael@0 1483 HB_gain_ref = (curr_bandwidth == OPUS_BANDWIDTH_SUPERWIDEBAND) ? 3000 : 3600;
michael@0 1484 HB_gain = SHL32((opus_val32)celt_rate, 9) / SHR32((opus_val32)celt_rate + st->stream_channels * HB_gain_ref, 6);
michael@0 1485 HB_gain = HB_gain < Q15ONE*6/7 ? HB_gain + Q15ONE/7 : Q15ONE;
michael@0 1486 }
michael@0 1487 } else {
michael@0 1488 /* SILK gets all bits */
michael@0 1489 st->silk_mode.bitRate = total_bitRate;
michael@0 1490 }
michael@0 1491
michael@0 1492 /* Surround masking for SILK */
michael@0 1493 if (st->energy_masking && st->use_vbr && !st->lfe)
michael@0 1494 {
michael@0 1495 opus_val32 mask_sum=0;
michael@0 1496 opus_val16 masking_depth;
michael@0 1497 opus_int32 rate_offset;
michael@0 1498 int c;
michael@0 1499 int end = 17;
michael@0 1500 opus_int16 srate = 16000;
michael@0 1501 if (st->bandwidth == OPUS_BANDWIDTH_NARROWBAND)
michael@0 1502 {
michael@0 1503 end = 13;
michael@0 1504 srate = 8000;
michael@0 1505 } else if (st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND)
michael@0 1506 {
michael@0 1507 end = 15;
michael@0 1508 srate = 12000;
michael@0 1509 }
michael@0 1510 for (c=0;c<st->channels;c++)
michael@0 1511 {
michael@0 1512 for(i=0;i<end;i++)
michael@0 1513 {
michael@0 1514 opus_val16 mask;
michael@0 1515 mask = MAX16(MIN16(st->energy_masking[21*c+i],
michael@0 1516 QCONST16(.5f, DB_SHIFT)), -QCONST16(2.0f, DB_SHIFT));
michael@0 1517 if (mask > 0)
michael@0 1518 mask = HALF16(mask);
michael@0 1519 mask_sum += mask;
michael@0 1520 }
michael@0 1521 }
michael@0 1522 /* Conservative rate reduction, we cut the masking in half */
michael@0 1523 masking_depth = mask_sum / end*st->channels;
michael@0 1524 masking_depth += QCONST16(.2f, DB_SHIFT);
michael@0 1525 rate_offset = (opus_int32)PSHR32(MULT16_16(srate, masking_depth), DB_SHIFT);
michael@0 1526 rate_offset = MAX32(rate_offset, -2*st->silk_mode.bitRate/3);
michael@0 1527 /* Split the rate change between the SILK and CELT part for hybrid. */
michael@0 1528 if (st->bandwidth==OPUS_BANDWIDTH_SUPERWIDEBAND || st->bandwidth==OPUS_BANDWIDTH_FULLBAND)
michael@0 1529 st->silk_mode.bitRate += 3*rate_offset/5;
michael@0 1530 else
michael@0 1531 st->silk_mode.bitRate += rate_offset;
michael@0 1532 bytes_target += rate_offset * frame_size / (8 * st->Fs);
michael@0 1533 }
michael@0 1534
michael@0 1535 st->silk_mode.payloadSize_ms = 1000 * frame_size / st->Fs;
michael@0 1536 st->silk_mode.nChannelsAPI = st->channels;
michael@0 1537 st->silk_mode.nChannelsInternal = st->stream_channels;
michael@0 1538 if (curr_bandwidth == OPUS_BANDWIDTH_NARROWBAND) {
michael@0 1539 st->silk_mode.desiredInternalSampleRate = 8000;
michael@0 1540 } else if (curr_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) {
michael@0 1541 st->silk_mode.desiredInternalSampleRate = 12000;
michael@0 1542 } else {
michael@0 1543 silk_assert( st->mode == MODE_HYBRID || curr_bandwidth == OPUS_BANDWIDTH_WIDEBAND );
michael@0 1544 st->silk_mode.desiredInternalSampleRate = 16000;
michael@0 1545 }
michael@0 1546 if( st->mode == MODE_HYBRID ) {
michael@0 1547 /* Don't allow bandwidth reduction at lowest bitrates in hybrid mode */
michael@0 1548 st->silk_mode.minInternalSampleRate = 16000;
michael@0 1549 } else {
michael@0 1550 st->silk_mode.minInternalSampleRate = 8000;
michael@0 1551 }
michael@0 1552
michael@0 1553 if (st->mode == MODE_SILK_ONLY)
michael@0 1554 {
michael@0 1555 opus_int32 effective_max_rate = max_rate;
michael@0 1556 st->silk_mode.maxInternalSampleRate = 16000;
michael@0 1557 if (frame_rate > 50)
michael@0 1558 effective_max_rate = effective_max_rate*2/3;
michael@0 1559 if (effective_max_rate < 13000)
michael@0 1560 {
michael@0 1561 st->silk_mode.maxInternalSampleRate = 12000;
michael@0 1562 st->silk_mode.desiredInternalSampleRate = IMIN(12000, st->silk_mode.desiredInternalSampleRate);
michael@0 1563 }
michael@0 1564 if (effective_max_rate < 9600)
michael@0 1565 {
michael@0 1566 st->silk_mode.maxInternalSampleRate = 8000;
michael@0 1567 st->silk_mode.desiredInternalSampleRate = IMIN(8000, st->silk_mode.desiredInternalSampleRate);
michael@0 1568 }
michael@0 1569 } else {
michael@0 1570 st->silk_mode.maxInternalSampleRate = 16000;
michael@0 1571 }
michael@0 1572
michael@0 1573 st->silk_mode.useCBR = !st->use_vbr;
michael@0 1574
michael@0 1575 /* Call SILK encoder for the low band */
michael@0 1576 nBytes = IMIN(1275, max_data_bytes-1-redundancy_bytes);
michael@0 1577
michael@0 1578 st->silk_mode.maxBits = nBytes*8;
michael@0 1579 /* Only allow up to 90% of the bits for hybrid mode*/
michael@0 1580 if (st->mode == MODE_HYBRID)
michael@0 1581 st->silk_mode.maxBits = (opus_int32)st->silk_mode.maxBits*9/10;
michael@0 1582 if (st->silk_mode.useCBR)
michael@0 1583 {
michael@0 1584 st->silk_mode.maxBits = (st->silk_mode.bitRate * frame_size / (st->Fs * 8))*8;
michael@0 1585 /* Reduce the initial target to make it easier to reach the CBR rate */
michael@0 1586 st->silk_mode.bitRate = IMAX(1, st->silk_mode.bitRate-2000);
michael@0 1587 }
michael@0 1588
michael@0 1589 if (prefill)
michael@0 1590 {
michael@0 1591 opus_int32 zero=0;
michael@0 1592 int prefill_offset;
michael@0 1593 /* Use a smooth onset for the SILK prefill to avoid the encoder trying to encode
michael@0 1594 a discontinuity. The exact location is what we need to avoid leaving any "gap"
michael@0 1595 in the audio when mixing with the redundant CELT frame. Here we can afford to
michael@0 1596 overwrite st->delay_buffer because the only thing that uses it before it gets
michael@0 1597 rewritten is tmp_prefill[] and even then only the part after the ramp really
michael@0 1598 gets used (rather than sent to the encoder and discarded) */
michael@0 1599 prefill_offset = st->channels*(st->encoder_buffer-st->delay_compensation-st->Fs/400);
michael@0 1600 gain_fade(st->delay_buffer+prefill_offset, st->delay_buffer+prefill_offset,
michael@0 1601 0, Q15ONE, celt_mode->overlap, st->Fs/400, st->channels, celt_mode->window, st->Fs);
michael@0 1602 for(i=0;i<prefill_offset;i++)
michael@0 1603 st->delay_buffer[i]=0;
michael@0 1604 #ifdef FIXED_POINT
michael@0 1605 pcm_silk = st->delay_buffer;
michael@0 1606 #else
michael@0 1607 for (i=0;i<st->encoder_buffer*st->channels;i++)
michael@0 1608 pcm_silk[i] = FLOAT2INT16(st->delay_buffer[i]);
michael@0 1609 #endif
michael@0 1610 silk_Encode( silk_enc, &st->silk_mode, pcm_silk, st->encoder_buffer, NULL, &zero, 1 );
michael@0 1611 }
michael@0 1612
michael@0 1613 #ifdef FIXED_POINT
michael@0 1614 pcm_silk = pcm_buf+total_buffer*st->channels;
michael@0 1615 #else
michael@0 1616 for (i=0;i<frame_size*st->channels;i++)
michael@0 1617 pcm_silk[i] = FLOAT2INT16(pcm_buf[total_buffer*st->channels + i]);
michael@0 1618 #endif
michael@0 1619 ret = silk_Encode( silk_enc, &st->silk_mode, pcm_silk, frame_size, &enc, &nBytes, 0 );
michael@0 1620 if( ret ) {
michael@0 1621 /*fprintf (stderr, "SILK encode error: %d\n", ret);*/
michael@0 1622 /* Handle error */
michael@0 1623 RESTORE_STACK;
michael@0 1624 return OPUS_INTERNAL_ERROR;
michael@0 1625 }
michael@0 1626 if (nBytes==0)
michael@0 1627 {
michael@0 1628 st->rangeFinal = 0;
michael@0 1629 data[-1] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels);
michael@0 1630 RESTORE_STACK;
michael@0 1631 return 1;
michael@0 1632 }
michael@0 1633 /* Extract SILK internal bandwidth for signaling in first byte */
michael@0 1634 if( st->mode == MODE_SILK_ONLY ) {
michael@0 1635 if( st->silk_mode.internalSampleRate == 8000 ) {
michael@0 1636 curr_bandwidth = OPUS_BANDWIDTH_NARROWBAND;
michael@0 1637 } else if( st->silk_mode.internalSampleRate == 12000 ) {
michael@0 1638 curr_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND;
michael@0 1639 } else if( st->silk_mode.internalSampleRate == 16000 ) {
michael@0 1640 curr_bandwidth = OPUS_BANDWIDTH_WIDEBAND;
michael@0 1641 }
michael@0 1642 } else {
michael@0 1643 silk_assert( st->silk_mode.internalSampleRate == 16000 );
michael@0 1644 }
michael@0 1645
michael@0 1646 st->silk_mode.opusCanSwitch = st->silk_mode.switchReady;
michael@0 1647 /* FIXME: How do we allocate the redundancy for CBR? */
michael@0 1648 if (st->silk_mode.opusCanSwitch)
michael@0 1649 {
michael@0 1650 redundancy = 1;
michael@0 1651 celt_to_silk = 0;
michael@0 1652 st->silk_bw_switch = 1;
michael@0 1653 }
michael@0 1654 }
michael@0 1655
michael@0 1656 /* CELT processing */
michael@0 1657 {
michael@0 1658 int endband=21;
michael@0 1659
michael@0 1660 switch(curr_bandwidth)
michael@0 1661 {
michael@0 1662 case OPUS_BANDWIDTH_NARROWBAND:
michael@0 1663 endband = 13;
michael@0 1664 break;
michael@0 1665 case OPUS_BANDWIDTH_MEDIUMBAND:
michael@0 1666 case OPUS_BANDWIDTH_WIDEBAND:
michael@0 1667 endband = 17;
michael@0 1668 break;
michael@0 1669 case OPUS_BANDWIDTH_SUPERWIDEBAND:
michael@0 1670 endband = 19;
michael@0 1671 break;
michael@0 1672 case OPUS_BANDWIDTH_FULLBAND:
michael@0 1673 endband = 21;
michael@0 1674 break;
michael@0 1675 }
michael@0 1676 celt_encoder_ctl(celt_enc, CELT_SET_END_BAND(endband));
michael@0 1677 celt_encoder_ctl(celt_enc, CELT_SET_CHANNELS(st->stream_channels));
michael@0 1678 }
michael@0 1679 celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX));
michael@0 1680 if (st->mode != MODE_SILK_ONLY)
michael@0 1681 {
michael@0 1682 opus_val32 celt_pred=2;
michael@0 1683 celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0));
michael@0 1684 /* We may still decide to disable prediction later */
michael@0 1685 if (st->silk_mode.reducedDependency)
michael@0 1686 celt_pred = 0;
michael@0 1687 celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(celt_pred));
michael@0 1688
michael@0 1689 if (st->mode == MODE_HYBRID)
michael@0 1690 {
michael@0 1691 int len;
michael@0 1692
michael@0 1693 len = (ec_tell(&enc)+7)>>3;
michael@0 1694 if (redundancy)
michael@0 1695 len += st->mode == MODE_HYBRID ? 3 : 1;
michael@0 1696 if( st->use_vbr ) {
michael@0 1697 nb_compr_bytes = len + bytes_target - (st->silk_mode.bitRate * frame_size) / (8 * st->Fs);
michael@0 1698 } else {
michael@0 1699 /* check if SILK used up too much */
michael@0 1700 nb_compr_bytes = len > bytes_target ? len : bytes_target;
michael@0 1701 }
michael@0 1702 } else {
michael@0 1703 if (st->use_vbr)
michael@0 1704 {
michael@0 1705 opus_int32 bonus=0;
michael@0 1706 #ifndef DISABLE_FLOAT_API
michael@0 1707 if (st->variable_duration==OPUS_FRAMESIZE_VARIABLE && frame_size != st->Fs/50)
michael@0 1708 {
michael@0 1709 bonus = (60*st->stream_channels+40)*(st->Fs/frame_size-50);
michael@0 1710 if (analysis_info.valid)
michael@0 1711 bonus = (opus_int32)(bonus*(1.f+.5f*analysis_info.tonality));
michael@0 1712 }
michael@0 1713 #endif
michael@0 1714 celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1));
michael@0 1715 celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(st->vbr_constraint));
michael@0 1716 celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps+bonus));
michael@0 1717 nb_compr_bytes = max_data_bytes-1-redundancy_bytes;
michael@0 1718 } else {
michael@0 1719 nb_compr_bytes = bytes_target;
michael@0 1720 }
michael@0 1721 }
michael@0 1722
michael@0 1723 } else {
michael@0 1724 nb_compr_bytes = 0;
michael@0 1725 }
michael@0 1726
michael@0 1727 ALLOC(tmp_prefill, st->channels*st->Fs/400, opus_val16);
michael@0 1728 if (st->mode != MODE_SILK_ONLY && st->mode != st->prev_mode && st->prev_mode > 0)
michael@0 1729 {
michael@0 1730 for (i=0;i<st->channels*st->Fs/400;i++)
michael@0 1731 tmp_prefill[i] = st->delay_buffer[(st->encoder_buffer-total_buffer-st->Fs/400)*st->channels + i];
michael@0 1732 }
michael@0 1733
michael@0 1734 for (i=0;i<st->channels*(st->encoder_buffer-(frame_size+total_buffer));i++)
michael@0 1735 st->delay_buffer[i] = st->delay_buffer[i+st->channels*frame_size];
michael@0 1736 for (;i<st->encoder_buffer*st->channels;i++)
michael@0 1737 st->delay_buffer[i] = pcm_buf[(frame_size+total_buffer-st->encoder_buffer)*st->channels+i];
michael@0 1738
michael@0 1739 /* gain_fade() and stereo_fade() need to be after the buffer copying
michael@0 1740 because we don't want any of this to affect the SILK part */
michael@0 1741 if( st->prev_HB_gain < Q15ONE || HB_gain < Q15ONE ) {
michael@0 1742 gain_fade(pcm_buf, pcm_buf,
michael@0 1743 st->prev_HB_gain, HB_gain, celt_mode->overlap, frame_size, st->channels, celt_mode->window, st->Fs);
michael@0 1744 }
michael@0 1745 st->prev_HB_gain = HB_gain;
michael@0 1746 if (st->mode != MODE_HYBRID || st->stream_channels==1)
michael@0 1747 st->silk_mode.stereoWidth_Q14 = IMIN((1<<14),2*IMAX(0,equiv_rate-30000));
michael@0 1748 if( !st->energy_masking && st->channels == 2 ) {
michael@0 1749 /* Apply stereo width reduction (at low bitrates) */
michael@0 1750 if( st->hybrid_stereo_width_Q14 < (1 << 14) || st->silk_mode.stereoWidth_Q14 < (1 << 14) ) {
michael@0 1751 opus_val16 g1, g2;
michael@0 1752 g1 = st->hybrid_stereo_width_Q14;
michael@0 1753 g2 = (opus_val16)(st->silk_mode.stereoWidth_Q14);
michael@0 1754 #ifdef FIXED_POINT
michael@0 1755 g1 = g1==16384 ? Q15ONE : SHL16(g1,1);
michael@0 1756 g2 = g2==16384 ? Q15ONE : SHL16(g2,1);
michael@0 1757 #else
michael@0 1758 g1 *= (1.f/16384);
michael@0 1759 g2 *= (1.f/16384);
michael@0 1760 #endif
michael@0 1761 stereo_fade(pcm_buf, pcm_buf, g1, g2, celt_mode->overlap,
michael@0 1762 frame_size, st->channels, celt_mode->window, st->Fs);
michael@0 1763 st->hybrid_stereo_width_Q14 = st->silk_mode.stereoWidth_Q14;
michael@0 1764 }
michael@0 1765 }
michael@0 1766
michael@0 1767 if ( st->mode != MODE_CELT_ONLY && ec_tell(&enc)+17+20*(st->mode == MODE_HYBRID) <= 8*(max_data_bytes-1))
michael@0 1768 {
michael@0 1769 /* For SILK mode, the redundancy is inferred from the length */
michael@0 1770 if (st->mode == MODE_HYBRID && (redundancy || ec_tell(&enc)+37 <= 8*nb_compr_bytes))
michael@0 1771 ec_enc_bit_logp(&enc, redundancy, 12);
michael@0 1772 if (redundancy)
michael@0 1773 {
michael@0 1774 int max_redundancy;
michael@0 1775 ec_enc_bit_logp(&enc, celt_to_silk, 1);
michael@0 1776 if (st->mode == MODE_HYBRID)
michael@0 1777 max_redundancy = (max_data_bytes-1)-nb_compr_bytes;
michael@0 1778 else
michael@0 1779 max_redundancy = (max_data_bytes-1)-((ec_tell(&enc)+7)>>3);
michael@0 1780 /* Target the same bit-rate for redundancy as for the rest,
michael@0 1781 up to a max of 257 bytes */
michael@0 1782 redundancy_bytes = IMIN(max_redundancy, st->bitrate_bps/1600);
michael@0 1783 redundancy_bytes = IMIN(257, IMAX(2, redundancy_bytes));
michael@0 1784 if (st->mode == MODE_HYBRID)
michael@0 1785 ec_enc_uint(&enc, redundancy_bytes-2, 256);
michael@0 1786 }
michael@0 1787 } else {
michael@0 1788 redundancy = 0;
michael@0 1789 }
michael@0 1790
michael@0 1791 if (!redundancy)
michael@0 1792 {
michael@0 1793 st->silk_bw_switch = 0;
michael@0 1794 redundancy_bytes = 0;
michael@0 1795 }
michael@0 1796 if (st->mode != MODE_CELT_ONLY)start_band=17;
michael@0 1797
michael@0 1798 if (st->mode == MODE_SILK_ONLY)
michael@0 1799 {
michael@0 1800 ret = (ec_tell(&enc)+7)>>3;
michael@0 1801 ec_enc_done(&enc);
michael@0 1802 nb_compr_bytes = ret;
michael@0 1803 } else {
michael@0 1804 nb_compr_bytes = IMIN((max_data_bytes-1)-redundancy_bytes, nb_compr_bytes);
michael@0 1805 ec_enc_shrink(&enc, nb_compr_bytes);
michael@0 1806 }
michael@0 1807
michael@0 1808 #ifndef DISABLE_FLOAT_API
michael@0 1809 if (redundancy || st->mode != MODE_SILK_ONLY)
michael@0 1810 celt_encoder_ctl(celt_enc, CELT_SET_ANALYSIS(&analysis_info));
michael@0 1811 #endif
michael@0 1812
michael@0 1813 /* 5 ms redundant frame for CELT->SILK */
michael@0 1814 if (redundancy && celt_to_silk)
michael@0 1815 {
michael@0 1816 int err;
michael@0 1817 celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0));
michael@0 1818 celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0));
michael@0 1819 err = celt_encode_with_ec(celt_enc, pcm_buf, st->Fs/200, data+nb_compr_bytes, redundancy_bytes, NULL);
michael@0 1820 if (err < 0)
michael@0 1821 {
michael@0 1822 RESTORE_STACK;
michael@0 1823 return OPUS_INTERNAL_ERROR;
michael@0 1824 }
michael@0 1825 celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&redundant_rng));
michael@0 1826 celt_encoder_ctl(celt_enc, OPUS_RESET_STATE);
michael@0 1827 }
michael@0 1828
michael@0 1829 celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(start_band));
michael@0 1830
michael@0 1831 if (st->mode != MODE_SILK_ONLY)
michael@0 1832 {
michael@0 1833 if (st->mode != st->prev_mode && st->prev_mode > 0)
michael@0 1834 {
michael@0 1835 unsigned char dummy[2];
michael@0 1836 celt_encoder_ctl(celt_enc, OPUS_RESET_STATE);
michael@0 1837
michael@0 1838 /* Prefilling */
michael@0 1839 celt_encode_with_ec(celt_enc, tmp_prefill, st->Fs/400, dummy, 2, NULL);
michael@0 1840 celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0));
michael@0 1841 }
michael@0 1842 /* If false, we already busted the budget and we'll end up with a "PLC packet" */
michael@0 1843 if (ec_tell(&enc) <= 8*nb_compr_bytes)
michael@0 1844 {
michael@0 1845 ret = celt_encode_with_ec(celt_enc, pcm_buf, frame_size, NULL, nb_compr_bytes, &enc);
michael@0 1846 if (ret < 0)
michael@0 1847 {
michael@0 1848 RESTORE_STACK;
michael@0 1849 return OPUS_INTERNAL_ERROR;
michael@0 1850 }
michael@0 1851 }
michael@0 1852 }
michael@0 1853
michael@0 1854 /* 5 ms redundant frame for SILK->CELT */
michael@0 1855 if (redundancy && !celt_to_silk)
michael@0 1856 {
michael@0 1857 int err;
michael@0 1858 unsigned char dummy[2];
michael@0 1859 int N2, N4;
michael@0 1860 N2 = st->Fs/200;
michael@0 1861 N4 = st->Fs/400;
michael@0 1862
michael@0 1863 celt_encoder_ctl(celt_enc, OPUS_RESET_STATE);
michael@0 1864 celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0));
michael@0 1865 celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0));
michael@0 1866
michael@0 1867 /* NOTE: We could speed this up slightly (at the expense of code size) by just adding a function that prefills the buffer */
michael@0 1868 celt_encode_with_ec(celt_enc, pcm_buf+st->channels*(frame_size-N2-N4), N4, dummy, 2, NULL);
michael@0 1869
michael@0 1870 err = celt_encode_with_ec(celt_enc, pcm_buf+st->channels*(frame_size-N2), N2, data+nb_compr_bytes, redundancy_bytes, NULL);
michael@0 1871 if (err < 0)
michael@0 1872 {
michael@0 1873 RESTORE_STACK;
michael@0 1874 return OPUS_INTERNAL_ERROR;
michael@0 1875 }
michael@0 1876 celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&redundant_rng));
michael@0 1877 }
michael@0 1878
michael@0 1879
michael@0 1880
michael@0 1881 /* Signalling the mode in the first byte */
michael@0 1882 data--;
michael@0 1883 data[0] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels);
michael@0 1884
michael@0 1885 st->rangeFinal = enc.rng ^ redundant_rng;
michael@0 1886
michael@0 1887 if (to_celt)
michael@0 1888 st->prev_mode = MODE_CELT_ONLY;
michael@0 1889 else
michael@0 1890 st->prev_mode = st->mode;
michael@0 1891 st->prev_channels = st->stream_channels;
michael@0 1892 st->prev_framesize = frame_size;
michael@0 1893
michael@0 1894 st->first = 0;
michael@0 1895
michael@0 1896 /* In the unlikely case that the SILK encoder busted its target, tell
michael@0 1897 the decoder to call the PLC */
michael@0 1898 if (ec_tell(&enc) > (max_data_bytes-1)*8)
michael@0 1899 {
michael@0 1900 if (max_data_bytes < 2)
michael@0 1901 {
michael@0 1902 RESTORE_STACK;
michael@0 1903 return OPUS_BUFFER_TOO_SMALL;
michael@0 1904 }
michael@0 1905 data[1] = 0;
michael@0 1906 ret = 1;
michael@0 1907 st->rangeFinal = 0;
michael@0 1908 } else if (st->mode==MODE_SILK_ONLY&&!redundancy)
michael@0 1909 {
michael@0 1910 /*When in LPC only mode it's perfectly
michael@0 1911 reasonable to strip off trailing zero bytes as
michael@0 1912 the required range decoder behavior is to
michael@0 1913 fill these in. This can't be done when the MDCT
michael@0 1914 modes are used because the decoder needs to know
michael@0 1915 the actual length for allocation purposes.*/
michael@0 1916 while(ret>2&&data[ret]==0)ret--;
michael@0 1917 }
michael@0 1918 /* Count ToC and redundancy */
michael@0 1919 ret += 1+redundancy_bytes;
michael@0 1920 if (!st->use_vbr)
michael@0 1921 {
michael@0 1922 if (opus_packet_pad(data, ret, max_data_bytes) != OPUS_OK)
michael@0 1923
michael@0 1924 {
michael@0 1925 RESTORE_STACK;
michael@0 1926 return OPUS_INTERNAL_ERROR;
michael@0 1927 }
michael@0 1928 ret = max_data_bytes;
michael@0 1929 }
michael@0 1930 RESTORE_STACK;
michael@0 1931 return ret;
michael@0 1932 }
michael@0 1933
michael@0 1934 #ifdef FIXED_POINT
michael@0 1935
michael@0 1936 #ifndef DISABLE_FLOAT_API
michael@0 1937 opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int analysis_frame_size,
michael@0 1938 unsigned char *data, opus_int32 max_data_bytes)
michael@0 1939 {
michael@0 1940 int i, ret;
michael@0 1941 int frame_size;
michael@0 1942 int delay_compensation;
michael@0 1943 VARDECL(opus_int16, in);
michael@0 1944 ALLOC_STACK;
michael@0 1945
michael@0 1946 if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY)
michael@0 1947 delay_compensation = 0;
michael@0 1948 else
michael@0 1949 delay_compensation = st->delay_compensation;
michael@0 1950 frame_size = compute_frame_size(pcm, analysis_frame_size,
michael@0 1951 st->variable_duration, st->channels, st->Fs, st->bitrate_bps,
michael@0 1952 delay_compensation, downmix_float, st->analysis.subframe_mem);
michael@0 1953
michael@0 1954 ALLOC(in, frame_size*st->channels, opus_int16);
michael@0 1955
michael@0 1956 for (i=0;i<frame_size*st->channels;i++)
michael@0 1957 in[i] = FLOAT2INT16(pcm[i]);
michael@0 1958 ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, 16, pcm, analysis_frame_size, 0, -2, st->channels, downmix_float);
michael@0 1959 RESTORE_STACK;
michael@0 1960 return ret;
michael@0 1961 }
michael@0 1962 #endif
michael@0 1963
michael@0 1964 opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int analysis_frame_size,
michael@0 1965 unsigned char *data, opus_int32 out_data_bytes)
michael@0 1966 {
michael@0 1967 int frame_size;
michael@0 1968 int delay_compensation;
michael@0 1969 if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY)
michael@0 1970 delay_compensation = 0;
michael@0 1971 else
michael@0 1972 delay_compensation = st->delay_compensation;
michael@0 1973 frame_size = compute_frame_size(pcm, analysis_frame_size,
michael@0 1974 st->variable_duration, st->channels, st->Fs, st->bitrate_bps,
michael@0 1975 delay_compensation, downmix_int
michael@0 1976 #ifndef DISABLE_FLOAT_API
michael@0 1977 , st->analysis.subframe_mem
michael@0 1978 #endif
michael@0 1979 );
michael@0 1980 return opus_encode_native(st, pcm, frame_size, data, out_data_bytes, 16, pcm, analysis_frame_size, 0, -2, st->channels, downmix_int);
michael@0 1981 }
michael@0 1982
michael@0 1983 #else
michael@0 1984 opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int analysis_frame_size,
michael@0 1985 unsigned char *data, opus_int32 max_data_bytes)
michael@0 1986 {
michael@0 1987 int i, ret;
michael@0 1988 int frame_size;
michael@0 1989 int delay_compensation;
michael@0 1990 VARDECL(float, in);
michael@0 1991 ALLOC_STACK;
michael@0 1992
michael@0 1993 if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY)
michael@0 1994 delay_compensation = 0;
michael@0 1995 else
michael@0 1996 delay_compensation = st->delay_compensation;
michael@0 1997 frame_size = compute_frame_size(pcm, analysis_frame_size,
michael@0 1998 st->variable_duration, st->channels, st->Fs, st->bitrate_bps,
michael@0 1999 delay_compensation, downmix_int, st->analysis.subframe_mem);
michael@0 2000
michael@0 2001 ALLOC(in, frame_size*st->channels, float);
michael@0 2002
michael@0 2003 for (i=0;i<frame_size*st->channels;i++)
michael@0 2004 in[i] = (1.0f/32768)*pcm[i];
michael@0 2005 ret = opus_encode_native(st, in, frame_size, data, max_data_bytes, 16, pcm, analysis_frame_size, 0, -2, st->channels, downmix_int);
michael@0 2006 RESTORE_STACK;
michael@0 2007 return ret;
michael@0 2008 }
michael@0 2009 opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int analysis_frame_size,
michael@0 2010 unsigned char *data, opus_int32 out_data_bytes)
michael@0 2011 {
michael@0 2012 int frame_size;
michael@0 2013 int delay_compensation;
michael@0 2014 if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY)
michael@0 2015 delay_compensation = 0;
michael@0 2016 else
michael@0 2017 delay_compensation = st->delay_compensation;
michael@0 2018 frame_size = compute_frame_size(pcm, analysis_frame_size,
michael@0 2019 st->variable_duration, st->channels, st->Fs, st->bitrate_bps,
michael@0 2020 delay_compensation, downmix_float, st->analysis.subframe_mem);
michael@0 2021 return opus_encode_native(st, pcm, frame_size, data, out_data_bytes, 24,
michael@0 2022 pcm, analysis_frame_size, 0, -2, st->channels, downmix_float);
michael@0 2023 }
michael@0 2024 #endif
michael@0 2025
michael@0 2026
michael@0 2027 int opus_encoder_ctl(OpusEncoder *st, int request, ...)
michael@0 2028 {
michael@0 2029 int ret;
michael@0 2030 CELTEncoder *celt_enc;
michael@0 2031 va_list ap;
michael@0 2032
michael@0 2033 ret = OPUS_OK;
michael@0 2034 va_start(ap, request);
michael@0 2035
michael@0 2036 celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset);
michael@0 2037
michael@0 2038 switch (request)
michael@0 2039 {
michael@0 2040 case OPUS_SET_APPLICATION_REQUEST:
michael@0 2041 {
michael@0 2042 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2043 if ( (value != OPUS_APPLICATION_VOIP && value != OPUS_APPLICATION_AUDIO
michael@0 2044 && value != OPUS_APPLICATION_RESTRICTED_LOWDELAY)
michael@0 2045 || (!st->first && st->application != value))
michael@0 2046 {
michael@0 2047 ret = OPUS_BAD_ARG;
michael@0 2048 break;
michael@0 2049 }
michael@0 2050 st->application = value;
michael@0 2051 }
michael@0 2052 break;
michael@0 2053 case OPUS_GET_APPLICATION_REQUEST:
michael@0 2054 {
michael@0 2055 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2056 if (!value)
michael@0 2057 {
michael@0 2058 goto bad_arg;
michael@0 2059 }
michael@0 2060 *value = st->application;
michael@0 2061 }
michael@0 2062 break;
michael@0 2063 case OPUS_SET_BITRATE_REQUEST:
michael@0 2064 {
michael@0 2065 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2066 if (value != OPUS_AUTO && value != OPUS_BITRATE_MAX)
michael@0 2067 {
michael@0 2068 if (value <= 0)
michael@0 2069 goto bad_arg;
michael@0 2070 else if (value <= 500)
michael@0 2071 value = 500;
michael@0 2072 else if (value > (opus_int32)300000*st->channels)
michael@0 2073 value = (opus_int32)300000*st->channels;
michael@0 2074 }
michael@0 2075 st->user_bitrate_bps = value;
michael@0 2076 }
michael@0 2077 break;
michael@0 2078 case OPUS_GET_BITRATE_REQUEST:
michael@0 2079 {
michael@0 2080 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2081 if (!value)
michael@0 2082 {
michael@0 2083 goto bad_arg;
michael@0 2084 }
michael@0 2085 *value = user_bitrate_to_bitrate(st, st->prev_framesize, 1276);
michael@0 2086 }
michael@0 2087 break;
michael@0 2088 case OPUS_SET_FORCE_CHANNELS_REQUEST:
michael@0 2089 {
michael@0 2090 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2091 if((value<1 || value>st->channels) && value != OPUS_AUTO)
michael@0 2092 {
michael@0 2093 goto bad_arg;
michael@0 2094 }
michael@0 2095 st->force_channels = value;
michael@0 2096 }
michael@0 2097 break;
michael@0 2098 case OPUS_GET_FORCE_CHANNELS_REQUEST:
michael@0 2099 {
michael@0 2100 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2101 if (!value)
michael@0 2102 {
michael@0 2103 goto bad_arg;
michael@0 2104 }
michael@0 2105 *value = st->force_channels;
michael@0 2106 }
michael@0 2107 break;
michael@0 2108 case OPUS_SET_MAX_BANDWIDTH_REQUEST:
michael@0 2109 {
michael@0 2110 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2111 if (value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND)
michael@0 2112 {
michael@0 2113 goto bad_arg;
michael@0 2114 }
michael@0 2115 st->max_bandwidth = value;
michael@0 2116 if (st->max_bandwidth == OPUS_BANDWIDTH_NARROWBAND) {
michael@0 2117 st->silk_mode.maxInternalSampleRate = 8000;
michael@0 2118 } else if (st->max_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) {
michael@0 2119 st->silk_mode.maxInternalSampleRate = 12000;
michael@0 2120 } else {
michael@0 2121 st->silk_mode.maxInternalSampleRate = 16000;
michael@0 2122 }
michael@0 2123 }
michael@0 2124 break;
michael@0 2125 case OPUS_GET_MAX_BANDWIDTH_REQUEST:
michael@0 2126 {
michael@0 2127 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2128 if (!value)
michael@0 2129 {
michael@0 2130 goto bad_arg;
michael@0 2131 }
michael@0 2132 *value = st->max_bandwidth;
michael@0 2133 }
michael@0 2134 break;
michael@0 2135 case OPUS_SET_BANDWIDTH_REQUEST:
michael@0 2136 {
michael@0 2137 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2138 if ((value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) && value != OPUS_AUTO)
michael@0 2139 {
michael@0 2140 goto bad_arg;
michael@0 2141 }
michael@0 2142 st->user_bandwidth = value;
michael@0 2143 if (st->user_bandwidth == OPUS_BANDWIDTH_NARROWBAND) {
michael@0 2144 st->silk_mode.maxInternalSampleRate = 8000;
michael@0 2145 } else if (st->user_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) {
michael@0 2146 st->silk_mode.maxInternalSampleRate = 12000;
michael@0 2147 } else {
michael@0 2148 st->silk_mode.maxInternalSampleRate = 16000;
michael@0 2149 }
michael@0 2150 }
michael@0 2151 break;
michael@0 2152 case OPUS_GET_BANDWIDTH_REQUEST:
michael@0 2153 {
michael@0 2154 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2155 if (!value)
michael@0 2156 {
michael@0 2157 goto bad_arg;
michael@0 2158 }
michael@0 2159 *value = st->bandwidth;
michael@0 2160 }
michael@0 2161 break;
michael@0 2162 case OPUS_SET_DTX_REQUEST:
michael@0 2163 {
michael@0 2164 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2165 if(value<0 || value>1)
michael@0 2166 {
michael@0 2167 goto bad_arg;
michael@0 2168 }
michael@0 2169 st->silk_mode.useDTX = value;
michael@0 2170 }
michael@0 2171 break;
michael@0 2172 case OPUS_GET_DTX_REQUEST:
michael@0 2173 {
michael@0 2174 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2175 if (!value)
michael@0 2176 {
michael@0 2177 goto bad_arg;
michael@0 2178 }
michael@0 2179 *value = st->silk_mode.useDTX;
michael@0 2180 }
michael@0 2181 break;
michael@0 2182 case OPUS_SET_COMPLEXITY_REQUEST:
michael@0 2183 {
michael@0 2184 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2185 if(value<0 || value>10)
michael@0 2186 {
michael@0 2187 goto bad_arg;
michael@0 2188 }
michael@0 2189 st->silk_mode.complexity = value;
michael@0 2190 celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(value));
michael@0 2191 }
michael@0 2192 break;
michael@0 2193 case OPUS_GET_COMPLEXITY_REQUEST:
michael@0 2194 {
michael@0 2195 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2196 if (!value)
michael@0 2197 {
michael@0 2198 goto bad_arg;
michael@0 2199 }
michael@0 2200 *value = st->silk_mode.complexity;
michael@0 2201 }
michael@0 2202 break;
michael@0 2203 case OPUS_SET_INBAND_FEC_REQUEST:
michael@0 2204 {
michael@0 2205 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2206 if(value<0 || value>1)
michael@0 2207 {
michael@0 2208 goto bad_arg;
michael@0 2209 }
michael@0 2210 st->silk_mode.useInBandFEC = value;
michael@0 2211 }
michael@0 2212 break;
michael@0 2213 case OPUS_GET_INBAND_FEC_REQUEST:
michael@0 2214 {
michael@0 2215 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2216 if (!value)
michael@0 2217 {
michael@0 2218 goto bad_arg;
michael@0 2219 }
michael@0 2220 *value = st->silk_mode.useInBandFEC;
michael@0 2221 }
michael@0 2222 break;
michael@0 2223 case OPUS_SET_PACKET_LOSS_PERC_REQUEST:
michael@0 2224 {
michael@0 2225 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2226 if (value < 0 || value > 100)
michael@0 2227 {
michael@0 2228 goto bad_arg;
michael@0 2229 }
michael@0 2230 st->silk_mode.packetLossPercentage = value;
michael@0 2231 celt_encoder_ctl(celt_enc, OPUS_SET_PACKET_LOSS_PERC(value));
michael@0 2232 }
michael@0 2233 break;
michael@0 2234 case OPUS_GET_PACKET_LOSS_PERC_REQUEST:
michael@0 2235 {
michael@0 2236 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2237 if (!value)
michael@0 2238 {
michael@0 2239 goto bad_arg;
michael@0 2240 }
michael@0 2241 *value = st->silk_mode.packetLossPercentage;
michael@0 2242 }
michael@0 2243 break;
michael@0 2244 case OPUS_SET_VBR_REQUEST:
michael@0 2245 {
michael@0 2246 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2247 if(value<0 || value>1)
michael@0 2248 {
michael@0 2249 goto bad_arg;
michael@0 2250 }
michael@0 2251 st->use_vbr = value;
michael@0 2252 st->silk_mode.useCBR = 1-value;
michael@0 2253 }
michael@0 2254 break;
michael@0 2255 case OPUS_GET_VBR_REQUEST:
michael@0 2256 {
michael@0 2257 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2258 if (!value)
michael@0 2259 {
michael@0 2260 goto bad_arg;
michael@0 2261 }
michael@0 2262 *value = st->use_vbr;
michael@0 2263 }
michael@0 2264 break;
michael@0 2265 case OPUS_SET_VOICE_RATIO_REQUEST:
michael@0 2266 {
michael@0 2267 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2268 if (value<-1 || value>100)
michael@0 2269 {
michael@0 2270 goto bad_arg;
michael@0 2271 }
michael@0 2272 st->voice_ratio = value;
michael@0 2273 }
michael@0 2274 break;
michael@0 2275 case OPUS_GET_VOICE_RATIO_REQUEST:
michael@0 2276 {
michael@0 2277 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2278 if (!value)
michael@0 2279 {
michael@0 2280 goto bad_arg;
michael@0 2281 }
michael@0 2282 *value = st->voice_ratio;
michael@0 2283 }
michael@0 2284 break;
michael@0 2285 case OPUS_SET_VBR_CONSTRAINT_REQUEST:
michael@0 2286 {
michael@0 2287 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2288 if(value<0 || value>1)
michael@0 2289 {
michael@0 2290 goto bad_arg;
michael@0 2291 }
michael@0 2292 st->vbr_constraint = value;
michael@0 2293 }
michael@0 2294 break;
michael@0 2295 case OPUS_GET_VBR_CONSTRAINT_REQUEST:
michael@0 2296 {
michael@0 2297 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2298 if (!value)
michael@0 2299 {
michael@0 2300 goto bad_arg;
michael@0 2301 }
michael@0 2302 *value = st->vbr_constraint;
michael@0 2303 }
michael@0 2304 break;
michael@0 2305 case OPUS_SET_SIGNAL_REQUEST:
michael@0 2306 {
michael@0 2307 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2308 if(value!=OPUS_AUTO && value!=OPUS_SIGNAL_VOICE && value!=OPUS_SIGNAL_MUSIC)
michael@0 2309 {
michael@0 2310 goto bad_arg;
michael@0 2311 }
michael@0 2312 st->signal_type = value;
michael@0 2313 }
michael@0 2314 break;
michael@0 2315 case OPUS_GET_SIGNAL_REQUEST:
michael@0 2316 {
michael@0 2317 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2318 if (!value)
michael@0 2319 {
michael@0 2320 goto bad_arg;
michael@0 2321 }
michael@0 2322 *value = st->signal_type;
michael@0 2323 }
michael@0 2324 break;
michael@0 2325 case OPUS_GET_LOOKAHEAD_REQUEST:
michael@0 2326 {
michael@0 2327 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2328 if (!value)
michael@0 2329 {
michael@0 2330 goto bad_arg;
michael@0 2331 }
michael@0 2332 *value = st->Fs/400;
michael@0 2333 if (st->application != OPUS_APPLICATION_RESTRICTED_LOWDELAY)
michael@0 2334 *value += st->delay_compensation;
michael@0 2335 }
michael@0 2336 break;
michael@0 2337 case OPUS_GET_SAMPLE_RATE_REQUEST:
michael@0 2338 {
michael@0 2339 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2340 if (!value)
michael@0 2341 {
michael@0 2342 goto bad_arg;
michael@0 2343 }
michael@0 2344 *value = st->Fs;
michael@0 2345 }
michael@0 2346 break;
michael@0 2347 case OPUS_GET_FINAL_RANGE_REQUEST:
michael@0 2348 {
michael@0 2349 opus_uint32 *value = va_arg(ap, opus_uint32*);
michael@0 2350 if (!value)
michael@0 2351 {
michael@0 2352 goto bad_arg;
michael@0 2353 }
michael@0 2354 *value = st->rangeFinal;
michael@0 2355 }
michael@0 2356 break;
michael@0 2357 case OPUS_SET_LSB_DEPTH_REQUEST:
michael@0 2358 {
michael@0 2359 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2360 if (value<8 || value>24)
michael@0 2361 {
michael@0 2362 goto bad_arg;
michael@0 2363 }
michael@0 2364 st->lsb_depth=value;
michael@0 2365 }
michael@0 2366 break;
michael@0 2367 case OPUS_GET_LSB_DEPTH_REQUEST:
michael@0 2368 {
michael@0 2369 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2370 if (!value)
michael@0 2371 {
michael@0 2372 goto bad_arg;
michael@0 2373 }
michael@0 2374 *value = st->lsb_depth;
michael@0 2375 }
michael@0 2376 break;
michael@0 2377 case OPUS_SET_EXPERT_FRAME_DURATION_REQUEST:
michael@0 2378 {
michael@0 2379 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2380 if (value != OPUS_FRAMESIZE_ARG && value != OPUS_FRAMESIZE_2_5_MS &&
michael@0 2381 value != OPUS_FRAMESIZE_5_MS && value != OPUS_FRAMESIZE_10_MS &&
michael@0 2382 value != OPUS_FRAMESIZE_20_MS && value != OPUS_FRAMESIZE_40_MS &&
michael@0 2383 value != OPUS_FRAMESIZE_60_MS && value != OPUS_FRAMESIZE_VARIABLE)
michael@0 2384 {
michael@0 2385 goto bad_arg;
michael@0 2386 }
michael@0 2387 st->variable_duration = value;
michael@0 2388 celt_encoder_ctl(celt_enc, OPUS_SET_EXPERT_FRAME_DURATION(value));
michael@0 2389 }
michael@0 2390 break;
michael@0 2391 case OPUS_GET_EXPERT_FRAME_DURATION_REQUEST:
michael@0 2392 {
michael@0 2393 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2394 if (!value)
michael@0 2395 {
michael@0 2396 goto bad_arg;
michael@0 2397 }
michael@0 2398 *value = st->variable_duration;
michael@0 2399 }
michael@0 2400 break;
michael@0 2401 case OPUS_SET_PREDICTION_DISABLED_REQUEST:
michael@0 2402 {
michael@0 2403 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2404 if (value > 1 || value < 0)
michael@0 2405 goto bad_arg;
michael@0 2406 st->silk_mode.reducedDependency = value;
michael@0 2407 }
michael@0 2408 break;
michael@0 2409 case OPUS_GET_PREDICTION_DISABLED_REQUEST:
michael@0 2410 {
michael@0 2411 opus_int32 *value = va_arg(ap, opus_int32*);
michael@0 2412 if (!value)
michael@0 2413 goto bad_arg;
michael@0 2414 *value = st->silk_mode.reducedDependency;
michael@0 2415 }
michael@0 2416 break;
michael@0 2417 case OPUS_RESET_STATE:
michael@0 2418 {
michael@0 2419 void *silk_enc;
michael@0 2420 silk_EncControlStruct dummy;
michael@0 2421 silk_enc = (char*)st+st->silk_enc_offset;
michael@0 2422
michael@0 2423 OPUS_CLEAR((char*)&st->OPUS_ENCODER_RESET_START,
michael@0 2424 sizeof(OpusEncoder)-
michael@0 2425 ((char*)&st->OPUS_ENCODER_RESET_START - (char*)st));
michael@0 2426
michael@0 2427 celt_encoder_ctl(celt_enc, OPUS_RESET_STATE);
michael@0 2428 silk_InitEncoder( silk_enc, st->arch, &dummy );
michael@0 2429 st->stream_channels = st->channels;
michael@0 2430 st->hybrid_stereo_width_Q14 = 1 << 14;
michael@0 2431 st->prev_HB_gain = Q15ONE;
michael@0 2432 st->first = 1;
michael@0 2433 st->mode = MODE_HYBRID;
michael@0 2434 st->bandwidth = OPUS_BANDWIDTH_FULLBAND;
michael@0 2435 st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 );
michael@0 2436 }
michael@0 2437 break;
michael@0 2438 case OPUS_SET_FORCE_MODE_REQUEST:
michael@0 2439 {
michael@0 2440 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2441 if ((value < MODE_SILK_ONLY || value > MODE_CELT_ONLY) && value != OPUS_AUTO)
michael@0 2442 {
michael@0 2443 goto bad_arg;
michael@0 2444 }
michael@0 2445 st->user_forced_mode = value;
michael@0 2446 }
michael@0 2447 break;
michael@0 2448 case OPUS_SET_LFE_REQUEST:
michael@0 2449 {
michael@0 2450 opus_int32 value = va_arg(ap, opus_int32);
michael@0 2451 st->lfe = value;
michael@0 2452 ret = celt_encoder_ctl(celt_enc, OPUS_SET_LFE(value));
michael@0 2453 }
michael@0 2454 break;
michael@0 2455 case OPUS_SET_ENERGY_MASK_REQUEST:
michael@0 2456 {
michael@0 2457 opus_val16 *value = va_arg(ap, opus_val16*);
michael@0 2458 st->energy_masking = value;
michael@0 2459 ret = celt_encoder_ctl(celt_enc, OPUS_SET_ENERGY_MASK(value));
michael@0 2460 }
michael@0 2461 break;
michael@0 2462
michael@0 2463 case CELT_GET_MODE_REQUEST:
michael@0 2464 {
michael@0 2465 const CELTMode ** value = va_arg(ap, const CELTMode**);
michael@0 2466 if (!value)
michael@0 2467 {
michael@0 2468 goto bad_arg;
michael@0 2469 }
michael@0 2470 ret = celt_encoder_ctl(celt_enc, CELT_GET_MODE(value));
michael@0 2471 }
michael@0 2472 break;
michael@0 2473 default:
michael@0 2474 /* fprintf(stderr, "unknown opus_encoder_ctl() request: %d", request);*/
michael@0 2475 ret = OPUS_UNIMPLEMENTED;
michael@0 2476 break;
michael@0 2477 }
michael@0 2478 va_end(ap);
michael@0 2479 return ret;
michael@0 2480 bad_arg:
michael@0 2481 va_end(ap);
michael@0 2482 return OPUS_BAD_ARG;
michael@0 2483 }
michael@0 2484
michael@0 2485 void opus_encoder_destroy(OpusEncoder *st)
michael@0 2486 {
michael@0 2487 opus_free(st);
michael@0 2488 }

mercurial