Alexandre Lision | 744f742 | 2013-09-25 11:39:37 -0400 | [diff] [blame] | 1 | /* Copyright (c) 2010-2011 Xiph.Org Foundation, Skype Limited |
| 2 | Written by Jean-Marc Valin and Koen Vos */ |
| 3 | /* |
| 4 | Redistribution and use in source and binary forms, with or without |
| 5 | modification, are permitted provided that the following conditions |
| 6 | are met: |
| 7 | |
| 8 | - Redistributions of source code must retain the above copyright |
| 9 | notice, this list of conditions and the following disclaimer. |
| 10 | |
| 11 | - Redistributions in binary form must reproduce the above copyright |
| 12 | notice, this list of conditions and the following disclaimer in the |
| 13 | documentation and/or other materials provided with the distribution. |
| 14 | |
| 15 | THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS |
| 16 | ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT |
| 17 | LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR |
| 18 | A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER |
| 19 | OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, |
| 20 | EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, |
| 21 | PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR |
| 22 | PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF |
| 23 | LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING |
| 24 | NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS |
| 25 | SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. |
| 26 | */ |
| 27 | |
| 28 | #ifdef HAVE_CONFIG_H |
| 29 | #include "config.h" |
| 30 | #endif |
| 31 | |
| 32 | #include <stdarg.h> |
| 33 | #include "celt.h" |
| 34 | #include "entenc.h" |
| 35 | #include "modes.h" |
| 36 | #include "API.h" |
| 37 | #include "stack_alloc.h" |
| 38 | #include "float_cast.h" |
| 39 | #include "opus.h" |
| 40 | #include "arch.h" |
| 41 | #include "opus_private.h" |
| 42 | #include "os_support.h" |
| 43 | |
| 44 | #include "tuning_parameters.h" |
| 45 | #ifdef FIXED_POINT |
| 46 | #include "fixed/structs_FIX.h" |
| 47 | #else |
| 48 | #include "float/structs_FLP.h" |
| 49 | #endif |
| 50 | |
| 51 | #define MAX_ENCODER_BUFFER 480 |
| 52 | |
| 53 | struct OpusEncoder { |
| 54 | int celt_enc_offset; |
| 55 | int silk_enc_offset; |
| 56 | silk_EncControlStruct silk_mode; |
| 57 | int application; |
| 58 | int channels; |
| 59 | int delay_compensation; |
| 60 | int force_channels; |
| 61 | int signal_type; |
| 62 | int user_bandwidth; |
| 63 | int max_bandwidth; |
| 64 | int user_forced_mode; |
| 65 | int voice_ratio; |
| 66 | opus_int32 Fs; |
| 67 | int use_vbr; |
| 68 | int vbr_constraint; |
| 69 | opus_int32 bitrate_bps; |
| 70 | opus_int32 user_bitrate_bps; |
| 71 | int encoder_buffer; |
| 72 | |
| 73 | #define OPUS_ENCODER_RESET_START stream_channels |
| 74 | int stream_channels; |
| 75 | opus_int16 hybrid_stereo_width_Q14; |
| 76 | opus_int32 variable_HP_smth2_Q15; |
| 77 | opus_val32 hp_mem[4]; |
| 78 | int mode; |
| 79 | int prev_mode; |
| 80 | int prev_channels; |
| 81 | int prev_framesize; |
| 82 | int bandwidth; |
| 83 | int silk_bw_switch; |
| 84 | /* Sampling rate (at the API level) */ |
| 85 | int first; |
| 86 | opus_val16 delay_buffer[MAX_ENCODER_BUFFER*2]; |
| 87 | |
| 88 | opus_uint32 rangeFinal; |
| 89 | }; |
| 90 | |
| 91 | /* Transition tables for the voice and music. First column is the |
| 92 | middle (memoriless) threshold. The second column is the hysteresis |
| 93 | (difference with the middle) */ |
| 94 | static const opus_int32 mono_voice_bandwidth_thresholds[8] = { |
| 95 | 11000, 1000, /* NB<->MB */ |
| 96 | 14000, 1000, /* MB<->WB */ |
| 97 | 21000, 2000, /* WB<->SWB */ |
| 98 | 29000, 2000, /* SWB<->FB */ |
| 99 | }; |
| 100 | static const opus_int32 mono_music_bandwidth_thresholds[8] = { |
| 101 | 14000, 1000, /* MB not allowed */ |
| 102 | 18000, 2000, /* MB<->WB */ |
| 103 | 24000, 2000, /* WB<->SWB */ |
| 104 | 33000, 2000, /* SWB<->FB */ |
| 105 | }; |
| 106 | static const opus_int32 stereo_voice_bandwidth_thresholds[8] = { |
| 107 | 11000, 1000, /* NB<->MB */ |
| 108 | 14000, 1000, /* MB<->WB */ |
| 109 | 21000, 2000, /* WB<->SWB */ |
| 110 | 32000, 2000, /* SWB<->FB */ |
| 111 | }; |
| 112 | static const opus_int32 stereo_music_bandwidth_thresholds[8] = { |
| 113 | 14000, 1000, /* MB not allowed */ |
| 114 | 18000, 2000, /* MB<->WB */ |
| 115 | 24000, 2000, /* WB<->SWB */ |
| 116 | 48000, 2000, /* SWB<->FB */ |
| 117 | }; |
| 118 | /* Threshold bit-rates for switching between mono and stereo */ |
| 119 | static const opus_int32 stereo_voice_threshold = 26000; |
| 120 | static const opus_int32 stereo_music_threshold = 36000; |
| 121 | |
| 122 | /* Threshold bit-rate for switching between SILK/hybrid and CELT-only */ |
| 123 | static const opus_int32 mode_thresholds[2][2] = { |
| 124 | /* voice */ /* music */ |
| 125 | { 48000, 24000}, /* mono */ |
| 126 | { 48000, 24000}, /* stereo */ |
| 127 | }; |
| 128 | |
| 129 | int opus_encoder_get_size(int channels) |
| 130 | { |
| 131 | int silkEncSizeBytes, celtEncSizeBytes; |
| 132 | int ret; |
| 133 | if (channels<1 || channels > 2) |
| 134 | return 0; |
| 135 | ret = silk_Get_Encoder_Size( &silkEncSizeBytes ); |
| 136 | if (ret) |
| 137 | return 0; |
| 138 | silkEncSizeBytes = align(silkEncSizeBytes); |
| 139 | celtEncSizeBytes = celt_encoder_get_size(channels); |
| 140 | return align(sizeof(OpusEncoder))+silkEncSizeBytes+celtEncSizeBytes; |
| 141 | } |
| 142 | |
| 143 | int opus_encoder_init(OpusEncoder* st, opus_int32 Fs, int channels, int application) |
| 144 | { |
| 145 | void *silk_enc; |
| 146 | CELTEncoder *celt_enc; |
| 147 | int err; |
| 148 | int ret, silkEncSizeBytes; |
| 149 | |
| 150 | if((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000)||(channels!=1&&channels!=2)|| |
| 151 | (application != OPUS_APPLICATION_VOIP && application != OPUS_APPLICATION_AUDIO |
| 152 | && application != OPUS_APPLICATION_RESTRICTED_LOWDELAY)) |
| 153 | return OPUS_BAD_ARG; |
| 154 | |
| 155 | OPUS_CLEAR((char*)st, opus_encoder_get_size(channels)); |
| 156 | /* Create SILK encoder */ |
| 157 | ret = silk_Get_Encoder_Size( &silkEncSizeBytes ); |
| 158 | if (ret) |
| 159 | return OPUS_BAD_ARG; |
| 160 | silkEncSizeBytes = align(silkEncSizeBytes); |
| 161 | st->silk_enc_offset = align(sizeof(OpusEncoder)); |
| 162 | st->celt_enc_offset = st->silk_enc_offset+silkEncSizeBytes; |
| 163 | silk_enc = (char*)st+st->silk_enc_offset; |
| 164 | celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); |
| 165 | |
| 166 | st->stream_channels = st->channels = channels; |
| 167 | |
| 168 | st->Fs = Fs; |
| 169 | |
| 170 | ret = silk_InitEncoder( silk_enc, &st->silk_mode ); |
| 171 | if(ret)return OPUS_INTERNAL_ERROR; |
| 172 | |
| 173 | /* default SILK parameters */ |
| 174 | st->silk_mode.nChannelsAPI = channels; |
| 175 | st->silk_mode.nChannelsInternal = channels; |
| 176 | st->silk_mode.API_sampleRate = st->Fs; |
| 177 | st->silk_mode.maxInternalSampleRate = 16000; |
| 178 | st->silk_mode.minInternalSampleRate = 8000; |
| 179 | st->silk_mode.desiredInternalSampleRate = 16000; |
| 180 | st->silk_mode.payloadSize_ms = 20; |
| 181 | st->silk_mode.bitRate = 25000; |
| 182 | st->silk_mode.packetLossPercentage = 0; |
| 183 | st->silk_mode.complexity = 10; |
| 184 | st->silk_mode.useInBandFEC = 0; |
| 185 | st->silk_mode.useDTX = 0; |
| 186 | st->silk_mode.useCBR = 0; |
| 187 | |
| 188 | /* Create CELT encoder */ |
| 189 | /* Initialize CELT encoder */ |
| 190 | err = celt_encoder_init(celt_enc, Fs, channels); |
| 191 | if(err!=OPUS_OK)return OPUS_INTERNAL_ERROR; |
| 192 | |
| 193 | celt_encoder_ctl(celt_enc, CELT_SET_SIGNALLING(0)); |
| 194 | celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(10)); |
| 195 | |
| 196 | st->use_vbr = 1; |
| 197 | /* Makes constrained VBR the default (safer for real-time use) */ |
| 198 | st->vbr_constraint = 1; |
| 199 | st->user_bitrate_bps = OPUS_AUTO; |
| 200 | st->bitrate_bps = 3000+Fs*channels; |
| 201 | st->application = application; |
| 202 | st->signal_type = OPUS_AUTO; |
| 203 | st->user_bandwidth = OPUS_AUTO; |
| 204 | st->max_bandwidth = OPUS_BANDWIDTH_FULLBAND; |
| 205 | st->force_channels = OPUS_AUTO; |
| 206 | st->user_forced_mode = OPUS_AUTO; |
| 207 | st->voice_ratio = -1; |
| 208 | st->encoder_buffer = st->Fs/100; |
| 209 | |
| 210 | /* Delay compensation of 4 ms (2.5 ms for SILK's extra look-ahead |
| 211 | + 1.5 ms for SILK resamplers and stereo prediction) */ |
| 212 | st->delay_compensation = st->Fs/250; |
| 213 | |
| 214 | st->hybrid_stereo_width_Q14 = 1 << 14; |
| 215 | st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); |
| 216 | st->first = 1; |
| 217 | st->mode = MODE_HYBRID; |
| 218 | st->bandwidth = OPUS_BANDWIDTH_FULLBAND; |
| 219 | |
| 220 | return OPUS_OK; |
| 221 | } |
| 222 | |
| 223 | static int pad_frame(unsigned char *data, opus_int32 len, opus_int32 new_len) |
| 224 | { |
| 225 | if (len == new_len) |
| 226 | return 0; |
| 227 | if (len > new_len) |
| 228 | return 1; |
| 229 | |
| 230 | if ((data[0]&0x3)==0) |
| 231 | { |
| 232 | int i; |
| 233 | int padding, nb_255s; |
| 234 | |
| 235 | padding = new_len - len; |
| 236 | if (padding >= 2) |
| 237 | { |
| 238 | nb_255s = (padding-2)/255; |
| 239 | |
| 240 | for (i=len-1;i>=1;i--) |
| 241 | data[i+nb_255s+2] = data[i]; |
| 242 | data[0] |= 0x3; |
| 243 | data[1] = 0x41; |
| 244 | for (i=0;i<nb_255s;i++) |
| 245 | data[i+2] = 255; |
| 246 | data[nb_255s+2] = padding-255*nb_255s-2; |
| 247 | for (i=len+3+nb_255s;i<new_len;i++) |
| 248 | data[i] = 0; |
| 249 | } else { |
| 250 | for (i=len-1;i>=1;i--) |
| 251 | data[i+1] = data[i]; |
| 252 | data[0] |= 0x3; |
| 253 | data[1] = 1; |
| 254 | } |
| 255 | return 0; |
| 256 | } else { |
| 257 | return 1; |
| 258 | } |
| 259 | } |
| 260 | |
| 261 | static unsigned char gen_toc(int mode, int framerate, int bandwidth, int channels) |
| 262 | { |
| 263 | int period; |
| 264 | unsigned char toc; |
| 265 | period = 0; |
| 266 | while (framerate < 400) |
| 267 | { |
| 268 | framerate <<= 1; |
| 269 | period++; |
| 270 | } |
| 271 | if (mode == MODE_SILK_ONLY) |
| 272 | { |
| 273 | toc = (bandwidth-OPUS_BANDWIDTH_NARROWBAND)<<5; |
| 274 | toc |= (period-2)<<3; |
| 275 | } else if (mode == MODE_CELT_ONLY) |
| 276 | { |
| 277 | int tmp = bandwidth-OPUS_BANDWIDTH_MEDIUMBAND; |
| 278 | if (tmp < 0) |
| 279 | tmp = 0; |
| 280 | toc = 0x80; |
| 281 | toc |= tmp << 5; |
| 282 | toc |= period<<3; |
| 283 | } else /* Hybrid */ |
| 284 | { |
| 285 | toc = 0x60; |
| 286 | toc |= (bandwidth-OPUS_BANDWIDTH_SUPERWIDEBAND)<<4; |
| 287 | toc |= (period-2)<<3; |
| 288 | } |
| 289 | toc |= (channels==2)<<2; |
| 290 | return toc; |
| 291 | } |
| 292 | |
| 293 | #ifndef FIXED_POINT |
| 294 | static void silk_biquad_float( |
| 295 | const opus_val16 *in, /* I: Input signal */ |
| 296 | const opus_int32 *B_Q28, /* I: MA coefficients [3] */ |
| 297 | const opus_int32 *A_Q28, /* I: AR coefficients [2] */ |
| 298 | opus_val32 *S, /* I/O: State vector [2] */ |
| 299 | opus_val16 *out, /* O: Output signal */ |
| 300 | const opus_int32 len, /* I: Signal length (must be even) */ |
| 301 | int stride |
| 302 | ) |
| 303 | { |
| 304 | /* DIRECT FORM II TRANSPOSED (uses 2 element state vector) */ |
| 305 | opus_int k; |
| 306 | opus_val32 vout; |
| 307 | opus_val32 inval; |
| 308 | opus_val32 A[2], B[3]; |
| 309 | |
| 310 | A[0] = (opus_val32)(A_Q28[0] * (1.f/((opus_int32)1<<28))); |
| 311 | A[1] = (opus_val32)(A_Q28[1] * (1.f/((opus_int32)1<<28))); |
| 312 | B[0] = (opus_val32)(B_Q28[0] * (1.f/((opus_int32)1<<28))); |
| 313 | B[1] = (opus_val32)(B_Q28[1] * (1.f/((opus_int32)1<<28))); |
| 314 | B[2] = (opus_val32)(B_Q28[2] * (1.f/((opus_int32)1<<28))); |
| 315 | |
| 316 | /* Negate A_Q28 values and split in two parts */ |
| 317 | |
| 318 | for( k = 0; k < len; k++ ) { |
| 319 | /* S[ 0 ], S[ 1 ]: Q12 */ |
| 320 | inval = in[ k*stride ]; |
| 321 | vout = S[ 0 ] + B[0]*inval; |
| 322 | |
| 323 | S[ 0 ] = S[1] - vout*A[0] + B[1]*inval; |
| 324 | |
| 325 | S[ 1 ] = - vout*A[1] + B[2]*inval; |
| 326 | |
| 327 | /* Scale back to Q0 and saturate */ |
| 328 | out[ k*stride ] = vout; |
| 329 | } |
| 330 | } |
| 331 | #endif |
| 332 | |
| 333 | 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) |
| 334 | { |
| 335 | opus_int32 B_Q28[ 3 ], A_Q28[ 2 ]; |
| 336 | opus_int32 Fc_Q19, r_Q28, r_Q22; |
| 337 | |
| 338 | silk_assert( cutoff_Hz <= silk_int32_MAX / SILK_FIX_CONST( 1.5 * 3.14159 / 1000, 19 ) ); |
| 339 | Fc_Q19 = silk_DIV32_16( silk_SMULBB( SILK_FIX_CONST( 1.5 * 3.14159 / 1000, 19 ), cutoff_Hz ), Fs/1000 ); |
| 340 | silk_assert( Fc_Q19 > 0 && Fc_Q19 < 32768 ); |
| 341 | |
| 342 | r_Q28 = SILK_FIX_CONST( 1.0, 28 ) - silk_MUL( SILK_FIX_CONST( 0.92, 9 ), Fc_Q19 ); |
| 343 | |
| 344 | /* b = r * [ 1; -2; 1 ]; */ |
| 345 | /* a = [ 1; -2 * r * ( 1 - 0.5 * Fc^2 ); r^2 ]; */ |
| 346 | B_Q28[ 0 ] = r_Q28; |
| 347 | B_Q28[ 1 ] = silk_LSHIFT( -r_Q28, 1 ); |
| 348 | B_Q28[ 2 ] = r_Q28; |
| 349 | |
| 350 | /* -r * ( 2 - Fc * Fc ); */ |
| 351 | r_Q22 = silk_RSHIFT( r_Q28, 6 ); |
| 352 | A_Q28[ 0 ] = silk_SMULWW( r_Q22, silk_SMULWW( Fc_Q19, Fc_Q19 ) - SILK_FIX_CONST( 2.0, 22 ) ); |
| 353 | A_Q28[ 1 ] = silk_SMULWW( r_Q22, r_Q22 ); |
| 354 | |
| 355 | #ifdef FIXED_POINT |
| 356 | silk_biquad_alt( in, B_Q28, A_Q28, hp_mem, out, len, channels ); |
| 357 | if( channels == 2 ) { |
| 358 | silk_biquad_alt( in+1, B_Q28, A_Q28, hp_mem+2, out+1, len, channels ); |
| 359 | } |
| 360 | #else |
| 361 | silk_biquad_float( in, B_Q28, A_Q28, hp_mem, out, len, channels ); |
| 362 | if( channels == 2 ) { |
| 363 | silk_biquad_float( in+1, B_Q28, A_Q28, hp_mem+2, out+1, len, channels ); |
| 364 | } |
| 365 | #endif |
| 366 | } |
| 367 | |
| 368 | static void stereo_fade(const opus_val16 *in, opus_val16 *out, opus_val16 g1, opus_val16 g2, |
| 369 | int overlap48, int frame_size, int channels, const opus_val16 *window, opus_int32 Fs) |
| 370 | { |
| 371 | int i; |
| 372 | int overlap; |
| 373 | int inc; |
| 374 | inc = 48000/Fs; |
| 375 | overlap=overlap48/inc; |
| 376 | g1 = Q15ONE-g1; |
| 377 | g2 = Q15ONE-g2; |
| 378 | for (i=0;i<overlap;i++) |
| 379 | { |
| 380 | opus_val32 diff; |
| 381 | opus_val16 g, w; |
| 382 | w = MULT16_16_Q15(window[i*inc], window[i*inc]); |
| 383 | g = SHR32(MAC16_16(MULT16_16(w,g2), |
| 384 | Q15ONE-w, g1), 15); |
| 385 | diff = EXTRACT16(HALF32((opus_val32)in[i*channels] - (opus_val32)in[i*channels+1])); |
| 386 | diff = MULT16_16_Q15(g, diff); |
| 387 | out[i*channels] = out[i*channels] - diff; |
| 388 | out[i*channels+1] = out[i*channels+1] + diff; |
| 389 | } |
| 390 | for (;i<frame_size;i++) |
| 391 | { |
| 392 | opus_val32 diff; |
| 393 | diff = EXTRACT16(HALF32((opus_val32)in[i*channels] - (opus_val32)in[i*channels+1])); |
| 394 | diff = MULT16_16_Q15(g2, diff); |
| 395 | out[i*channels] = out[i*channels] - diff; |
| 396 | out[i*channels+1] = out[i*channels+1] + diff; |
| 397 | } |
| 398 | } |
| 399 | |
| 400 | OpusEncoder *opus_encoder_create(opus_int32 Fs, int channels, int application, int *error) |
| 401 | { |
| 402 | int ret; |
| 403 | OpusEncoder *st; |
| 404 | if((Fs!=48000&&Fs!=24000&&Fs!=16000&&Fs!=12000&&Fs!=8000)||(channels!=1&&channels!=2)|| |
| 405 | (application != OPUS_APPLICATION_VOIP && application != OPUS_APPLICATION_AUDIO |
| 406 | && application != OPUS_APPLICATION_RESTRICTED_LOWDELAY)) |
| 407 | { |
| 408 | if (error) |
| 409 | *error = OPUS_BAD_ARG; |
| 410 | return NULL; |
| 411 | } |
| 412 | st = (OpusEncoder *)opus_alloc(opus_encoder_get_size(channels)); |
| 413 | if (st == NULL) |
| 414 | { |
| 415 | if (error) |
| 416 | *error = OPUS_ALLOC_FAIL; |
| 417 | return NULL; |
| 418 | } |
| 419 | ret = opus_encoder_init(st, Fs, channels, application); |
| 420 | if (error) |
| 421 | *error = ret; |
| 422 | if (ret != OPUS_OK) |
| 423 | { |
| 424 | opus_free(st); |
| 425 | st = NULL; |
| 426 | } |
| 427 | return st; |
| 428 | } |
| 429 | |
| 430 | static opus_int32 user_bitrate_to_bitrate(OpusEncoder *st, int frame_size, int max_data_bytes) |
| 431 | { |
| 432 | if(!frame_size)frame_size=st->Fs/400; |
| 433 | if (st->user_bitrate_bps==OPUS_AUTO) |
| 434 | return 60*st->Fs/frame_size + st->Fs*st->channels; |
| 435 | else if (st->user_bitrate_bps==OPUS_BITRATE_MAX) |
| 436 | return max_data_bytes*8*st->Fs/frame_size; |
| 437 | else |
| 438 | return st->user_bitrate_bps; |
| 439 | } |
| 440 | |
| 441 | #ifdef FIXED_POINT |
| 442 | #define opus_encode_native opus_encode |
| 443 | opus_int32 opus_encode(OpusEncoder *st, const opus_val16 *pcm, int frame_size, |
| 444 | unsigned char *data, opus_int32 out_data_bytes) |
| 445 | #else |
| 446 | #define opus_encode_native opus_encode_float |
| 447 | opus_int32 opus_encode_float(OpusEncoder *st, const opus_val16 *pcm, int frame_size, |
| 448 | unsigned char *data, opus_int32 out_data_bytes) |
| 449 | #endif |
| 450 | { |
| 451 | void *silk_enc; |
| 452 | CELTEncoder *celt_enc; |
| 453 | int i; |
| 454 | int ret=0; |
| 455 | opus_int32 nBytes; |
| 456 | ec_enc enc; |
| 457 | int bytes_target; |
| 458 | int prefill=0; |
| 459 | int start_band = 0; |
| 460 | int redundancy = 0; |
| 461 | int redundancy_bytes = 0; /* Number of bytes to use for redundancy frame */ |
| 462 | int celt_to_silk = 0; |
| 463 | VARDECL(opus_val16, pcm_buf); |
| 464 | int nb_compr_bytes; |
| 465 | int to_celt = 0; |
| 466 | opus_uint32 redundant_rng = 0; |
| 467 | int cutoff_Hz, hp_freq_smth1; |
| 468 | int voice_est; /* Probability of voice in Q7 */ |
| 469 | opus_int32 equiv_rate; |
| 470 | int delay_compensation; |
| 471 | int frame_rate; |
| 472 | opus_int32 max_rate; /* Max bitrate we're allowed to use */ |
| 473 | int curr_bandwidth; |
| 474 | opus_int32 max_data_bytes; /* Max number of bytes we're allowed to use */ |
| 475 | VARDECL(opus_val16, tmp_prefill); |
| 476 | |
| 477 | ALLOC_STACK; |
| 478 | |
| 479 | max_data_bytes = IMIN(1276, out_data_bytes); |
| 480 | |
| 481 | st->rangeFinal = 0; |
| 482 | if (400*frame_size != st->Fs && 200*frame_size != st->Fs && 100*frame_size != st->Fs && |
| 483 | 50*frame_size != st->Fs && 25*frame_size != st->Fs && 50*frame_size != 3*st->Fs) |
| 484 | { |
| 485 | RESTORE_STACK; |
| 486 | return OPUS_BAD_ARG; |
| 487 | } |
| 488 | if (max_data_bytes<=0) |
| 489 | { |
| 490 | RESTORE_STACK; |
| 491 | return OPUS_BAD_ARG; |
| 492 | } |
| 493 | silk_enc = (char*)st+st->silk_enc_offset; |
| 494 | celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); |
| 495 | |
| 496 | if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) |
| 497 | delay_compensation = 0; |
| 498 | else |
| 499 | delay_compensation = st->delay_compensation; |
| 500 | |
| 501 | st->bitrate_bps = user_bitrate_to_bitrate(st, frame_size, max_data_bytes); |
| 502 | |
| 503 | frame_rate = st->Fs/frame_size; |
| 504 | if (max_data_bytes<3 || st->bitrate_bps < 3*frame_rate*8 |
| 505 | || (frame_rate<50 && (max_data_bytes*frame_rate<300 || st->bitrate_bps < 2400))) |
| 506 | { |
| 507 | /*If the space is too low to do something useful, emit 'PLC' frames.*/ |
| 508 | int tocmode = st->mode; |
| 509 | int bw = st->bandwidth == 0 ? OPUS_BANDWIDTH_NARROWBAND : st->bandwidth; |
| 510 | if (tocmode==0) |
| 511 | tocmode = MODE_SILK_ONLY; |
| 512 | if (frame_rate>100) |
| 513 | tocmode = MODE_CELT_ONLY; |
| 514 | if (frame_rate < 50) |
| 515 | tocmode = MODE_SILK_ONLY; |
| 516 | if(tocmode==MODE_SILK_ONLY&&bw>OPUS_BANDWIDTH_WIDEBAND) |
| 517 | bw=OPUS_BANDWIDTH_WIDEBAND; |
| 518 | else if (tocmode==MODE_CELT_ONLY&&bw==OPUS_BANDWIDTH_MEDIUMBAND) |
| 519 | bw=OPUS_BANDWIDTH_NARROWBAND; |
| 520 | else if (bw<=OPUS_BANDWIDTH_SUPERWIDEBAND) |
| 521 | bw=OPUS_BANDWIDTH_SUPERWIDEBAND; |
| 522 | data[0] = gen_toc(tocmode, frame_rate, bw, st->stream_channels); |
| 523 | RESTORE_STACK; |
| 524 | return 1; |
| 525 | } |
| 526 | if (!st->use_vbr) |
| 527 | { |
| 528 | int cbrBytes; |
| 529 | cbrBytes = IMIN( (st->bitrate_bps + 4*frame_rate)/(8*frame_rate) , max_data_bytes); |
| 530 | st->bitrate_bps = cbrBytes * (8*frame_rate); |
| 531 | max_data_bytes = cbrBytes; |
| 532 | } |
| 533 | max_rate = frame_rate*max_data_bytes*8; |
| 534 | |
| 535 | /* Equivalent 20-ms rate for mode/channel/bandwidth decisions */ |
| 536 | equiv_rate = st->bitrate_bps - 60*(st->Fs/frame_size - 50); |
| 537 | |
| 538 | if (st->signal_type == OPUS_SIGNAL_VOICE) |
| 539 | voice_est = 127; |
| 540 | else if (st->signal_type == OPUS_SIGNAL_MUSIC) |
| 541 | voice_est = 0; |
| 542 | else if (st->voice_ratio >= 0) |
| 543 | voice_est = st->voice_ratio*327>>8; |
| 544 | else if (st->application == OPUS_APPLICATION_VOIP) |
| 545 | voice_est = 115; |
| 546 | else |
| 547 | voice_est = 48; |
| 548 | |
| 549 | if (st->force_channels!=OPUS_AUTO && st->channels == 2) |
| 550 | { |
| 551 | st->stream_channels = st->force_channels; |
| 552 | } else { |
| 553 | #ifdef FUZZING |
| 554 | /* Random mono/stereo decision */ |
| 555 | if (st->channels == 2 && (rand()&0x1F)==0) |
| 556 | st->stream_channels = 3-st->stream_channels; |
| 557 | #else |
| 558 | /* Rate-dependent mono-stereo decision */ |
| 559 | if (st->channels == 2) |
| 560 | { |
| 561 | opus_int32 stereo_threshold; |
| 562 | stereo_threshold = stereo_music_threshold + ((voice_est*voice_est*(stereo_voice_threshold-stereo_music_threshold))>>14); |
| 563 | if (st->stream_channels == 2) |
| 564 | stereo_threshold -= 4000; |
| 565 | else |
| 566 | stereo_threshold += 4000; |
| 567 | st->stream_channels = (equiv_rate > stereo_threshold) ? 2 : 1; |
| 568 | } else { |
| 569 | st->stream_channels = st->channels; |
| 570 | } |
| 571 | #endif |
| 572 | } |
| 573 | |
| 574 | /* Mode selection depending on application and signal type */ |
| 575 | if (st->application == OPUS_APPLICATION_RESTRICTED_LOWDELAY) |
| 576 | { |
| 577 | st->mode = MODE_CELT_ONLY; |
| 578 | } else if (st->user_forced_mode == OPUS_AUTO) |
| 579 | { |
| 580 | #ifdef FUZZING |
| 581 | /* Random mode switching */ |
| 582 | if ((rand()&0xF)==0) |
| 583 | { |
| 584 | if ((rand()&0x1)==0) |
| 585 | st->mode = MODE_CELT_ONLY; |
| 586 | else |
| 587 | st->mode = MODE_SILK_ONLY; |
| 588 | } else { |
| 589 | if (st->prev_mode==MODE_CELT_ONLY) |
| 590 | st->mode = MODE_CELT_ONLY; |
| 591 | else |
| 592 | st->mode = MODE_SILK_ONLY; |
| 593 | } |
| 594 | #else |
| 595 | int chan; |
| 596 | opus_int32 mode_voice, mode_music; |
| 597 | opus_int32 threshold; |
| 598 | |
| 599 | chan = (st->channels==2) && st->force_channels!=1; |
| 600 | mode_voice = mode_thresholds[chan][0]; |
| 601 | mode_music = mode_thresholds[chan][1]; |
| 602 | threshold = mode_music + ((voice_est*voice_est*(mode_voice-mode_music))>>14); |
| 603 | |
| 604 | /* Hysteresis */ |
| 605 | if (st->prev_mode == MODE_CELT_ONLY) |
| 606 | threshold -= 4000; |
| 607 | else if (st->prev_mode>0) |
| 608 | threshold += 4000; |
| 609 | |
| 610 | st->mode = (equiv_rate >= threshold) ? MODE_CELT_ONLY: MODE_SILK_ONLY; |
| 611 | |
| 612 | /* When FEC is enabled and there's enough packet loss, use SILK */ |
| 613 | if (st->silk_mode.useInBandFEC && st->silk_mode.packetLossPercentage > (128-voice_est)>>4) |
| 614 | st->mode = MODE_SILK_ONLY; |
| 615 | /* When encoding voice and DTX is enabled, set the encoder to SILK mode (at least for now) */ |
| 616 | if (st->silk_mode.useDTX && voice_est > 100) |
| 617 | st->mode = MODE_SILK_ONLY; |
| 618 | #endif |
| 619 | } else { |
| 620 | st->mode = st->user_forced_mode; |
| 621 | } |
| 622 | |
| 623 | /* Override the chosen mode to make sure we meet the requested frame size */ |
| 624 | if (st->mode != MODE_CELT_ONLY && frame_size < st->Fs/100) |
| 625 | st->mode = MODE_CELT_ONLY; |
| 626 | |
| 627 | if (st->stream_channels == 1 && st->prev_channels ==2 && st->silk_mode.toMono==0 |
| 628 | && st->mode != MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY) |
| 629 | { |
| 630 | /* Delay stereo->mono transition by two frames so that SILK can do a smooth downmix */ |
| 631 | st->silk_mode.toMono = 1; |
| 632 | st->stream_channels = 2; |
| 633 | } else { |
| 634 | st->silk_mode.toMono = 0; |
| 635 | } |
| 636 | |
| 637 | if (st->prev_mode > 0 && |
| 638 | ((st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) || |
| 639 | (st->mode == MODE_CELT_ONLY && st->prev_mode != MODE_CELT_ONLY))) |
| 640 | { |
| 641 | redundancy = 1; |
| 642 | celt_to_silk = (st->mode != MODE_CELT_ONLY); |
| 643 | if (!celt_to_silk) |
| 644 | { |
| 645 | /* Switch to SILK/hybrid if frame size is 10 ms or more*/ |
| 646 | if (frame_size >= st->Fs/100) |
| 647 | { |
| 648 | st->mode = st->prev_mode; |
| 649 | to_celt = 1; |
| 650 | } else { |
| 651 | redundancy=0; |
| 652 | } |
| 653 | } |
| 654 | } |
| 655 | /* For the first frame at a new SILK bandwidth */ |
| 656 | if (st->silk_bw_switch) |
| 657 | { |
| 658 | redundancy = 1; |
| 659 | celt_to_silk = 1; |
| 660 | st->silk_bw_switch = 0; |
| 661 | prefill=1; |
| 662 | } |
| 663 | |
| 664 | if (redundancy) |
| 665 | { |
| 666 | /* Fair share of the max size allowed */ |
| 667 | redundancy_bytes = IMIN(257, max_data_bytes*(opus_int32)(st->Fs/200)/(frame_size+st->Fs/200)); |
| 668 | /* For VBR, target the actual bitrate (subject to the limit above) */ |
| 669 | if (st->use_vbr) |
| 670 | redundancy_bytes = IMIN(redundancy_bytes, st->bitrate_bps/1600); |
| 671 | } |
| 672 | |
| 673 | if (st->mode != MODE_CELT_ONLY && st->prev_mode == MODE_CELT_ONLY) |
| 674 | { |
| 675 | silk_EncControlStruct dummy; |
| 676 | silk_InitEncoder( silk_enc, &dummy); |
| 677 | prefill=1; |
| 678 | } |
| 679 | |
| 680 | /* Automatic (rate-dependent) bandwidth selection */ |
| 681 | if (st->mode == MODE_CELT_ONLY || st->first || st->silk_mode.allowBandwidthSwitch) |
| 682 | { |
| 683 | const opus_int32 *voice_bandwidth_thresholds, *music_bandwidth_thresholds; |
| 684 | opus_int32 bandwidth_thresholds[8]; |
| 685 | int bandwidth = OPUS_BANDWIDTH_FULLBAND; |
| 686 | opus_int32 equiv_rate2; |
| 687 | |
| 688 | equiv_rate2 = equiv_rate; |
| 689 | if (st->mode != MODE_CELT_ONLY) |
| 690 | { |
| 691 | /* Adjust the threshold +/- 10% depending on complexity */ |
| 692 | equiv_rate2 = equiv_rate2 * (45+st->silk_mode.complexity)/50; |
| 693 | /* CBR is less efficient by ~1 kb/s */ |
| 694 | if (!st->use_vbr) |
| 695 | equiv_rate2 -= 1000; |
| 696 | } |
| 697 | if (st->channels==2 && st->force_channels!=1) |
| 698 | { |
| 699 | voice_bandwidth_thresholds = stereo_voice_bandwidth_thresholds; |
| 700 | music_bandwidth_thresholds = stereo_music_bandwidth_thresholds; |
| 701 | } else { |
| 702 | voice_bandwidth_thresholds = mono_voice_bandwidth_thresholds; |
| 703 | music_bandwidth_thresholds = mono_music_bandwidth_thresholds; |
| 704 | } |
| 705 | /* Interpolate bandwidth thresholds depending on voice estimation */ |
| 706 | for (i=0;i<8;i++) |
| 707 | { |
| 708 | bandwidth_thresholds[i] = music_bandwidth_thresholds[i] |
| 709 | + ((voice_est*voice_est*(voice_bandwidth_thresholds[i]-music_bandwidth_thresholds[i]))>>14); |
| 710 | } |
| 711 | do { |
| 712 | int threshold, hysteresis; |
| 713 | threshold = bandwidth_thresholds[2*(bandwidth-OPUS_BANDWIDTH_MEDIUMBAND)]; |
| 714 | hysteresis = bandwidth_thresholds[2*(bandwidth-OPUS_BANDWIDTH_MEDIUMBAND)+1]; |
| 715 | if (!st->first) |
| 716 | { |
| 717 | if (st->bandwidth >= bandwidth) |
| 718 | threshold -= hysteresis; |
| 719 | else |
| 720 | threshold += hysteresis; |
| 721 | } |
| 722 | if (equiv_rate2 >= threshold) |
| 723 | break; |
| 724 | } while (--bandwidth>OPUS_BANDWIDTH_NARROWBAND); |
| 725 | st->bandwidth = bandwidth; |
| 726 | /* Prevents any transition to SWB/FB until the SILK layer has fully |
| 727 | switched to WB mode and turned the variable LP filter off */ |
| 728 | if (!st->first && st->mode != MODE_CELT_ONLY && !st->silk_mode.inWBmodeWithoutVariableLP && st->bandwidth > OPUS_BANDWIDTH_WIDEBAND) |
| 729 | st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; |
| 730 | } |
| 731 | |
| 732 | if (st->bandwidth>st->max_bandwidth) |
| 733 | st->bandwidth = st->max_bandwidth; |
| 734 | |
| 735 | if (st->user_bandwidth != OPUS_AUTO) |
| 736 | st->bandwidth = st->user_bandwidth; |
| 737 | |
| 738 | /* This prevents us from using hybrid at unsafe CBR/max rates */ |
| 739 | if (st->mode != MODE_CELT_ONLY && max_rate < 15000) |
| 740 | { |
| 741 | st->bandwidth = IMIN(st->bandwidth, OPUS_BANDWIDTH_WIDEBAND); |
| 742 | } |
| 743 | |
| 744 | /* Prevents Opus from wasting bits on frequencies that are above |
| 745 | the Nyquist rate of the input signal */ |
| 746 | if (st->Fs <= 24000 && st->bandwidth > OPUS_BANDWIDTH_SUPERWIDEBAND) |
| 747 | st->bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND; |
| 748 | if (st->Fs <= 16000 && st->bandwidth > OPUS_BANDWIDTH_WIDEBAND) |
| 749 | st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; |
| 750 | if (st->Fs <= 12000 && st->bandwidth > OPUS_BANDWIDTH_MEDIUMBAND) |
| 751 | st->bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; |
| 752 | if (st->Fs <= 8000 && st->bandwidth > OPUS_BANDWIDTH_NARROWBAND) |
| 753 | st->bandwidth = OPUS_BANDWIDTH_NARROWBAND; |
| 754 | |
| 755 | /* If max_data_bytes represents less than 8 kb/s, switch to CELT-only mode */ |
| 756 | if (max_data_bytes < (frame_rate > 50 ? 12000 : 8000)*frame_size / (st->Fs * 8)) |
| 757 | st->mode = MODE_CELT_ONLY; |
| 758 | |
| 759 | /* CELT mode doesn't support mediumband, use wideband instead */ |
| 760 | if (st->mode == MODE_CELT_ONLY && st->bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) |
| 761 | st->bandwidth = OPUS_BANDWIDTH_WIDEBAND; |
| 762 | |
| 763 | /* Can't support higher than wideband for >20 ms frames */ |
| 764 | if (frame_size > st->Fs/50 && (st->mode == MODE_CELT_ONLY || st->bandwidth > OPUS_BANDWIDTH_WIDEBAND)) |
| 765 | { |
| 766 | VARDECL(unsigned char, tmp_data); |
| 767 | int nb_frames; |
| 768 | int bak_mode, bak_bandwidth, bak_channels, bak_to_mono; |
| 769 | OpusRepacketizer rp; |
| 770 | opus_int32 bytes_per_frame; |
| 771 | |
| 772 | |
| 773 | nb_frames = frame_size > st->Fs/25 ? 3 : 2; |
| 774 | bytes_per_frame = IMIN(1276,(out_data_bytes-3)/nb_frames); |
| 775 | |
| 776 | ALLOC(tmp_data, nb_frames*bytes_per_frame, unsigned char); |
| 777 | |
| 778 | opus_repacketizer_init(&rp); |
| 779 | |
| 780 | bak_mode = st->user_forced_mode; |
| 781 | bak_bandwidth = st->user_bandwidth; |
| 782 | bak_channels = st->force_channels; |
| 783 | |
| 784 | st->user_forced_mode = st->mode; |
| 785 | st->user_bandwidth = st->bandwidth; |
| 786 | st->force_channels = st->stream_channels; |
| 787 | bak_to_mono = st->silk_mode.toMono; |
| 788 | |
| 789 | if (bak_to_mono) |
| 790 | st->force_channels = 1; |
| 791 | else |
| 792 | st->prev_channels = st->stream_channels; |
| 793 | for (i=0;i<nb_frames;i++) |
| 794 | { |
| 795 | int tmp_len; |
| 796 | st->silk_mode.toMono = 0; |
| 797 | /* When switching from SILK/Hybrid to CELT, only ask for a switch at the last frame */ |
| 798 | if (to_celt && i==nb_frames-1) |
| 799 | st->user_forced_mode = MODE_CELT_ONLY; |
| 800 | tmp_len = opus_encode_native(st, pcm+i*(st->channels*st->Fs/50), st->Fs/50, tmp_data+i*bytes_per_frame, bytes_per_frame); |
| 801 | if (tmp_len<0) |
| 802 | { |
| 803 | RESTORE_STACK; |
| 804 | return OPUS_INTERNAL_ERROR; |
| 805 | } |
| 806 | ret = opus_repacketizer_cat(&rp, tmp_data+i*bytes_per_frame, tmp_len); |
| 807 | if (ret<0) |
| 808 | { |
| 809 | RESTORE_STACK; |
| 810 | return OPUS_INTERNAL_ERROR; |
| 811 | } |
| 812 | } |
| 813 | ret = opus_repacketizer_out(&rp, data, out_data_bytes); |
| 814 | if (ret<0) |
| 815 | { |
| 816 | RESTORE_STACK; |
| 817 | return OPUS_INTERNAL_ERROR; |
| 818 | } |
| 819 | st->user_forced_mode = bak_mode; |
| 820 | st->user_bandwidth = bak_bandwidth; |
| 821 | st->force_channels = bak_channels; |
| 822 | st->silk_mode.toMono = bak_to_mono; |
| 823 | RESTORE_STACK; |
| 824 | return ret; |
| 825 | } |
| 826 | |
| 827 | curr_bandwidth = st->bandwidth; |
| 828 | |
| 829 | /* Chooses the appropriate mode for speech |
| 830 | *NEVER* switch to/from CELT-only mode here as this will invalidate some assumptions */ |
| 831 | if (st->mode == MODE_SILK_ONLY && curr_bandwidth > OPUS_BANDWIDTH_WIDEBAND) |
| 832 | st->mode = MODE_HYBRID; |
| 833 | if (st->mode == MODE_HYBRID && curr_bandwidth <= OPUS_BANDWIDTH_WIDEBAND) |
| 834 | st->mode = MODE_SILK_ONLY; |
| 835 | |
| 836 | /* printf("%d %d %d %d\n", st->bitrate_bps, st->stream_channels, st->mode, curr_bandwidth); */ |
| 837 | bytes_target = IMIN(max_data_bytes-redundancy_bytes, st->bitrate_bps * frame_size / (st->Fs * 8)) - 1; |
| 838 | |
| 839 | data += 1; |
| 840 | |
| 841 | ec_enc_init(&enc, data, max_data_bytes-1); |
| 842 | |
| 843 | ALLOC(pcm_buf, (delay_compensation+frame_size)*st->channels, opus_val16); |
| 844 | for (i=0;i<delay_compensation*st->channels;i++) |
| 845 | pcm_buf[i] = st->delay_buffer[(st->encoder_buffer-delay_compensation)*st->channels+i]; |
| 846 | |
| 847 | if (st->mode == MODE_CELT_ONLY) |
| 848 | hp_freq_smth1 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); |
| 849 | else |
| 850 | hp_freq_smth1 = ((silk_encoder*)silk_enc)->state_Fxx[0].sCmn.variable_HP_smth1_Q15; |
| 851 | |
| 852 | st->variable_HP_smth2_Q15 = silk_SMLAWB( st->variable_HP_smth2_Q15, |
| 853 | hp_freq_smth1 - st->variable_HP_smth2_Q15, SILK_FIX_CONST( VARIABLE_HP_SMTH_COEF2, 16 ) ); |
| 854 | |
| 855 | /* convert from log scale to Hertz */ |
| 856 | cutoff_Hz = silk_log2lin( silk_RSHIFT( st->variable_HP_smth2_Q15, 8 ) ); |
| 857 | |
| 858 | if (st->application == OPUS_APPLICATION_VOIP) |
| 859 | { |
| 860 | hp_cutoff(pcm, cutoff_Hz, &pcm_buf[delay_compensation*st->channels], st->hp_mem, frame_size, st->channels, st->Fs); |
| 861 | } else { |
| 862 | for (i=0;i<frame_size*st->channels;i++) |
| 863 | pcm_buf[delay_compensation*st->channels + i] = pcm[i]; |
| 864 | } |
| 865 | |
| 866 | /* SILK processing */ |
| 867 | if (st->mode != MODE_CELT_ONLY) |
| 868 | { |
| 869 | #ifdef FIXED_POINT |
| 870 | const opus_int16 *pcm_silk; |
| 871 | #else |
| 872 | VARDECL(opus_int16, pcm_silk); |
| 873 | ALLOC(pcm_silk, st->channels*frame_size, opus_int16); |
| 874 | #endif |
| 875 | st->silk_mode.bitRate = 8*bytes_target*frame_rate; |
| 876 | if( st->mode == MODE_HYBRID ) { |
| 877 | st->silk_mode.bitRate /= st->stream_channels; |
| 878 | if( curr_bandwidth == OPUS_BANDWIDTH_SUPERWIDEBAND ) { |
| 879 | if( st->Fs == 100 * frame_size ) { |
| 880 | /* 24 kHz, 10 ms */ |
| 881 | st->silk_mode.bitRate = ( ( st->silk_mode.bitRate + 2000 + st->use_vbr * 1000 ) * 2 ) / 3; |
| 882 | } else { |
| 883 | /* 24 kHz, 20 ms */ |
| 884 | st->silk_mode.bitRate = ( ( st->silk_mode.bitRate + 1000 + st->use_vbr * 1000 ) * 2 ) / 3; |
| 885 | } |
| 886 | } else { |
| 887 | if( st->Fs == 100 * frame_size ) { |
| 888 | /* 48 kHz, 10 ms */ |
| 889 | st->silk_mode.bitRate = ( st->silk_mode.bitRate + 8000 + st->use_vbr * 3000 ) / 2; |
| 890 | } else { |
| 891 | /* 48 kHz, 20 ms */ |
| 892 | st->silk_mode.bitRate = ( st->silk_mode.bitRate + 9000 + st->use_vbr * 1000 ) / 2; |
| 893 | } |
| 894 | } |
| 895 | st->silk_mode.bitRate *= st->stream_channels; |
| 896 | /* don't let SILK use more than 80% */ |
| 897 | if( st->silk_mode.bitRate > ( st->bitrate_bps - 8*st->Fs/frame_size ) * 4/5 ) { |
| 898 | st->silk_mode.bitRate = ( st->bitrate_bps - 8*st->Fs/frame_size ) * 4/5; |
| 899 | } |
| 900 | } |
| 901 | |
| 902 | st->silk_mode.payloadSize_ms = 1000 * frame_size / st->Fs; |
| 903 | st->silk_mode.nChannelsAPI = st->channels; |
| 904 | st->silk_mode.nChannelsInternal = st->stream_channels; |
| 905 | if (curr_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { |
| 906 | st->silk_mode.desiredInternalSampleRate = 8000; |
| 907 | } else if (curr_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { |
| 908 | st->silk_mode.desiredInternalSampleRate = 12000; |
| 909 | } else { |
| 910 | silk_assert( st->mode == MODE_HYBRID || curr_bandwidth == OPUS_BANDWIDTH_WIDEBAND ); |
| 911 | st->silk_mode.desiredInternalSampleRate = 16000; |
| 912 | } |
| 913 | if( st->mode == MODE_HYBRID ) { |
| 914 | /* Don't allow bandwidth reduction at lowest bitrates in hybrid mode */ |
| 915 | st->silk_mode.minInternalSampleRate = 16000; |
| 916 | } else { |
| 917 | st->silk_mode.minInternalSampleRate = 8000; |
| 918 | } |
| 919 | |
| 920 | if (st->mode == MODE_SILK_ONLY) |
| 921 | { |
| 922 | opus_int32 effective_max_rate = max_rate; |
| 923 | st->silk_mode.maxInternalSampleRate = 16000; |
| 924 | if (frame_rate > 50) |
| 925 | effective_max_rate = effective_max_rate*2/3; |
| 926 | if (effective_max_rate < 13000) |
| 927 | { |
| 928 | st->silk_mode.maxInternalSampleRate = 12000; |
| 929 | st->silk_mode.desiredInternalSampleRate = IMIN(12000, st->silk_mode.desiredInternalSampleRate); |
| 930 | } |
| 931 | if (effective_max_rate < 9600) |
| 932 | { |
| 933 | st->silk_mode.maxInternalSampleRate = 8000; |
| 934 | st->silk_mode.desiredInternalSampleRate = IMIN(8000, st->silk_mode.desiredInternalSampleRate); |
| 935 | } |
| 936 | } else { |
| 937 | st->silk_mode.maxInternalSampleRate = 16000; |
| 938 | } |
| 939 | |
| 940 | st->silk_mode.useCBR = !st->use_vbr; |
| 941 | |
| 942 | /* Call SILK encoder for the low band */ |
| 943 | nBytes = IMIN(1275, max_data_bytes-1-redundancy_bytes); |
| 944 | |
| 945 | st->silk_mode.maxBits = nBytes*8; |
| 946 | /* Only allow up to 90% of the bits for hybrid mode*/ |
| 947 | if (st->mode == MODE_HYBRID) |
| 948 | st->silk_mode.maxBits = (opus_int32)st->silk_mode.maxBits*9/10; |
| 949 | if (st->silk_mode.useCBR) |
| 950 | { |
| 951 | st->silk_mode.maxBits = (st->silk_mode.bitRate * frame_size / (st->Fs * 8))*8; |
| 952 | /* Reduce the initial target to make it easier to reach the CBR rate */ |
| 953 | st->silk_mode.bitRate = IMAX(1, st->silk_mode.bitRate-2000); |
| 954 | } |
| 955 | |
| 956 | if (prefill) |
| 957 | { |
| 958 | opus_int32 zero=0; |
| 959 | #ifdef FIXED_POINT |
| 960 | pcm_silk = st->delay_buffer; |
| 961 | #else |
| 962 | for (i=0;i<st->encoder_buffer*st->channels;i++) |
| 963 | pcm_silk[i] = FLOAT2INT16(st->delay_buffer[i]); |
| 964 | #endif |
| 965 | silk_Encode( silk_enc, &st->silk_mode, pcm_silk, st->encoder_buffer, NULL, &zero, 1 ); |
| 966 | } |
| 967 | |
| 968 | #ifdef FIXED_POINT |
| 969 | pcm_silk = pcm_buf+delay_compensation*st->channels; |
| 970 | #else |
| 971 | for (i=0;i<frame_size*st->channels;i++) |
| 972 | pcm_silk[i] = FLOAT2INT16(pcm_buf[delay_compensation*st->channels + i]); |
| 973 | #endif |
| 974 | ret = silk_Encode( silk_enc, &st->silk_mode, pcm_silk, frame_size, &enc, &nBytes, 0 ); |
| 975 | if( ret ) { |
| 976 | /*fprintf (stderr, "SILK encode error: %d\n", ret);*/ |
| 977 | /* Handle error */ |
| 978 | RESTORE_STACK; |
| 979 | return OPUS_INTERNAL_ERROR; |
| 980 | } |
| 981 | if (nBytes==0) |
| 982 | { |
| 983 | st->rangeFinal = 0; |
| 984 | data[-1] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); |
| 985 | RESTORE_STACK; |
| 986 | return 1; |
| 987 | } |
| 988 | /* Extract SILK internal bandwidth for signaling in first byte */ |
| 989 | if( st->mode == MODE_SILK_ONLY ) { |
| 990 | if( st->silk_mode.internalSampleRate == 8000 ) { |
| 991 | curr_bandwidth = OPUS_BANDWIDTH_NARROWBAND; |
| 992 | } else if( st->silk_mode.internalSampleRate == 12000 ) { |
| 993 | curr_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND; |
| 994 | } else if( st->silk_mode.internalSampleRate == 16000 ) { |
| 995 | curr_bandwidth = OPUS_BANDWIDTH_WIDEBAND; |
| 996 | } |
| 997 | } else { |
| 998 | silk_assert( st->silk_mode.internalSampleRate == 16000 ); |
| 999 | } |
| 1000 | |
| 1001 | st->silk_mode.opusCanSwitch = st->silk_mode.switchReady; |
| 1002 | /* FIXME: How do we allocate the redundancy for CBR? */ |
| 1003 | if (st->silk_mode.opusCanSwitch) |
| 1004 | { |
| 1005 | redundancy = 1; |
| 1006 | celt_to_silk = 0; |
| 1007 | st->silk_bw_switch = 1; |
| 1008 | } |
| 1009 | } |
| 1010 | |
| 1011 | /* CELT processing */ |
| 1012 | { |
| 1013 | int endband=21; |
| 1014 | |
| 1015 | switch(curr_bandwidth) |
| 1016 | { |
| 1017 | case OPUS_BANDWIDTH_NARROWBAND: |
| 1018 | endband = 13; |
| 1019 | break; |
| 1020 | case OPUS_BANDWIDTH_MEDIUMBAND: |
| 1021 | case OPUS_BANDWIDTH_WIDEBAND: |
| 1022 | endband = 17; |
| 1023 | break; |
| 1024 | case OPUS_BANDWIDTH_SUPERWIDEBAND: |
| 1025 | endband = 19; |
| 1026 | break; |
| 1027 | case OPUS_BANDWIDTH_FULLBAND: |
| 1028 | endband = 21; |
| 1029 | break; |
| 1030 | } |
| 1031 | celt_encoder_ctl(celt_enc, CELT_SET_END_BAND(endband)); |
| 1032 | celt_encoder_ctl(celt_enc, CELT_SET_CHANNELS(st->stream_channels)); |
| 1033 | } |
| 1034 | celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(OPUS_BITRATE_MAX)); |
| 1035 | if (st->mode != MODE_SILK_ONLY) |
| 1036 | { |
| 1037 | celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); |
| 1038 | /* Allow prediction unless we decide to disable it later */ |
| 1039 | celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(2)); |
| 1040 | |
| 1041 | if (st->mode == MODE_HYBRID) |
| 1042 | { |
| 1043 | int len; |
| 1044 | |
| 1045 | len = (ec_tell(&enc)+7)>>3; |
| 1046 | if (redundancy) |
| 1047 | len += st->mode == MODE_HYBRID ? 3 : 1; |
| 1048 | if( st->use_vbr ) { |
| 1049 | nb_compr_bytes = len + bytes_target - (st->silk_mode.bitRate * frame_size) / (8 * st->Fs); |
| 1050 | } else { |
| 1051 | /* check if SILK used up too much */ |
| 1052 | nb_compr_bytes = len > bytes_target ? len : bytes_target; |
| 1053 | } |
| 1054 | } else { |
| 1055 | if (st->use_vbr) |
| 1056 | { |
| 1057 | celt_encoder_ctl(celt_enc, OPUS_SET_VBR(1)); |
| 1058 | celt_encoder_ctl(celt_enc, OPUS_SET_VBR_CONSTRAINT(st->vbr_constraint)); |
| 1059 | celt_encoder_ctl(celt_enc, OPUS_SET_BITRATE(st->bitrate_bps)); |
| 1060 | nb_compr_bytes = max_data_bytes-1-redundancy_bytes; |
| 1061 | } else { |
| 1062 | nb_compr_bytes = bytes_target; |
| 1063 | } |
| 1064 | } |
| 1065 | |
| 1066 | } else { |
| 1067 | nb_compr_bytes = 0; |
| 1068 | } |
| 1069 | |
| 1070 | ALLOC(tmp_prefill, st->channels*st->Fs/400, opus_val16); |
| 1071 | if (st->mode != MODE_SILK_ONLY && st->mode != st->prev_mode && st->prev_mode > 0) |
| 1072 | { |
| 1073 | for (i=0;i<st->channels*st->Fs/400;i++) |
| 1074 | tmp_prefill[i] = st->delay_buffer[(st->encoder_buffer-st->delay_compensation-st->Fs/400)*st->channels + i]; |
| 1075 | } |
| 1076 | |
| 1077 | for (i=0;i<st->channels*(st->encoder_buffer-(frame_size+delay_compensation));i++) |
| 1078 | st->delay_buffer[i] = st->delay_buffer[i+st->channels*frame_size]; |
| 1079 | for (;i<st->encoder_buffer*st->channels;i++) |
| 1080 | st->delay_buffer[i] = pcm_buf[(frame_size+delay_compensation-st->encoder_buffer)*st->channels+i]; |
| 1081 | |
| 1082 | |
| 1083 | if (st->mode != MODE_HYBRID || st->stream_channels==1) |
| 1084 | st->silk_mode.stereoWidth_Q14 = 1<<14; |
| 1085 | if( st->channels == 2 ) { |
| 1086 | /* Apply stereo width reduction (at low bitrates) */ |
| 1087 | if( st->hybrid_stereo_width_Q14 < (1 << 14) || st->silk_mode.stereoWidth_Q14 < (1 << 14) ) { |
| 1088 | opus_val16 g1, g2; |
| 1089 | const CELTMode *celt_mode; |
| 1090 | |
| 1091 | celt_encoder_ctl(celt_enc, CELT_GET_MODE(&celt_mode)); |
| 1092 | g1 = st->hybrid_stereo_width_Q14; |
| 1093 | g2 = (opus_val16)(st->silk_mode.stereoWidth_Q14); |
| 1094 | #ifdef FIXED_POINT |
| 1095 | g1 = g1==16384 ? Q15ONE : SHL16(g1,1); |
| 1096 | g2 = g2==16384 ? Q15ONE : SHL16(g2,1); |
| 1097 | #else |
| 1098 | g1 *= (1.f/16384); |
| 1099 | g2 *= (1.f/16384); |
| 1100 | #endif |
| 1101 | stereo_fade(pcm_buf, pcm_buf, g1, g2, celt_mode->overlap, |
| 1102 | frame_size, st->channels, celt_mode->window, st->Fs); |
| 1103 | st->hybrid_stereo_width_Q14 = st->silk_mode.stereoWidth_Q14; |
| 1104 | } |
| 1105 | } |
| 1106 | |
| 1107 | if ( st->mode != MODE_CELT_ONLY && ec_tell(&enc)+17+20*(st->mode == MODE_HYBRID) <= 8*(max_data_bytes-1)) |
| 1108 | { |
| 1109 | /* For SILK mode, the redundancy is inferred from the length */ |
| 1110 | if (st->mode == MODE_HYBRID && (redundancy || ec_tell(&enc)+37 <= 8*nb_compr_bytes)) |
| 1111 | ec_enc_bit_logp(&enc, redundancy, 12); |
| 1112 | if (redundancy) |
| 1113 | { |
| 1114 | int max_redundancy; |
| 1115 | ec_enc_bit_logp(&enc, celt_to_silk, 1); |
| 1116 | if (st->mode == MODE_HYBRID) |
| 1117 | max_redundancy = (max_data_bytes-1)-nb_compr_bytes-1; |
| 1118 | else |
| 1119 | max_redundancy = (max_data_bytes-1)-((ec_tell(&enc)+7)>>3); |
| 1120 | /* Target the same bit-rate for redundancy as for the rest, |
| 1121 | up to a max of 257 bytes */ |
| 1122 | redundancy_bytes = IMIN(max_redundancy, st->bitrate_bps/1600); |
| 1123 | redundancy_bytes = IMIN(257, IMAX(2, redundancy_bytes)); |
| 1124 | if (st->mode == MODE_HYBRID) |
| 1125 | ec_enc_uint(&enc, redundancy_bytes-2, 256); |
| 1126 | } |
| 1127 | } else { |
| 1128 | redundancy = 0; |
| 1129 | } |
| 1130 | |
| 1131 | if (!redundancy) |
| 1132 | { |
| 1133 | st->silk_bw_switch = 0; |
| 1134 | redundancy_bytes = 0; |
| 1135 | } |
| 1136 | if (st->mode != MODE_CELT_ONLY)start_band=17; |
| 1137 | |
| 1138 | if (st->mode == MODE_SILK_ONLY) |
| 1139 | { |
| 1140 | ret = (ec_tell(&enc)+7)>>3; |
| 1141 | ec_enc_done(&enc); |
| 1142 | nb_compr_bytes = ret; |
| 1143 | } else { |
| 1144 | nb_compr_bytes = IMIN((max_data_bytes-1)-redundancy_bytes, nb_compr_bytes); |
| 1145 | ec_enc_shrink(&enc, nb_compr_bytes); |
| 1146 | } |
| 1147 | |
| 1148 | |
| 1149 | /* 5 ms redundant frame for CELT->SILK */ |
| 1150 | if (redundancy && celt_to_silk) |
| 1151 | { |
| 1152 | int err; |
| 1153 | celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0)); |
| 1154 | celt_encoder_ctl(celt_enc, OPUS_SET_VBR(0)); |
| 1155 | err = celt_encode_with_ec(celt_enc, pcm_buf, st->Fs/200, data+nb_compr_bytes, redundancy_bytes, NULL); |
| 1156 | if (err < 0) |
| 1157 | { |
| 1158 | RESTORE_STACK; |
| 1159 | return OPUS_INTERNAL_ERROR; |
| 1160 | } |
| 1161 | celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&redundant_rng)); |
| 1162 | celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); |
| 1163 | } |
| 1164 | |
| 1165 | celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(start_band)); |
| 1166 | |
| 1167 | if (st->mode != MODE_SILK_ONLY) |
| 1168 | { |
| 1169 | if (st->mode != st->prev_mode && st->prev_mode > 0) |
| 1170 | { |
| 1171 | unsigned char dummy[2]; |
| 1172 | celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); |
| 1173 | |
| 1174 | /* Prefilling */ |
| 1175 | celt_encode_with_ec(celt_enc, tmp_prefill, st->Fs/400, dummy, 2, NULL); |
| 1176 | celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0)); |
| 1177 | } |
| 1178 | /* If false, we already busted the budget and we'll end up with a "PLC packet" */ |
| 1179 | if (ec_tell(&enc) <= 8*nb_compr_bytes) |
| 1180 | { |
| 1181 | ret = celt_encode_with_ec(celt_enc, pcm_buf, frame_size, NULL, nb_compr_bytes, &enc); |
| 1182 | if (ret < 0) |
| 1183 | { |
| 1184 | RESTORE_STACK; |
| 1185 | return OPUS_INTERNAL_ERROR; |
| 1186 | } |
| 1187 | } |
| 1188 | } |
| 1189 | |
| 1190 | /* 5 ms redundant frame for SILK->CELT */ |
| 1191 | if (redundancy && !celt_to_silk) |
| 1192 | { |
| 1193 | int err; |
| 1194 | unsigned char dummy[2]; |
| 1195 | int N2, N4; |
| 1196 | N2 = st->Fs/200; |
| 1197 | N4 = st->Fs/400; |
| 1198 | |
| 1199 | celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); |
| 1200 | celt_encoder_ctl(celt_enc, CELT_SET_START_BAND(0)); |
| 1201 | celt_encoder_ctl(celt_enc, CELT_SET_PREDICTION(0)); |
| 1202 | |
| 1203 | /* NOTE: We could speed this up slightly (at the expense of code size) by just adding a function that prefills the buffer */ |
| 1204 | celt_encode_with_ec(celt_enc, pcm_buf+st->channels*(frame_size-N2-N4), N4, dummy, 2, NULL); |
| 1205 | |
| 1206 | err = celt_encode_with_ec(celt_enc, pcm_buf+st->channels*(frame_size-N2), N2, data+nb_compr_bytes, redundancy_bytes, NULL); |
| 1207 | if (err < 0) |
| 1208 | { |
| 1209 | RESTORE_STACK; |
| 1210 | return OPUS_INTERNAL_ERROR; |
| 1211 | } |
| 1212 | celt_encoder_ctl(celt_enc, OPUS_GET_FINAL_RANGE(&redundant_rng)); |
| 1213 | } |
| 1214 | |
| 1215 | |
| 1216 | |
| 1217 | /* Signalling the mode in the first byte */ |
| 1218 | data--; |
| 1219 | data[0] = gen_toc(st->mode, st->Fs/frame_size, curr_bandwidth, st->stream_channels); |
| 1220 | |
| 1221 | st->rangeFinal = enc.rng ^ redundant_rng; |
| 1222 | |
| 1223 | if (to_celt) |
| 1224 | st->prev_mode = MODE_CELT_ONLY; |
| 1225 | else |
| 1226 | st->prev_mode = st->mode; |
| 1227 | st->prev_channels = st->stream_channels; |
| 1228 | st->prev_framesize = frame_size; |
| 1229 | |
| 1230 | st->first = 0; |
| 1231 | |
| 1232 | /* In the unlikely case that the SILK encoder busted its target, tell |
| 1233 | the decoder to call the PLC */ |
| 1234 | if (ec_tell(&enc) > (max_data_bytes-1)*8) |
| 1235 | { |
| 1236 | if (max_data_bytes < 2) |
| 1237 | { |
| 1238 | RESTORE_STACK; |
| 1239 | return OPUS_BUFFER_TOO_SMALL; |
| 1240 | } |
| 1241 | data[1] = 0; |
| 1242 | ret = 1; |
| 1243 | st->rangeFinal = 0; |
| 1244 | } else if (st->mode==MODE_SILK_ONLY&&!redundancy) |
| 1245 | { |
| 1246 | /*When in LPC only mode it's perfectly |
| 1247 | reasonable to strip off trailing zero bytes as |
| 1248 | the required range decoder behavior is to |
| 1249 | fill these in. This can't be done when the MDCT |
| 1250 | modes are used because the decoder needs to know |
| 1251 | the actual length for allocation purposes.*/ |
| 1252 | while(ret>2&&data[ret]==0)ret--; |
| 1253 | } |
| 1254 | /* Count ToC and redundancy */ |
| 1255 | ret += 1+redundancy_bytes; |
| 1256 | if (!st->use_vbr && ret >= 3) |
| 1257 | { |
| 1258 | if (pad_frame(data, ret, max_data_bytes)) |
| 1259 | { |
| 1260 | RESTORE_STACK; |
| 1261 | return OPUS_INTERNAL_ERROR; |
| 1262 | } |
| 1263 | ret = max_data_bytes; |
| 1264 | } |
| 1265 | RESTORE_STACK; |
| 1266 | return ret; |
| 1267 | } |
| 1268 | |
| 1269 | #ifdef FIXED_POINT |
| 1270 | |
| 1271 | #ifndef DISABLE_FLOAT_API |
| 1272 | opus_int32 opus_encode_float(OpusEncoder *st, const float *pcm, int frame_size, |
| 1273 | unsigned char *data, opus_int32 max_data_bytes) |
| 1274 | { |
| 1275 | int i, ret; |
| 1276 | VARDECL(opus_int16, in); |
| 1277 | ALLOC_STACK; |
| 1278 | |
| 1279 | if(frame_size<0) |
| 1280 | { |
| 1281 | RESTORE_STACK; |
| 1282 | return OPUS_BAD_ARG; |
| 1283 | } |
| 1284 | |
| 1285 | ALLOC(in, frame_size*st->channels, opus_int16); |
| 1286 | |
| 1287 | for (i=0;i<frame_size*st->channels;i++) |
| 1288 | in[i] = FLOAT2INT16(pcm[i]); |
| 1289 | ret = opus_encode(st, in, frame_size, data, max_data_bytes); |
| 1290 | RESTORE_STACK; |
| 1291 | return ret; |
| 1292 | } |
| 1293 | #endif |
| 1294 | |
| 1295 | #else |
| 1296 | opus_int32 opus_encode(OpusEncoder *st, const opus_int16 *pcm, int frame_size, |
| 1297 | unsigned char *data, opus_int32 max_data_bytes) |
| 1298 | { |
| 1299 | int i, ret; |
| 1300 | VARDECL(float, in); |
| 1301 | ALLOC_STACK; |
| 1302 | |
| 1303 | ALLOC(in, frame_size*st->channels, float); |
| 1304 | |
| 1305 | for (i=0;i<frame_size*st->channels;i++) |
| 1306 | in[i] = (1.0f/32768)*pcm[i]; |
| 1307 | ret = opus_encode_float(st, in, frame_size, data, max_data_bytes); |
| 1308 | RESTORE_STACK; |
| 1309 | return ret; |
| 1310 | } |
| 1311 | #endif |
| 1312 | |
| 1313 | |
| 1314 | int opus_encoder_ctl(OpusEncoder *st, int request, ...) |
| 1315 | { |
| 1316 | int ret; |
| 1317 | CELTEncoder *celt_enc; |
| 1318 | va_list ap; |
| 1319 | |
| 1320 | ret = OPUS_OK; |
| 1321 | va_start(ap, request); |
| 1322 | |
| 1323 | celt_enc = (CELTEncoder*)((char*)st+st->celt_enc_offset); |
| 1324 | |
| 1325 | switch (request) |
| 1326 | { |
| 1327 | case OPUS_SET_APPLICATION_REQUEST: |
| 1328 | { |
| 1329 | opus_int32 value = va_arg(ap, opus_int32); |
| 1330 | if ( (value != OPUS_APPLICATION_VOIP && value != OPUS_APPLICATION_AUDIO |
| 1331 | && value != OPUS_APPLICATION_RESTRICTED_LOWDELAY) |
| 1332 | || (!st->first && st->application != value)) |
| 1333 | { |
| 1334 | ret = OPUS_BAD_ARG; |
| 1335 | break; |
| 1336 | } |
| 1337 | st->application = value; |
| 1338 | } |
| 1339 | break; |
| 1340 | case OPUS_GET_APPLICATION_REQUEST: |
| 1341 | { |
| 1342 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1343 | *value = st->application; |
| 1344 | } |
| 1345 | break; |
| 1346 | case OPUS_SET_BITRATE_REQUEST: |
| 1347 | { |
| 1348 | opus_int32 value = va_arg(ap, opus_int32); |
| 1349 | if (value != OPUS_AUTO && value != OPUS_BITRATE_MAX) |
| 1350 | { |
| 1351 | if (value <= 0) |
| 1352 | goto bad_arg; |
| 1353 | else if (value <= 500) |
| 1354 | value = 500; |
| 1355 | else if (value > (opus_int32)300000*st->channels) |
| 1356 | value = (opus_int32)300000*st->channels; |
| 1357 | } |
| 1358 | st->user_bitrate_bps = value; |
| 1359 | } |
| 1360 | break; |
| 1361 | case OPUS_GET_BITRATE_REQUEST: |
| 1362 | { |
| 1363 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1364 | *value = user_bitrate_to_bitrate(st, st->prev_framesize, 1276); |
| 1365 | } |
| 1366 | break; |
| 1367 | case OPUS_SET_FORCE_CHANNELS_REQUEST: |
| 1368 | { |
| 1369 | opus_int32 value = va_arg(ap, opus_int32); |
| 1370 | if((value<1 || value>st->channels) && value != OPUS_AUTO) |
| 1371 | return OPUS_BAD_ARG; |
| 1372 | st->force_channels = value; |
| 1373 | } |
| 1374 | break; |
| 1375 | case OPUS_GET_FORCE_CHANNELS_REQUEST: |
| 1376 | { |
| 1377 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1378 | *value = st->force_channels; |
| 1379 | } |
| 1380 | break; |
| 1381 | case OPUS_SET_MAX_BANDWIDTH_REQUEST: |
| 1382 | { |
| 1383 | opus_int32 value = va_arg(ap, opus_int32); |
| 1384 | if (value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) |
| 1385 | return OPUS_BAD_ARG; |
| 1386 | st->max_bandwidth = value; |
| 1387 | if (st->max_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { |
| 1388 | st->silk_mode.maxInternalSampleRate = 8000; |
| 1389 | } else if (st->max_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { |
| 1390 | st->silk_mode.maxInternalSampleRate = 12000; |
| 1391 | } else { |
| 1392 | st->silk_mode.maxInternalSampleRate = 16000; |
| 1393 | } |
| 1394 | } |
| 1395 | break; |
| 1396 | case OPUS_GET_MAX_BANDWIDTH_REQUEST: |
| 1397 | { |
| 1398 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1399 | *value = st->max_bandwidth; |
| 1400 | } |
| 1401 | break; |
| 1402 | case OPUS_SET_BANDWIDTH_REQUEST: |
| 1403 | { |
| 1404 | opus_int32 value = va_arg(ap, opus_int32); |
| 1405 | if ((value < OPUS_BANDWIDTH_NARROWBAND || value > OPUS_BANDWIDTH_FULLBAND) && value != OPUS_AUTO) |
| 1406 | return OPUS_BAD_ARG; |
| 1407 | st->user_bandwidth = value; |
| 1408 | if (st->user_bandwidth == OPUS_BANDWIDTH_NARROWBAND) { |
| 1409 | st->silk_mode.maxInternalSampleRate = 8000; |
| 1410 | } else if (st->user_bandwidth == OPUS_BANDWIDTH_MEDIUMBAND) { |
| 1411 | st->silk_mode.maxInternalSampleRate = 12000; |
| 1412 | } else { |
| 1413 | st->silk_mode.maxInternalSampleRate = 16000; |
| 1414 | } |
| 1415 | } |
| 1416 | break; |
| 1417 | case OPUS_GET_BANDWIDTH_REQUEST: |
| 1418 | { |
| 1419 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1420 | *value = st->bandwidth; |
| 1421 | } |
| 1422 | break; |
| 1423 | case OPUS_SET_DTX_REQUEST: |
| 1424 | { |
| 1425 | opus_int32 value = va_arg(ap, opus_int32); |
| 1426 | if(value<0 || value>1) |
| 1427 | return OPUS_BAD_ARG; |
| 1428 | st->silk_mode.useDTX = value; |
| 1429 | } |
| 1430 | break; |
| 1431 | case OPUS_GET_DTX_REQUEST: |
| 1432 | { |
| 1433 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1434 | *value = st->silk_mode.useDTX; |
| 1435 | } |
| 1436 | break; |
| 1437 | case OPUS_SET_COMPLEXITY_REQUEST: |
| 1438 | { |
| 1439 | opus_int32 value = va_arg(ap, opus_int32); |
| 1440 | if(value<0 || value>10) |
| 1441 | return OPUS_BAD_ARG; |
| 1442 | st->silk_mode.complexity = value; |
| 1443 | celt_encoder_ctl(celt_enc, OPUS_SET_COMPLEXITY(value)); |
| 1444 | } |
| 1445 | break; |
| 1446 | case OPUS_GET_COMPLEXITY_REQUEST: |
| 1447 | { |
| 1448 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1449 | *value = st->silk_mode.complexity; |
| 1450 | } |
| 1451 | break; |
| 1452 | case OPUS_SET_INBAND_FEC_REQUEST: |
| 1453 | { |
| 1454 | opus_int32 value = va_arg(ap, opus_int32); |
| 1455 | if(value<0 || value>1) |
| 1456 | return OPUS_BAD_ARG; |
| 1457 | st->silk_mode.useInBandFEC = value; |
| 1458 | } |
| 1459 | break; |
| 1460 | case OPUS_GET_INBAND_FEC_REQUEST: |
| 1461 | { |
| 1462 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1463 | *value = st->silk_mode.useInBandFEC; |
| 1464 | } |
| 1465 | break; |
| 1466 | case OPUS_SET_PACKET_LOSS_PERC_REQUEST: |
| 1467 | { |
| 1468 | opus_int32 value = va_arg(ap, opus_int32); |
| 1469 | if (value < 0 || value > 100) |
| 1470 | return OPUS_BAD_ARG; |
| 1471 | st->silk_mode.packetLossPercentage = value; |
| 1472 | celt_encoder_ctl(celt_enc, OPUS_SET_PACKET_LOSS_PERC(value)); |
| 1473 | } |
| 1474 | break; |
| 1475 | case OPUS_GET_PACKET_LOSS_PERC_REQUEST: |
| 1476 | { |
| 1477 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1478 | *value = st->silk_mode.packetLossPercentage; |
| 1479 | } |
| 1480 | break; |
| 1481 | case OPUS_SET_VBR_REQUEST: |
| 1482 | { |
| 1483 | opus_int32 value = va_arg(ap, opus_int32); |
| 1484 | if(value<0 || value>1) |
| 1485 | return OPUS_BAD_ARG; |
| 1486 | st->use_vbr = value; |
| 1487 | st->silk_mode.useCBR = 1-value; |
| 1488 | } |
| 1489 | break; |
| 1490 | case OPUS_GET_VBR_REQUEST: |
| 1491 | { |
| 1492 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1493 | *value = st->use_vbr; |
| 1494 | } |
| 1495 | break; |
| 1496 | case OPUS_SET_VOICE_RATIO_REQUEST: |
| 1497 | { |
| 1498 | opus_int32 value = va_arg(ap, opus_int32); |
| 1499 | if (value>100 || value<-1) |
| 1500 | goto bad_arg; |
| 1501 | st->voice_ratio = value; |
| 1502 | } |
| 1503 | break; |
| 1504 | case OPUS_GET_VOICE_RATIO_REQUEST: |
| 1505 | { |
| 1506 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1507 | *value = st->voice_ratio; |
| 1508 | } |
| 1509 | break; |
| 1510 | case OPUS_SET_VBR_CONSTRAINT_REQUEST: |
| 1511 | { |
| 1512 | opus_int32 value = va_arg(ap, opus_int32); |
| 1513 | if(value<0 || value>1) |
| 1514 | return OPUS_BAD_ARG; |
| 1515 | st->vbr_constraint = value; |
| 1516 | } |
| 1517 | break; |
| 1518 | case OPUS_GET_VBR_CONSTRAINT_REQUEST: |
| 1519 | { |
| 1520 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1521 | *value = st->vbr_constraint; |
| 1522 | } |
| 1523 | break; |
| 1524 | case OPUS_SET_SIGNAL_REQUEST: |
| 1525 | { |
| 1526 | opus_int32 value = va_arg(ap, opus_int32); |
| 1527 | if(value!=OPUS_AUTO && value!=OPUS_SIGNAL_VOICE && value!=OPUS_SIGNAL_MUSIC) |
| 1528 | return OPUS_BAD_ARG; |
| 1529 | st->signal_type = value; |
| 1530 | } |
| 1531 | break; |
| 1532 | case OPUS_GET_SIGNAL_REQUEST: |
| 1533 | { |
| 1534 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1535 | *value = st->signal_type; |
| 1536 | } |
| 1537 | break; |
| 1538 | case OPUS_GET_LOOKAHEAD_REQUEST: |
| 1539 | { |
| 1540 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1541 | *value = st->Fs/400; |
| 1542 | if (st->application != OPUS_APPLICATION_RESTRICTED_LOWDELAY) |
| 1543 | *value += st->delay_compensation; |
| 1544 | } |
| 1545 | break; |
| 1546 | case OPUS_GET_SAMPLE_RATE_REQUEST: |
| 1547 | { |
| 1548 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1549 | if (value==NULL) |
| 1550 | { |
| 1551 | ret = OPUS_BAD_ARG; |
| 1552 | break; |
| 1553 | } |
| 1554 | *value = st->Fs; |
| 1555 | } |
| 1556 | break; |
| 1557 | case OPUS_GET_FINAL_RANGE_REQUEST: |
| 1558 | { |
| 1559 | opus_uint32 *value = va_arg(ap, opus_uint32*); |
| 1560 | *value = st->rangeFinal; |
| 1561 | } |
| 1562 | break; |
| 1563 | case OPUS_SET_LSB_DEPTH_REQUEST: |
| 1564 | { |
| 1565 | opus_int32 value = va_arg(ap, opus_int32); |
| 1566 | ret = celt_encoder_ctl(celt_enc, OPUS_SET_LSB_DEPTH(value)); |
| 1567 | } |
| 1568 | break; |
| 1569 | case OPUS_GET_LSB_DEPTH_REQUEST: |
| 1570 | { |
| 1571 | opus_int32 *value = va_arg(ap, opus_int32*); |
| 1572 | celt_encoder_ctl(celt_enc, OPUS_GET_LSB_DEPTH(value)); |
| 1573 | } |
| 1574 | break; |
| 1575 | case OPUS_RESET_STATE: |
| 1576 | { |
| 1577 | void *silk_enc; |
| 1578 | silk_EncControlStruct dummy; |
| 1579 | silk_enc = (char*)st+st->silk_enc_offset; |
| 1580 | |
| 1581 | OPUS_CLEAR((char*)&st->OPUS_ENCODER_RESET_START, |
| 1582 | sizeof(OpusEncoder)- |
| 1583 | ((char*)&st->OPUS_ENCODER_RESET_START - (char*)st)); |
| 1584 | |
| 1585 | celt_encoder_ctl(celt_enc, OPUS_RESET_STATE); |
| 1586 | silk_InitEncoder( silk_enc, &dummy ); |
| 1587 | st->stream_channels = st->channels; |
| 1588 | st->hybrid_stereo_width_Q14 = 1 << 14; |
| 1589 | st->first = 1; |
| 1590 | st->mode = MODE_HYBRID; |
| 1591 | st->bandwidth = OPUS_BANDWIDTH_FULLBAND; |
| 1592 | st->variable_HP_smth2_Q15 = silk_LSHIFT( silk_lin2log( VARIABLE_HP_MIN_CUTOFF_HZ ), 8 ); |
| 1593 | } |
| 1594 | break; |
| 1595 | case OPUS_SET_FORCE_MODE_REQUEST: |
| 1596 | { |
| 1597 | opus_int32 value = va_arg(ap, opus_int32); |
| 1598 | if ((value < MODE_SILK_ONLY || value > MODE_CELT_ONLY) && value != OPUS_AUTO) |
| 1599 | goto bad_arg; |
| 1600 | st->user_forced_mode = value; |
| 1601 | } |
| 1602 | break; |
| 1603 | default: |
| 1604 | /* fprintf(stderr, "unknown opus_encoder_ctl() request: %d", request);*/ |
| 1605 | ret = OPUS_UNIMPLEMENTED; |
| 1606 | break; |
| 1607 | } |
| 1608 | va_end(ap); |
| 1609 | return ret; |
| 1610 | bad_arg: |
| 1611 | va_end(ap); |
| 1612 | return OPUS_BAD_ARG; |
| 1613 | } |
| 1614 | |
| 1615 | void opus_encoder_destroy(OpusEncoder *st) |
| 1616 | { |
| 1617 | opus_free(st); |
| 1618 | } |