1/* 2 * DSP Group TrueSpeech compatible decoder 3 * Copyright (c) 2005 Konstantin Shishkov 4 * 5 * This file is part of FFmpeg. 6 * 7 * FFmpeg is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU Lesser General Public 9 * License as published by the Free Software Foundation; either 10 * version 2.1 of the License, or (at your option) any later version. 11 * 12 * FFmpeg is distributed in the hope that it will be useful, 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 15 * Lesser General Public License for more details. 16 * 17 * You should have received a copy of the GNU Lesser General Public 18 * License along with FFmpeg; if not, write to the Free Software 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 20 */ 21 22#include "libavutil/intreadwrite.h" 23#include "avcodec.h" 24 25#include "truespeech_data.h" 26/** 27 * @file 28 * TrueSpeech decoder. 29 */ 30 31/** 32 * TrueSpeech decoder context 33 */ 34typedef struct { 35 /* input data */ 36 int16_t vector[8]; //< input vector: 5/5/4/4/4/3/3/3 37 int offset1[2]; //< 8-bit value, used in one copying offset 38 int offset2[4]; //< 7-bit value, encodes offsets for copying and for two-point filter 39 int pulseoff[4]; //< 4-bit offset of pulse values block 40 int pulsepos[4]; //< 27-bit variable, encodes 7 pulse positions 41 int pulseval[4]; //< 7x2-bit pulse values 42 int flag; //< 1-bit flag, shows how to choose filters 43 /* temporary data */ 44 int filtbuf[146]; // some big vector used for storing filters 45 int prevfilt[8]; // filter from previous frame 46 int16_t tmp1[8]; // coefficients for adding to out 47 int16_t tmp2[8]; // coefficients for adding to out 48 int16_t tmp3[8]; // coefficients for adding to out 49 int16_t cvector[8]; // correlated input vector 50 int filtval; // gain value for one function 51 int16_t newvec[60]; // tmp vector 52 int16_t filters[32]; // filters for every subframe 53} TSContext; 54 55static av_cold int truespeech_decode_init(AVCodecContext * avctx) 56{ 57// TSContext *c = avctx->priv_data; 58 59 avctx->sample_fmt = SAMPLE_FMT_S16; 60 return 0; 61} 62 63static void truespeech_read_frame(TSContext *dec, const uint8_t *input) 64{ 65 uint32_t t; 66 67 /* first dword */ 68 t = AV_RL32(input); 69 input += 4; 70 71 dec->flag = t & 1; 72 73 dec->vector[0] = ts_codebook[0][(t >> 1) & 0x1F]; 74 dec->vector[1] = ts_codebook[1][(t >> 6) & 0x1F]; 75 dec->vector[2] = ts_codebook[2][(t >> 11) & 0xF]; 76 dec->vector[3] = ts_codebook[3][(t >> 15) & 0xF]; 77 dec->vector[4] = ts_codebook[4][(t >> 19) & 0xF]; 78 dec->vector[5] = ts_codebook[5][(t >> 23) & 0x7]; 79 dec->vector[6] = ts_codebook[6][(t >> 26) & 0x7]; 80 dec->vector[7] = ts_codebook[7][(t >> 29) & 0x7]; 81 82 /* second dword */ 83 t = AV_RL32(input); 84 input += 4; 85 86 dec->offset2[0] = (t >> 0) & 0x7F; 87 dec->offset2[1] = (t >> 7) & 0x7F; 88 dec->offset2[2] = (t >> 14) & 0x7F; 89 dec->offset2[3] = (t >> 21) & 0x7F; 90 91 dec->offset1[0] = ((t >> 28) & 0xF) << 4; 92 93 /* third dword */ 94 t = AV_RL32(input); 95 input += 4; 96 97 dec->pulseval[0] = (t >> 0) & 0x3FFF; 98 dec->pulseval[1] = (t >> 14) & 0x3FFF; 99 100 dec->offset1[1] = (t >> 28) & 0x0F; 101 102 /* fourth dword */ 103 t = AV_RL32(input); 104 input += 4; 105 106 dec->pulseval[2] = (t >> 0) & 0x3FFF; 107 dec->pulseval[3] = (t >> 14) & 0x3FFF; 108 109 dec->offset1[1] |= ((t >> 28) & 0x0F) << 4; 110 111 /* fifth dword */ 112 t = AV_RL32(input); 113 input += 4; 114 115 dec->pulsepos[0] = (t >> 4) & 0x7FFFFFF; 116 117 dec->pulseoff[0] = (t >> 0) & 0xF; 118 119 dec->offset1[0] |= (t >> 31) & 1; 120 121 /* sixth dword */ 122 t = AV_RL32(input); 123 input += 4; 124 125 dec->pulsepos[1] = (t >> 4) & 0x7FFFFFF; 126 127 dec->pulseoff[1] = (t >> 0) & 0xF; 128 129 dec->offset1[0] |= ((t >> 31) & 1) << 1; 130 131 /* seventh dword */ 132 t = AV_RL32(input); 133 input += 4; 134 135 dec->pulsepos[2] = (t >> 4) & 0x7FFFFFF; 136 137 dec->pulseoff[2] = (t >> 0) & 0xF; 138 139 dec->offset1[0] |= ((t >> 31) & 1) << 2; 140 141 /* eighth dword */ 142 t = AV_RL32(input); 143 input += 4; 144 145 dec->pulsepos[3] = (t >> 4) & 0x7FFFFFF; 146 147 dec->pulseoff[3] = (t >> 0) & 0xF; 148 149 dec->offset1[0] |= ((t >> 31) & 1) << 3; 150 151} 152 153static void truespeech_correlate_filter(TSContext *dec) 154{ 155 int16_t tmp[8]; 156 int i, j; 157 158 for(i = 0; i < 8; i++){ 159 if(i > 0){ 160 memcpy(tmp, dec->cvector, i * 2); 161 for(j = 0; j < i; j++) 162 dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) + 163 (dec->cvector[j] << 15) + 0x4000) >> 15; 164 } 165 dec->cvector[i] = (8 - dec->vector[i]) >> 3; 166 } 167 for(i = 0; i < 8; i++) 168 dec->cvector[i] = (dec->cvector[i] * ts_230[i]) >> 15; 169 170 dec->filtval = dec->vector[0]; 171} 172 173static void truespeech_filters_merge(TSContext *dec) 174{ 175 int i; 176 177 if(!dec->flag){ 178 for(i = 0; i < 8; i++){ 179 dec->filters[i + 0] = dec->prevfilt[i]; 180 dec->filters[i + 8] = dec->prevfilt[i]; 181 } 182 }else{ 183 for(i = 0; i < 8; i++){ 184 dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15; 185 dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15; 186 } 187 } 188 for(i = 0; i < 8; i++){ 189 dec->filters[i + 16] = dec->cvector[i]; 190 dec->filters[i + 24] = dec->cvector[i]; 191 } 192} 193 194static void truespeech_apply_twopoint_filter(TSContext *dec, int quart) 195{ 196 int16_t tmp[146 + 60], *ptr0, *ptr1; 197 const int16_t *filter; 198 int i, t, off; 199 200 t = dec->offset2[quart]; 201 if(t == 127){ 202 memset(dec->newvec, 0, 60 * 2); 203 return; 204 } 205 for(i = 0; i < 146; i++) 206 tmp[i] = dec->filtbuf[i]; 207 off = (t / 25) + dec->offset1[quart >> 1] + 18; 208 ptr0 = tmp + 145 - off; 209 ptr1 = tmp + 146; 210 filter = (const int16_t*)ts_240 + (t % 25) * 2; 211 for(i = 0; i < 60; i++){ 212 t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14; 213 ptr0++; 214 dec->newvec[i] = t; 215 ptr1[i] = t; 216 } 217} 218 219static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart) 220{ 221 int16_t tmp[7]; 222 int i, j, t; 223 const int16_t *ptr1; 224 int16_t *ptr2; 225 int coef; 226 227 memset(out, 0, 60 * 2); 228 for(i = 0; i < 7; i++) { 229 t = dec->pulseval[quart] & 3; 230 dec->pulseval[quart] >>= 2; 231 tmp[6 - i] = ts_562[dec->pulseoff[quart] * 4 + t]; 232 } 233 234 coef = dec->pulsepos[quart] >> 15; 235 ptr1 = (const int16_t*)ts_140 + 30; 236 ptr2 = tmp; 237 for(i = 0, j = 3; (i < 30) && (j > 0); i++){ 238 t = *ptr1++; 239 if(coef >= t) 240 coef -= t; 241 else{ 242 out[i] = *ptr2++; 243 ptr1 += 30; 244 j--; 245 } 246 } 247 coef = dec->pulsepos[quart] & 0x7FFF; 248 ptr1 = (const int16_t*)ts_140; 249 for(i = 30, j = 4; (i < 60) && (j > 0); i++){ 250 t = *ptr1++; 251 if(coef >= t) 252 coef -= t; 253 else{ 254 out[i] = *ptr2++; 255 ptr1 += 30; 256 j--; 257 } 258 } 259 260} 261 262static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart) 263{ 264 int i; 265 266 for(i = 0; i < 86; i++) 267 dec->filtbuf[i] = dec->filtbuf[i + 60]; 268 for(i = 0; i < 60; i++){ 269 dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3); 270 out[i] += dec->newvec[i]; 271 } 272} 273 274static void truespeech_synth(TSContext *dec, int16_t *out, int quart) 275{ 276 int i,k; 277 int t[8]; 278 int16_t *ptr0, *ptr1; 279 280 ptr0 = dec->tmp1; 281 ptr1 = dec->filters + quart * 8; 282 for(i = 0; i < 60; i++){ 283 int sum = 0; 284 for(k = 0; k < 8; k++) 285 sum += ptr0[k] * ptr1[k]; 286 sum = (sum + (out[i] << 12) + 0x800) >> 12; 287 out[i] = av_clip(sum, -0x7FFE, 0x7FFE); 288 for(k = 7; k > 0; k--) 289 ptr0[k] = ptr0[k - 1]; 290 ptr0[0] = out[i]; 291 } 292 293 for(i = 0; i < 8; i++) 294 t[i] = (ts_5E2[i] * ptr1[i]) >> 15; 295 296 ptr0 = dec->tmp2; 297 for(i = 0; i < 60; i++){ 298 int sum = 0; 299 for(k = 0; k < 8; k++) 300 sum += ptr0[k] * t[k]; 301 for(k = 7; k > 0; k--) 302 ptr0[k] = ptr0[k - 1]; 303 ptr0[0] = out[i]; 304 out[i] = ((out[i] << 12) - sum) >> 12; 305 } 306 307 for(i = 0; i < 8; i++) 308 t[i] = (ts_5F2[i] * ptr1[i]) >> 15; 309 310 ptr0 = dec->tmp3; 311 for(i = 0; i < 60; i++){ 312 int sum = out[i] << 12; 313 for(k = 0; k < 8; k++) 314 sum += ptr0[k] * t[k]; 315 for(k = 7; k > 0; k--) 316 ptr0[k] = ptr0[k - 1]; 317 ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); 318 319 sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum; 320 sum = sum - (sum >> 3); 321 out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE); 322 } 323} 324 325static void truespeech_save_prevvec(TSContext *c) 326{ 327 int i; 328 329 for(i = 0; i < 8; i++) 330 c->prevfilt[i] = c->cvector[i]; 331} 332 333static int truespeech_decode_frame(AVCodecContext *avctx, 334 void *data, int *data_size, 335 AVPacket *avpkt) 336{ 337 const uint8_t *buf = avpkt->data; 338 int buf_size = avpkt->size; 339 TSContext *c = avctx->priv_data; 340 341 int i, j; 342 short *samples = data; 343 int consumed = 0; 344 int16_t out_buf[240]; 345 int iterations; 346 347 if (!buf_size) 348 return 0; 349 350 iterations = FFMIN(buf_size / 32, *data_size / 480); 351 for(j = 0; j < iterations; j++) { 352 truespeech_read_frame(c, buf + consumed); 353 consumed += 32; 354 355 truespeech_correlate_filter(c); 356 truespeech_filters_merge(c); 357 358 memset(out_buf, 0, 240 * 2); 359 for(i = 0; i < 4; i++) { 360 truespeech_apply_twopoint_filter(c, i); 361 truespeech_place_pulses(c, out_buf + i * 60, i); 362 truespeech_update_filters(c, out_buf + i * 60, i); 363 truespeech_synth(c, out_buf + i * 60, i); 364 } 365 366 truespeech_save_prevvec(c); 367 368 /* finally output decoded frame */ 369 for(i = 0; i < 240; i++) 370 *samples++ = out_buf[i]; 371 372 } 373 374 *data_size = consumed * 15; 375 376 return consumed; 377} 378 379AVCodec truespeech_decoder = { 380 "truespeech", 381 AVMEDIA_TYPE_AUDIO, 382 CODEC_ID_TRUESPEECH, 383 sizeof(TSContext), 384 truespeech_decode_init, 385 NULL, 386 NULL, 387 truespeech_decode_frame, 388 .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"), 389}; 390