libavcodec/truespeech.c

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