libavcodec/truespeech.c

   1 /*
   2  * DSP Group TrueSpeech compatible decoder
   3  * Copyright (c) 2005 Konstantin Shishkov
   4  *
   5  * This file is part of Libav.
   6  *
   7  * Libav 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  * Libav 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 Libav; 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/channel_layout.h"
  23 #include "libavutil/intreadwrite.h"
  24 #include "avcodec.h"
  25 #include "dsputil.h"
  26 #include "get_bits.h"
  27 #include "internal.h"
  28
  29 #include "truespeech_data.h"
  30 /**
  31  * @file
  32  * TrueSpeech decoder.
  33  */
  34
  35 /**
  36  * TrueSpeech decoder context
  37  */
  38 typedef struct {
  39     AVFrame frame;
  40     DSPContext dsp;
  41     /* input data */
  42     DECLARE_ALIGNED(16, uint8_t, buffer)[32];
  43     int16_t vector[8];  ///< input vector: 5/5/4/4/4/3/3/3
  44     int offset1[2];     ///< 8-bit value, used in one copying offset
  45     int offset2[4];     ///< 7-bit value, encodes offsets for copying and for two-point filter
  46     int pulseoff[4];    ///< 4-bit offset of pulse values block
  47     int pulsepos[4];    ///< 27-bit variable, encodes 7 pulse positions
  48     int pulseval[4];    ///< 7x2-bit pulse values
  49     int flag;           ///< 1-bit flag, shows how to choose filters
  50     /* temporary data */
  51     int filtbuf[146];   // some big vector used for storing filters
  52     int prevfilt[8];    // filter from previous frame
  53     int16_t tmp1[8];    // coefficients for adding to out
  54     int16_t tmp2[8];    // coefficients for adding to out
  55     int16_t tmp3[8];    // coefficients for adding to out
  56     int16_t cvector[8]; // correlated input vector
  57     int filtval;        // gain value for one function
  58     int16_t newvec[60]; // tmp vector
  59     int16_t filters[32]; // filters for every subframe
  60 } TSContext;
  61
  62 static av_cold int truespeech_decode_init(AVCodecContext * avctx)
  63 {
  64     TSContext *c = avctx->priv_data;
  65
  66     if (avctx->channels != 1) {
  67         av_log_ask_for_sample(avctx, "Unsupported channel count: %d\n", avctx->channels);
  68         return AVERROR_PATCHWELCOME;
  69     }
  70
  71     avctx->channel_layout = AV_CH_LAYOUT_MONO;
  72     avctx->sample_fmt     = AV_SAMPLE_FMT_S16;
  73
  74     ff_dsputil_init(&c->dsp, avctx);
  75
  76     avcodec_get_frame_defaults(&c->frame);
  77     avctx->coded_frame = &c->frame;
  78
  79     return 0;
  80 }
  81
  82 static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
  83 {
  84     GetBitContext gb;
  85
  86     dec->dsp.bswap_buf((uint32_t *)dec->buffer, (const uint32_t *)input, 8);
  87     init_get_bits(&gb, dec->buffer, 32 * 8);
  88
  89     dec->vector[7] = ts_codebook[7][get_bits(&gb, 3)];
  90     dec->vector[6] = ts_codebook[6][get_bits(&gb, 3)];
  91     dec->vector[5] = ts_codebook[5][get_bits(&gb, 3)];
  92     dec->vector[4] = ts_codebook[4][get_bits(&gb, 4)];
  93     dec->vector[3] = ts_codebook[3][get_bits(&gb, 4)];
  94     dec->vector[2] = ts_codebook[2][get_bits(&gb, 4)];
  95     dec->vector[1] = ts_codebook[1][get_bits(&gb, 5)];
  96     dec->vector[0] = ts_codebook[0][get_bits(&gb, 5)];
  97     dec->flag      = get_bits1(&gb);
  98
  99     dec->offset1[0] = get_bits(&gb, 4) << 4;
 100     dec->offset2[3] = get_bits(&gb, 7);
 101     dec->offset2[2] = get_bits(&gb, 7);
 102     dec->offset2[1] = get_bits(&gb, 7);
 103     dec->offset2[0] = get_bits(&gb, 7);
 104
 105     dec->offset1[1]  = get_bits(&gb, 4);
 106     dec->pulseval[1] = get_bits(&gb, 14);
 107     dec->pulseval[0] = get_bits(&gb, 14);
 108
 109     dec->offset1[1] |= get_bits(&gb, 4) << 4;
 110     dec->pulseval[3] = get_bits(&gb, 14);
 111     dec->pulseval[2] = get_bits(&gb, 14);
 112
 113     dec->offset1[0] |= get_bits1(&gb);
 114     dec->pulsepos[0] = get_bits_long(&gb, 27);
 115     dec->pulseoff[0] = get_bits(&gb, 4);
 116
 117     dec->offset1[0] |= get_bits1(&gb) << 1;
 118     dec->pulsepos[1] = get_bits_long(&gb, 27);
 119     dec->pulseoff[1] = get_bits(&gb, 4);
 120
 121     dec->offset1[0] |= get_bits1(&gb) << 2;
 122     dec->pulsepos[2] = get_bits_long(&gb, 27);
 123     dec->pulseoff[2] = get_bits(&gb, 4);
 124
 125     dec->offset1[0] |= get_bits1(&gb) << 3;
 126     dec->pulsepos[3] = get_bits_long(&gb, 27);
 127     dec->pulseoff[3] = get_bits(&gb, 4);
 128 }
 129
 130 static void truespeech_correlate_filter(TSContext *dec)
 131 {
 132     int16_t tmp[8];
 133     int i, j;
 134
 135     for(i = 0; i < 8; i++){
 136         if(i > 0){
 137             memcpy(tmp, dec->cvector, i * sizeof(*tmp));
 138             for(j = 0; j < i; j++)
 139                 dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
 140                                    (dec->cvector[j] << 15) + 0x4000) >> 15;
 141         }
 142         dec->cvector[i] = (8 - dec->vector[i]) >> 3;
 143     }
 144     for(i = 0; i < 8; i++)
 145         dec->cvector[i] = (dec->cvector[i] * ts_decay_994_1000[i]) >> 15;
 146
 147     dec->filtval = dec->vector[0];
 148 }
 149
 150 static void truespeech_filters_merge(TSContext *dec)
 151 {
 152     int i;
 153
 154     if(!dec->flag){
 155         for(i = 0; i < 8; i++){
 156             dec->filters[i + 0] = dec->prevfilt[i];
 157             dec->filters[i + 8] = dec->prevfilt[i];
 158         }
 159     }else{
 160         for(i = 0; i < 8; i++){
 161             dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
 162             dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
 163         }
 164     }
 165     for(i = 0; i < 8; i++){
 166         dec->filters[i + 16] = dec->cvector[i];
 167         dec->filters[i + 24] = dec->cvector[i];
 168     }
 169 }
 170
 171 static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
 172 {
 173     int16_t tmp[146 + 60], *ptr0, *ptr1;
 174     const int16_t *filter;
 175     int i, t, off;
 176
 177     t = dec->offset2[quart];
 178     if(t == 127){
 179         memset(dec->newvec, 0, 60 * sizeof(*dec->newvec));
 180         return;
 181     }
 182     for(i = 0; i < 146; i++)
 183         tmp[i] = dec->filtbuf[i];
 184     off = (t / 25) + dec->offset1[quart >> 1] + 18;
 185     off = av_clip(off, 0, 145);
 186     ptr0 = tmp + 145 - off;
 187     ptr1 = tmp + 146;
 188     filter = ts_order2_coeffs + (t % 25) * 2;
 189     for(i = 0; i < 60; i++){
 190         t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
 191         ptr0++;
 192         dec->newvec[i] = t;
 193         ptr1[i] = t;
 194     }
 195 }
 196
 197 static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
 198 {
 199     int16_t tmp[7];
 200     int i, j, t;
 201     const int16_t *ptr1;
 202     int16_t *ptr2;
 203     int coef;
 204
 205     memset(out, 0, 60 * sizeof(*out));
 206     for(i = 0; i < 7; i++) {
 207         t = dec->pulseval[quart] & 3;
 208         dec->pulseval[quart] >>= 2;
 209         tmp[6 - i] = ts_pulse_scales[dec->pulseoff[quart] * 4 + t];
 210     }
 211
 212     coef = dec->pulsepos[quart] >> 15;
 213     ptr1 = ts_pulse_values + 30;
 214     ptr2 = tmp;
 215     for(i = 0, j = 3; (i < 30) && (j > 0); i++){
 216         t = *ptr1++;
 217         if(coef >= t)
 218             coef -= t;
 219         else{
 220             out[i] = *ptr2++;
 221             ptr1 += 30;
 222             j--;
 223         }
 224     }
 225     coef = dec->pulsepos[quart] & 0x7FFF;
 226     ptr1 = ts_pulse_values;
 227     for(i = 30, j = 4; (i < 60) && (j > 0); i++){
 228         t = *ptr1++;
 229         if(coef >= t)
 230             coef -= t;
 231         else{
 232             out[i] = *ptr2++;
 233             ptr1 += 30;
 234             j--;
 235         }
 236     }
 237
 238 }
 239
 240 static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
 241 {
 242     int i;
 243
 244     memmove(dec->filtbuf, &dec->filtbuf[60], 86 * sizeof(*dec->filtbuf));
 245     for(i = 0; i < 60; i++){
 246         dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
 247         out[i] += dec->newvec[i];
 248     }
 249 }
 250
 251 static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
 252 {
 253     int i,k;
 254     int t[8];
 255     int16_t *ptr0, *ptr1;
 256
 257     ptr0 = dec->tmp1;
 258     ptr1 = dec->filters + quart * 8;
 259     for(i = 0; i < 60; i++){
 260         int sum = 0;
 261         for(k = 0; k < 8; k++)
 262             sum += ptr0[k] * ptr1[k];
 263         sum = (sum + (out[i] << 12) + 0x800) >> 12;
 264         out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
 265         for(k = 7; k > 0; k--)
 266             ptr0[k] = ptr0[k - 1];
 267         ptr0[0] = out[i];
 268     }
 269
 270     for(i = 0; i < 8; i++)
 271         t[i] = (ts_decay_35_64[i] * ptr1[i]) >> 15;
 272
 273     ptr0 = dec->tmp2;
 274     for(i = 0; i < 60; i++){
 275         int sum = 0;
 276         for(k = 0; k < 8; k++)
 277             sum += ptr0[k] * t[k];
 278         for(k = 7; k > 0; k--)
 279             ptr0[k] = ptr0[k - 1];
 280         ptr0[0] = out[i];
 281         out[i] = ((out[i] << 12) - sum) >> 12;
 282     }
 283
 284     for(i = 0; i < 8; i++)
 285         t[i] = (ts_decay_3_4[i] * ptr1[i]) >> 15;
 286
 287     ptr0 = dec->tmp3;
 288     for(i = 0; i < 60; i++){
 289         int sum = out[i] << 12;
 290         for(k = 0; k < 8; k++)
 291             sum += ptr0[k] * t[k];
 292         for(k = 7; k > 0; k--)
 293             ptr0[k] = ptr0[k - 1];
 294         ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
 295
 296         sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
 297         sum = sum - (sum >> 3);
 298         out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
 299     }
 300 }
 301
 302 static void truespeech_save_prevvec(TSContext *c)
 303 {
 304     int i;
 305
 306     for(i = 0; i < 8; i++)
 307         c->prevfilt[i] = c->cvector[i];
 308 }
 309
 310 static int truespeech_decode_frame(AVCodecContext *avctx, void *data,
 311                                    int *got_frame_ptr, AVPacket *avpkt)
 312 {
 313     const uint8_t *buf = avpkt->data;
 314     int buf_size = avpkt->size;
 315     TSContext *c = avctx->priv_data;
 316
 317     int i, j;
 318     int16_t *samples;
 319     int iterations, ret;
 320
 321     iterations = buf_size / 32;
 322
 323     if (!iterations) {
 324         av_log(avctx, AV_LOG_ERROR,
 325                "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
 326         return -1;
 327     }
 328
 329     /* get output buffer */
 330     c->frame.nb_samples = iterations * 240;
 331     if ((ret = ff_get_buffer(avctx, &c->frame)) < 0) {
 332         av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
 333         return ret;
 334     }
 335     samples = (int16_t *)c->frame.data[0];
 336
 337     memset(samples, 0, iterations * 240 * sizeof(*samples));
 338
 339     for(j = 0; j < iterations; j++) {
 340         truespeech_read_frame(c, buf);
 341         buf += 32;
 342
 343         truespeech_correlate_filter(c);
 344         truespeech_filters_merge(c);
 345
 346         for(i = 0; i < 4; i++) {
 347             truespeech_apply_twopoint_filter(c, i);
 348             truespeech_place_pulses  (c, samples, i);
 349             truespeech_update_filters(c, samples, i);
 350             truespeech_synth         (c, samples, i);
 351             samples += 60;
 352         }
 353
 354         truespeech_save_prevvec(c);
 355     }
 356
 357     *got_frame_ptr   = 1;
 358     *(AVFrame *)data = c->frame;
 359
 360     return buf_size;
 361 }
 362
 363 AVCodec ff_truespeech_decoder = {
 364     .name           = "truespeech",
 365     .type           = AVMEDIA_TYPE_AUDIO,
 366     .id             = AV_CODEC_ID_TRUESPEECH,
 367     .priv_data_size = sizeof(TSContext),
 368     .init           = truespeech_decode_init,
 369     .decode         = truespeech_decode_frame,
 370     .capabilities   = CODEC_CAP_DR1,
 371     .long_name      = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
 372 };