truespeech.c 10.3 KB
Newer Older
1 2 3 4
/*
 * DSP Group TrueSpeech compatible decoder
 * Copyright (c) 2005 Konstantin Shishkov
 *
5 6 7
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
8 9
 * modify it under the terms of the GNU Lesser General Public
 * 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 14 15 16 17
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * 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 24 25 26
#include "avcodec.h"

#include "truespeech_data.h"
/**
27
 * @file
28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54
 * TrueSpeech decoder.
 */

/**
 * TrueSpeech decoder context
 */
typedef struct {
    /* input data */
    int16_t vector[8];  //< input vector: 5/5/4/4/4/3/3/3
    int offset1[2];     //< 8-bit value, used in one copying offset
    int offset2[4];     //< 7-bit value, encodes offsets for copying and for two-point filter
    int pulseoff[4];    //< 4-bit offset of pulse values block
    int pulsepos[4];    //< 27-bit variable, encodes 7 pulse positions
    int pulseval[4];    //< 7x2-bit pulse values
    int flag;           //< 1-bit flag, shows how to choose filters
    /* temporary data */
    int filtbuf[146];   // some big vector used for storing filters
    int prevfilt[8];    // filter from previous frame
    int16_t tmp1[8];    // coefficients for adding to out
    int16_t tmp2[8];    // coefficients for adding to out
    int16_t tmp3[8];    // coefficients for adding to out
    int16_t cvector[8]; // correlated input vector
    int filtval;        // gain value for one function
    int16_t newvec[60]; // tmp vector
    int16_t filters[32]; // filters for every subframe
} TSContext;

55
static av_cold int truespeech_decode_init(AVCodecContext * avctx)
56 57 58
{
//    TSContext *c = avctx->priv_data;

59
    avctx->sample_fmt = SAMPLE_FMT_S16;
60 61 62
    return 0;
}

Michael Niedermayer's avatar
Michael Niedermayer committed
63
static void truespeech_read_frame(TSContext *dec, const uint8_t *input)
64 65 66 67
{
    uint32_t t;

    /* first dword */
68
    t = AV_RL32(input);
69 70 71 72 73 74 75 76 77 78 79 80 81 82
    input += 4;

    dec->flag = t & 1;

    dec->vector[0] = ts_codebook[0][(t >>  1) & 0x1F];
    dec->vector[1] = ts_codebook[1][(t >>  6) & 0x1F];
    dec->vector[2] = ts_codebook[2][(t >> 11) &  0xF];
    dec->vector[3] = ts_codebook[3][(t >> 15) &  0xF];
    dec->vector[4] = ts_codebook[4][(t >> 19) &  0xF];
    dec->vector[5] = ts_codebook[5][(t >> 23) &  0x7];
    dec->vector[6] = ts_codebook[6][(t >> 26) &  0x7];
    dec->vector[7] = ts_codebook[7][(t >> 29) &  0x7];

    /* second dword */
83
    t = AV_RL32(input);
84 85 86 87 88 89 90 91 92 93
    input += 4;

    dec->offset2[0] = (t >>  0) & 0x7F;
    dec->offset2[1] = (t >>  7) & 0x7F;
    dec->offset2[2] = (t >> 14) & 0x7F;
    dec->offset2[3] = (t >> 21) & 0x7F;

    dec->offset1[0] = ((t >> 28) & 0xF) << 4;

    /* third dword */
94
    t = AV_RL32(input);
95 96 97 98 99 100 101 102
    input += 4;

    dec->pulseval[0] = (t >>  0) & 0x3FFF;
    dec->pulseval[1] = (t >> 14) & 0x3FFF;

    dec->offset1[1] = (t >> 28) & 0x0F;

    /* fourth dword */
103
    t = AV_RL32(input);
104 105 106 107 108 109 110 111
    input += 4;

    dec->pulseval[2] = (t >>  0) & 0x3FFF;
    dec->pulseval[3] = (t >> 14) & 0x3FFF;

    dec->offset1[1] |= ((t >> 28) & 0x0F) << 4;

    /* fifth dword */
112
    t = AV_RL32(input);
113 114 115 116 117 118 119 120 121
    input += 4;

    dec->pulsepos[0] = (t >> 4) & 0x7FFFFFF;

    dec->pulseoff[0] = (t >> 0) & 0xF;

    dec->offset1[0] |= (t >> 31) & 1;

    /* sixth dword */
122
    t = AV_RL32(input);
123 124 125 126 127 128 129 130 131
    input += 4;

    dec->pulsepos[1] = (t >> 4) & 0x7FFFFFF;

    dec->pulseoff[1] = (t >> 0) & 0xF;

    dec->offset1[0] |= ((t >> 31) & 1) << 1;

    /* seventh dword */
132
    t = AV_RL32(input);
133 134 135 136 137 138 139 140 141
    input += 4;

    dec->pulsepos[2] = (t >> 4) & 0x7FFFFFF;

    dec->pulseoff[2] = (t >> 0) & 0xF;

    dec->offset1[0] |= ((t >> 31) & 1) << 2;

    /* eighth dword */
142
    t = AV_RL32(input);
143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195
    input += 4;

    dec->pulsepos[3] = (t >> 4) & 0x7FFFFFF;

    dec->pulseoff[3] = (t >> 0) & 0xF;

    dec->offset1[0] |= ((t >> 31) & 1) << 3;

}

static void truespeech_correlate_filter(TSContext *dec)
{
    int16_t tmp[8];
    int i, j;

    for(i = 0; i < 8; i++){
        if(i > 0){
            memcpy(tmp, dec->cvector, i * 2);
            for(j = 0; j < i; j++)
                dec->cvector[j] = ((tmp[i - j - 1] * dec->vector[i]) +
                                   (dec->cvector[j] << 15) + 0x4000) >> 15;
        }
        dec->cvector[i] = (8 - dec->vector[i]) >> 3;
    }
    for(i = 0; i < 8; i++)
        dec->cvector[i] = (dec->cvector[i] * ts_230[i]) >> 15;

    dec->filtval = dec->vector[0];
}

static void truespeech_filters_merge(TSContext *dec)
{
    int i;

    if(!dec->flag){
        for(i = 0; i < 8; i++){
            dec->filters[i + 0] = dec->prevfilt[i];
            dec->filters[i + 8] = dec->prevfilt[i];
        }
    }else{
        for(i = 0; i < 8; i++){
            dec->filters[i + 0]=(dec->cvector[i] * 21846 + dec->prevfilt[i] * 10923 + 16384) >> 15;
            dec->filters[i + 8]=(dec->cvector[i] * 10923 + dec->prevfilt[i] * 21846 + 16384) >> 15;
        }
    }
    for(i = 0; i < 8; i++){
        dec->filters[i + 16] = dec->cvector[i];
        dec->filters[i + 24] = dec->cvector[i];
    }
}

static void truespeech_apply_twopoint_filter(TSContext *dec, int quart)
{
196 197
    int16_t tmp[146 + 60], *ptr0, *ptr1;
    const int16_t *filter;
198 199 200 201 202 203 204 205 206 207 208 209
    int i, t, off;

    t = dec->offset2[quart];
    if(t == 127){
        memset(dec->newvec, 0, 60 * 2);
        return;
    }
    for(i = 0; i < 146; i++)
        tmp[i] = dec->filtbuf[i];
    off = (t / 25) + dec->offset1[quart >> 1] + 18;
    ptr0 = tmp + 145 - off;
    ptr1 = tmp + 146;
210
    filter = (const int16_t*)ts_240 + (t % 25) * 2;
211 212 213 214 215 216 217 218 219 220 221 222
    for(i = 0; i < 60; i++){
        t = (ptr0[0] * filter[0] + ptr0[1] * filter[1] + 0x2000) >> 14;
        ptr0++;
        dec->newvec[i] = t;
        ptr1[i] = t;
    }
}

static void truespeech_place_pulses(TSContext *dec, int16_t *out, int quart)
{
    int16_t tmp[7];
    int i, j, t;
223 224
    const int16_t *ptr1;
    int16_t *ptr2;
225 226 227 228 229 230 231 232 233 234
    int coef;

    memset(out, 0, 60 * 2);
    for(i = 0; i < 7; i++) {
        t = dec->pulseval[quart] & 3;
        dec->pulseval[quart] >>= 2;
        tmp[6 - i] = ts_562[dec->pulseoff[quart] * 4 + t];
    }

    coef = dec->pulsepos[quart] >> 15;
235
    ptr1 = (const int16_t*)ts_140 + 30;
236 237 238 239 240 241 242 243 244 245 246 247
    ptr2 = tmp;
    for(i = 0, j = 3; (i < 30) && (j > 0); i++){
        t = *ptr1++;
        if(coef >= t)
            coef -= t;
        else{
            out[i] = *ptr2++;
            ptr1 += 30;
            j--;
        }
    }
    coef = dec->pulsepos[quart] & 0x7FFF;
248
    ptr1 = (const int16_t*)ts_140;
249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286
    for(i = 30, j = 4; (i < 60) && (j > 0); i++){
        t = *ptr1++;
        if(coef >= t)
            coef -= t;
        else{
            out[i] = *ptr2++;
            ptr1 += 30;
            j--;
        }
    }

}

static void truespeech_update_filters(TSContext *dec, int16_t *out, int quart)
{
    int i;

    for(i = 0; i < 86; i++)
        dec->filtbuf[i] = dec->filtbuf[i + 60];
    for(i = 0; i < 60; i++){
        dec->filtbuf[i + 86] = out[i] + dec->newvec[i] - (dec->newvec[i] >> 3);
        out[i] += dec->newvec[i];
    }
}

static void truespeech_synth(TSContext *dec, int16_t *out, int quart)
{
    int i,k;
    int t[8];
    int16_t *ptr0, *ptr1;

    ptr0 = dec->tmp1;
    ptr1 = dec->filters + quart * 8;
    for(i = 0; i < 60; i++){
        int sum = 0;
        for(k = 0; k < 8; k++)
            sum += ptr0[k] * ptr1[k];
        sum = (sum + (out[i] << 12) + 0x800) >> 12;
287
        out[i] = av_clip(sum, -0x7FFE, 0x7FFE);
288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316
        for(k = 7; k > 0; k--)
            ptr0[k] = ptr0[k - 1];
        ptr0[0] = out[i];
    }

    for(i = 0; i < 8; i++)
        t[i] = (ts_5E2[i] * ptr1[i]) >> 15;

    ptr0 = dec->tmp2;
    for(i = 0; i < 60; i++){
        int sum = 0;
        for(k = 0; k < 8; k++)
            sum += ptr0[k] * t[k];
        for(k = 7; k > 0; k--)
            ptr0[k] = ptr0[k - 1];
        ptr0[0] = out[i];
        out[i] = ((out[i] << 12) - sum) >> 12;
    }

    for(i = 0; i < 8; i++)
        t[i] = (ts_5F2[i] * ptr1[i]) >> 15;

    ptr0 = dec->tmp3;
    for(i = 0; i < 60; i++){
        int sum = out[i] << 12;
        for(k = 0; k < 8; k++)
            sum += ptr0[k] * t[k];
        for(k = 7; k > 0; k--)
            ptr0[k] = ptr0[k - 1];
317
        ptr0[0] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
318 319 320

        sum = ((ptr0[1] * (dec->filtval - (dec->filtval >> 2))) >> 4) + sum;
        sum = sum - (sum >> 3);
321
        out[i] = av_clip((sum + 0x800) >> 12, -0x7FFE, 0x7FFE);
322 323 324 325 326 327 328 329 330 331 332 333 334
    }
}

static void truespeech_save_prevvec(TSContext *c)
{
    int i;

    for(i = 0; i < 8; i++)
        c->prevfilt[i] = c->cvector[i];
}

static int truespeech_decode_frame(AVCodecContext *avctx,
                void *data, int *data_size,
335
                AVPacket *avpkt)
336
{
337 338
    const uint8_t *buf = avpkt->data;
    int buf_size = avpkt->size;
339 340
    TSContext *c = avctx->priv_data;

341
    int i, j;
342 343 344
    short *samples = data;
    int consumed = 0;
    int16_t out_buf[240];
345
    int iterations;
346 347 348 349

    if (!buf_size)
        return 0;

350 351 352 353 354
    if (buf_size < 32) {
        av_log(avctx, AV_LOG_ERROR,
               "Too small input buffer (%d bytes), need at least 32 bytes\n", buf_size);
        return -1;
    }
355 356
    iterations = FFMIN(buf_size / 32, *data_size / 480);
    for(j = 0; j < iterations; j++) {
357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380
        truespeech_read_frame(c, buf + consumed);
        consumed += 32;

        truespeech_correlate_filter(c);
        truespeech_filters_merge(c);

        memset(out_buf, 0, 240 * 2);
        for(i = 0; i < 4; i++) {
            truespeech_apply_twopoint_filter(c, i);
            truespeech_place_pulses(c, out_buf + i * 60, i);
            truespeech_update_filters(c, out_buf + i * 60, i);
            truespeech_synth(c, out_buf + i * 60, i);
        }

        truespeech_save_prevvec(c);

        /* finally output decoded frame */
        for(i = 0; i < 240; i++)
            *samples++ = out_buf[i];

    }

    *data_size = consumed * 15;

381
    return consumed;
382 383 384 385
}

AVCodec truespeech_decoder = {
    "truespeech",
386
    AVMEDIA_TYPE_AUDIO,
387 388 389 390 391 392
    CODEC_ID_TRUESPEECH,
    sizeof(TSContext),
    truespeech_decode_init,
    NULL,
    NULL,
    truespeech_decode_frame,
393
    .long_name = NULL_IF_CONFIG_SMALL("DSP Group TrueSpeech"),
394
};