metasound.c 13.1 KB
Newer Older
1 2 3 4 5 6
/*
 * Voxware MetaSound decoder
 * Copyright (c) 2013 Konstantin Shishkov
 * based on TwinVQ decoder
 * Copyright (c) 2009 Vitor Sessak
 *
7
 * This file is part of FFmpeg.
8
 *
9
 * FFmpeg is free software; you can redistribute it and/or
10 11 12 13
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
14
 * FFmpeg is distributed in the hope that it will be useful,
15 16 17 18 19
 * 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
20
 * License along with FFmpeg; if not, write to the Free Software
21 22 23
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

24
#include <inttypes.h>
25 26 27 28 29
#include <math.h>
#include <stdint.h>

#include "libavutil/channel_layout.h"
#include "libavutil/float_dsp.h"
30 31

#define BITSTREAM_READER_LE
32 33
#include "avcodec.h"
#include "fft.h"
34
#include "get_bits.h"
35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167
#include "internal.h"
#include "lsp.h"
#include "sinewin.h"

#include "twinvq.h"
#include "metasound_data.h"

static void add_peak(float period, int width, const float *shape,
                     float ppc_gain, float *speech, int len)
{
    int i, j, center;
    const float *shape_end = shape + len;

    // First peak centered around zero
    for (i = 0; i < width / 2; i++)
        speech[i] += ppc_gain * *shape++;

    for (i = 1; i < ROUNDED_DIV(len, width); i++) {
        center = (int)(i * period + 0.5);
        for (j = -width / 2; j < (width + 1) / 2; j++)
            speech[j + center] += ppc_gain * *shape++;
    }

    // For the last block, be careful not to go beyond the end of the buffer
    center = (int)(i * period + 0.5);
    for (j = -width / 2; j < (width + 1) / 2 && shape < shape_end; j++)
        speech[j + center] += ppc_gain * *shape++;
}

static void decode_ppc(TwinVQContext *tctx, int period_coef, int g_coef,
                       const float *shape, float *speech)
{
    const TwinVQModeTab *mtab = tctx->mtab;
    int isampf       = tctx->avctx->sample_rate / 1000;
    int ibps         = tctx->avctx->bit_rate / (1000 * tctx->avctx->channels);
    int width;

    float ratio = (float)mtab->size / isampf;
    float min_period, max_period, period_range, period;
    float some_mult;

    float pgain_base, pgain_step, ppc_gain;

    if (tctx->avctx->channels == 1) {
        min_period = log2(ratio * 0.2);
        max_period = min_period + log2(6);
    } else {
        min_period = (int)(ratio * 0.2 * 400     + 0.5) / 400.0;
        max_period = (int)(ratio * 0.2 * 400 * 6 + 0.5) / 400.0;
    }
    period_range = max_period - min_period;
    period       = min_period + period_coef * period_range /
                   ((1 << mtab->ppc_period_bit) - 1);
    if (tctx->avctx->channels == 1)
        period = powf(2.0, period);
    else
        period = (int)(period * 400 + 0.5) / 400.0;

    switch (isampf) {
    case  8: some_mult = 2.0; break;
    case 11: some_mult = 3.0; break;
    case 16: some_mult = 3.0; break;
    case 22: some_mult = ibps == 32 ? 2.0 : 4.0; break;
    case 44: some_mult = 8.0; break;
    default: some_mult = 4.0;
    }

    width = (int)(some_mult / (mtab->size / period) * mtab->ppc_shape_len);
    if (isampf == 22 && ibps == 32)
        width = (int)((2.0 / period + 1) * width + 0.5);

    pgain_base = tctx->avctx->channels == 2 ? 25000.0 : 20000.0;
    pgain_step = pgain_base / ((1 << mtab->pgain_bit) - 1);
    ppc_gain   = 1.0 / 8192 *
                 twinvq_mulawinv(pgain_step * g_coef + pgain_step / 2,
                                 pgain_base, TWINVQ_PGAIN_MU);

    add_peak(period, width, shape, ppc_gain, speech, mtab->ppc_shape_len);
}

static void dec_bark_env(TwinVQContext *tctx, const uint8_t *in, int use_hist,
                         int ch, float *out, float gain,
                         enum TwinVQFrameType ftype)
{
    const TwinVQModeTab *mtab = tctx->mtab;
    int i, j;
    float *hist     = tctx->bark_hist[ftype][ch];
    float val       = ((const float []) { 0.4, 0.35, 0.28 })[ftype];
    int bark_n_coef = mtab->fmode[ftype].bark_n_coef;
    int fw_cb_len   = mtab->fmode[ftype].bark_env_size / bark_n_coef;
    int idx         = 0;

    if (tctx->avctx->channels == 1)
        val = 0.5;
    for (i = 0; i < fw_cb_len; i++)
        for (j = 0; j < bark_n_coef; j++, idx++) {
            float tmp2 = mtab->fmode[ftype].bark_cb[fw_cb_len * in[j] + i] *
                         (1.0 / 2048);
            float st;

            if (tctx->avctx->channels == 1)
                st = use_hist ?
                    tmp2 + val * hist[idx] + 1.0 : tmp2 + 1.0;
            else
                st = use_hist ? (1.0 - val) * tmp2 + val * hist[idx] + 1.0
                              : tmp2 + 1.0;

            hist[idx] = tmp2;
            if (st < 0.1)
                st = 0.1;

            twinvq_memset_float(out, st * gain,
                                mtab->fmode[ftype].bark_tab[idx]);
            out += mtab->fmode[ftype].bark_tab[idx];
        }
}

static void read_cb_data(TwinVQContext *tctx, GetBitContext *gb,
                         uint8_t *dst, enum TwinVQFrameType ftype)
{
    int i;

    for (i = 0; i < tctx->n_div[ftype]; i++) {
        int bs_second_part = (i >= tctx->bits_main_spec_change[ftype]);

        *dst++ = get_bits(gb, tctx->bits_main_spec[0][ftype][bs_second_part]);
        *dst++ = get_bits(gb, tctx->bits_main_spec[1][ftype][bs_second_part]);
    }
}

static int metasound_read_bitstream(AVCodecContext *avctx, TwinVQContext *tctx,
                                    const uint8_t *buf, int buf_size)
{
168
    TwinVQFrameData     *bits;
169 170 171 172 173 174 175 176
    const TwinVQModeTab *mtab = tctx->mtab;
    int channels              = tctx->avctx->channels;
    int sub;
    GetBitContext gb;
    int i, j, k;

    init_get_bits(&gb, buf, buf_size * 8);

177 178 179
    for (tctx->cur_frame = 0; tctx->cur_frame < tctx->frames_per_packet;
         tctx->cur_frame++) {
        bits = tctx->bits + tctx->cur_frame;
180

181
        bits->window_type = get_bits(&gb, TWINVQ_WINDOW_TYPE_BITS);
182

183 184 185 186
        if (bits->window_type > 8) {
            av_log(avctx, AV_LOG_ERROR, "Invalid window type, broken sample?\n");
            return AVERROR_INVALIDDATA;
        }
187

188
        bits->ftype = ff_twinvq_wtype_to_ftype_table[tctx->bits[tctx->cur_frame].window_type];
189

190
        sub = mtab->fmode[bits->ftype].sub;
191

192
        if (bits->ftype != TWINVQ_FT_SHORT && !tctx->is_6kbps)
193
            get_bits(&gb, 2);
194

195
        read_cb_data(tctx, &gb, bits->main_coeffs, bits->ftype);
196

197 198 199 200 201
        for (i = 0; i < channels; i++)
            for (j = 0; j < sub; j++)
                for (k = 0; k < mtab->fmode[bits->ftype].bark_n_coef; k++)
                    bits->bark1[i][j][k] =
                        get_bits(&gb, mtab->fmode[bits->ftype].bark_n_bit);
202 203 204

        for (i = 0; i < channels; i++)
            for (j = 0; j < sub; j++)
205 206 207 208 209 210 211 212 213 214 215 216
                bits->bark_use_hist[i][j] = get_bits1(&gb);

        if (bits->ftype == TWINVQ_FT_LONG) {
            for (i = 0; i < channels; i++)
                bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
        } else {
            for (i = 0; i < channels; i++) {
                bits->gain_bits[i] = get_bits(&gb, TWINVQ_GAIN_BITS);
                for (j = 0; j < sub; j++)
                    bits->sub_gain_bits[i * sub + j] =
                        get_bits(&gb, TWINVQ_SUB_GAIN_BITS);
            }
217 218
        }

219 220 221
        for (i = 0; i < channels; i++) {
            bits->lpc_hist_idx[i] = get_bits(&gb, mtab->lsp_bit0);
            bits->lpc_idx1[i]     = get_bits(&gb, mtab->lsp_bit1);
222

223 224 225
            for (j = 0; j < mtab->lsp_split; j++)
                bits->lpc_idx2[i][j] = get_bits(&gb, mtab->lsp_bit2);
        }
226

227 228 229 230 231 232
        if (bits->ftype == TWINVQ_FT_LONG) {
            read_cb_data(tctx, &gb, bits->ppc_coeffs, 3);
            for (i = 0; i < channels; i++) {
                bits->p_coef[i] = get_bits(&gb, mtab->ppc_period_bit);
                bits->g_coef[i] = get_bits(&gb, mtab->pgain_bit);
            }
233
        }
234 235 236 237

        // subframes are aligned to nibbles
        if (get_bits_count(&gb) & 3)
            skip_bits(&gb, 4 - (get_bits_count(&gb) & 3));
238 239
    }

240
    return (get_bits_count(&gb) + 7) / 8;
241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287
}

typedef struct MetasoundProps {
    uint32_t tag;
    int      bit_rate;
    int      channels;
    int      sample_rate;
} MetasoundProps;

static const MetasoundProps codec_props[] = {
    { MKTAG('V','X','0','3'),  6, 1,  8000 },
    { MKTAG('V','X','0','4'), 12, 2,  8000 },

    { MKTAG('V','O','X','i'),  8, 1,  8000 },
    { MKTAG('V','O','X','j'), 10, 1, 11025 },
    { MKTAG('V','O','X','k'), 16, 1, 16000 },
    { MKTAG('V','O','X','L'), 24, 1, 22050 },
    { MKTAG('V','O','X','q'), 32, 1, 44100 },
    { MKTAG('V','O','X','r'), 40, 1, 44100 },
    { MKTAG('V','O','X','s'), 48, 1, 44100 },
    { MKTAG('V','O','X','t'), 16, 2,  8000 },
    { MKTAG('V','O','X','u'), 20, 2, 11025 },
    { MKTAG('V','O','X','v'), 32, 2, 16000 },
    { MKTAG('V','O','X','w'), 48, 2, 22050 },
    { MKTAG('V','O','X','x'), 64, 2, 44100 },
    { MKTAG('V','O','X','y'), 80, 2, 44100 },
    { MKTAG('V','O','X','z'), 96, 2, 44100 },

    { 0, 0, 0, 0 }
};

static av_cold int metasound_decode_init(AVCodecContext *avctx)
{
    int isampf, ibps;
    TwinVQContext *tctx = avctx->priv_data;
    uint32_t tag;
    const MetasoundProps *props = codec_props;

    if (!avctx->extradata || avctx->extradata_size < 16) {
        av_log(avctx, AV_LOG_ERROR, "Missing or incomplete extradata\n");
        return AVERROR_INVALIDDATA;
    }

    tag = AV_RL32(avctx->extradata + 12);

    for (;;) {
        if (!props->tag) {
288
            av_log(avctx, AV_LOG_ERROR, "Could not find tag %08"PRIX32"\n", tag);
289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311
            return AVERROR_INVALIDDATA;
        }
        if (props->tag == tag) {
            avctx->sample_rate = props->sample_rate;
            avctx->channels    = props->channels;
            avctx->bit_rate    = props->bit_rate * 1000;
            isampf             = avctx->sample_rate / 1000;
            break;
        }
        props++;
    }

    if (avctx->channels <= 0 || avctx->channels > TWINVQ_CHANNELS_MAX) {
        av_log(avctx, AV_LOG_ERROR, "Unsupported number of channels: %i\n",
               avctx->channels);
        return AVERROR_INVALIDDATA;
    }
    avctx->channel_layout = avctx->channels == 1 ? AV_CH_LAYOUT_MONO
                                                 : AV_CH_LAYOUT_STEREO;

    ibps = avctx->bit_rate / (1000 * avctx->channels);

    switch ((avctx->channels << 16) + (isampf << 8) + ibps) {
312 313 314 315 316 317
    case (1 << 16) + ( 8 << 8) +  6:
        tctx->mtab = &ff_metasound_mode0806;
        break;
    case (2 << 16) + ( 8 << 8) +  6:
        tctx->mtab = &ff_metasound_mode0806s;
        break;
318 319 320
    case (1 << 16) + ( 8 << 8) +  8:
        tctx->mtab = &ff_metasound_mode0808;
        break;
321 322 323 324 325 326 327 328 329
    case (2 << 16) + ( 8 << 8) +  8:
        tctx->mtab = &ff_metasound_mode0808s;
        break;
    case (1 << 16) + (11 << 8) + 10:
        tctx->mtab = &ff_metasound_mode1110;
        break;
    case (2 << 16) + (11 << 8) + 10:
        tctx->mtab = &ff_metasound_mode1110s;
        break;
330 331 332
    case (1 << 16) + (16 << 8) + 16:
        tctx->mtab = &ff_metasound_mode1616;
        break;
333 334 335
    case (2 << 16) + (16 << 8) + 16:
        tctx->mtab = &ff_metasound_mode1616s;
        break;
336 337 338 339 340 341
    case (1 << 16) + (22 << 8) + 24:
        tctx->mtab = &ff_metasound_mode2224;
        break;
    case (2 << 16) + (22 << 8) + 24:
        tctx->mtab = &ff_metasound_mode2224s;
        break;
342 343 344
    case (1 << 16) + (44 << 8) + 32:
        tctx->mtab = &ff_metasound_mode4432;
        break;
345 346 347 348 349 350 351 352 353 354 355 356
    case (2 << 16) + (44 << 8) + 32:
        tctx->mtab = &ff_metasound_mode4432s;
        break;
    case (1 << 16) + (44 << 8) + 40:
        tctx->mtab = &ff_metasound_mode4440;
        break;
    case (2 << 16) + (44 << 8) + 40:
        tctx->mtab = &ff_metasound_mode4440s;
        break;
    case (1 << 16) + (44 << 8) + 48:
        tctx->mtab = &ff_metasound_mode4448;
        break;
357 358 359 360 361 362
    case (2 << 16) + (44 << 8) + 48:
        tctx->mtab = &ff_metasound_mode4448s;
        break;
    default:
        av_log(avctx, AV_LOG_ERROR,
               "This version does not support %d kHz - %d kbit/s/ch mode.\n",
363
               isampf, ibps);
364 365 366 367 368 369 370
        return AVERROR(ENOSYS);
    }

    tctx->codec          = TWINVQ_CODEC_METASOUND;
    tctx->read_bitstream = metasound_read_bitstream;
    tctx->dec_bark_env   = dec_bark_env;
    tctx->decode_ppc     = decode_ppc;
371 372
    tctx->frame_size     = avctx->bit_rate * tctx->mtab->size
                                           / avctx->sample_rate;
373
    tctx->is_6kbps       = ibps == 6;
374 375 376 377 378 379 380 381 382 383 384 385 386

    return ff_twinvq_decode_init(avctx);
}

AVCodec ff_metasound_decoder = {
    .name           = "metasound",
    .long_name      = NULL_IF_CONFIG_SMALL("Voxware MetaSound"),
    .type           = AVMEDIA_TYPE_AUDIO,
    .id             = AV_CODEC_ID_METASOUND,
    .priv_data_size = sizeof(TwinVQContext),
    .init           = metasound_decode_init,
    .close          = ff_twinvq_decode_close,
    .decode         = ff_twinvq_decode_frame,
387
    .capabilities   = AV_CODEC_CAP_DR1,
388 389 390
    .sample_fmts    = (const enum AVSampleFormat[]) { AV_SAMPLE_FMT_FLTP,
                                                      AV_SAMPLE_FMT_NONE },
};