ra144enc.c 18.9 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22
/*
 * Real Audio 1.0 (14.4K) encoder
 * Copyright (c) 2010 Francesco Lavra <francescolavra@interfree.it>
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * 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.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * 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
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
23
 * @file
24 25 26 27
 * Real Audio 1.0 (14.4K) encoder
 * @author Francesco Lavra <francescolavra@interfree.it>
 */

28
#include <float.h>
29 30

#include "avcodec.h"
31
#include "audio_frame_queue.h"
32
#include "celp_filters.h"
33
#include "internal.h"
34
#include "mathops.h"
35 36 37
#include "put_bits.h"
#include "ra144.h"

38 39 40 41
static av_cold int ra144_encode_close(AVCodecContext *avctx)
{
    RA144Context *ractx = avctx->priv_data;
    ff_lpc_end(&ractx->lpc_ctx);
42
    ff_af_queue_close(&ractx->afq);
43 44 45 46
    return 0;
}


47 48 49
static av_cold int ra144_encode_init(AVCodecContext * avctx)
{
    RA144Context *ractx;
50
    int ret;
51 52 53 54 55 56 57

    if (avctx->channels != 1) {
        av_log(avctx, AV_LOG_ERROR, "invalid number of channels: %d\n",
               avctx->channels);
        return -1;
    }
    avctx->frame_size = NBLOCKS * BLOCKSIZE;
58
    avctx->initial_padding = avctx->frame_size;
59 60 61 62 63
    avctx->bit_rate = 8000;
    ractx = avctx->priv_data;
    ractx->lpc_coef[0] = ractx->lpc_tables[0];
    ractx->lpc_coef[1] = ractx->lpc_tables[1];
    ractx->avctx = avctx;
64
    ff_audiodsp_init(&ractx->adsp);
65
    ret = ff_lpc_init(&ractx->lpc_ctx, avctx->frame_size, LPC_ORDER,
66
                      FF_LPC_TYPE_LEVINSON);
67 68
    if (ret < 0)
        goto error;
69

70 71
    ff_af_queue_init(avctx, &ractx->afq);

72
    return 0;
73 74 75
error:
    ra144_encode_close(avctx);
    return ret;
76 77 78 79
}


/**
80
 * Quantize a value by searching a sorted table for the element with the
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
 * nearest value
 *
 * @param value value to quantize
 * @param table array containing the quantization table
 * @param size size of the quantization table
 * @return index of the quantization table corresponding to the element with the
 *         nearest value
 */
static int quantize(int value, const int16_t *table, unsigned int size)
{
    unsigned int low = 0, high = size - 1;

    while (1) {
        int index = (low + high) >> 1;
        int error = table[index] - value;

        if (index == low)
            return table[high] + error > value ? low : high;
        if (error > 0) {
            high = index;
        } else {
            low = index;
        }
    }
}


/**
109
 * Orthogonalize a vector to another vector
110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129
 *
 * @param v vector to orthogonalize
 * @param u vector against which orthogonalization is performed
 */
static void orthogonalize(float *v, const float *u)
{
    int i;
    float num = 0, den = 0;

    for (i = 0; i < BLOCKSIZE; i++) {
        num += v[i] * u[i];
        den += u[i] * u[i];
    }
    num /= den;
    for (i = 0; i < BLOCKSIZE; i++)
        v[i] -= num * u[i];
}


/**
130
 * Calculate match score and gain of an LPC-filtered vector with respect to
131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168
 * input data, possibly othogonalizing it to up to 2 other vectors
 *
 * @param work array used to calculate the filtered vector
 * @param coefs coefficients of the LPC filter
 * @param vect original vector
 * @param ortho1 first vector against which orthogonalization is performed
 * @param ortho2 second vector against which orthogonalization is performed
 * @param data input data
 * @param score pointer to variable where match score is returned
 * @param gain pointer to variable where gain is returned
 */
static void get_match_score(float *work, const float *coefs, float *vect,
                            const float *ortho1, const float *ortho2,
                            const float *data, float *score, float *gain)
{
    float c, g;
    int i;

    ff_celp_lp_synthesis_filterf(work, coefs, vect, BLOCKSIZE, LPC_ORDER);
    if (ortho1)
        orthogonalize(work, ortho1);
    if (ortho2)
        orthogonalize(work, ortho2);
    c = g = 0;
    for (i = 0; i < BLOCKSIZE; i++) {
        g += work[i] * work[i];
        c += data[i] * work[i];
    }
    if (c <= 0) {
        *score = 0;
        return;
    }
    *gain = c / g;
    *score = *gain * c;
}


/**
169
 * Create a vector from the adaptive codebook at a given lag value
170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188
 *
 * @param vect array where vector is stored
 * @param cb adaptive codebook
 * @param lag lag value
 */
static void create_adapt_vect(float *vect, const int16_t *cb, int lag)
{
    int i;

    cb += BUFFERSIZE - lag;
    for (i = 0; i < FFMIN(BLOCKSIZE, lag); i++)
        vect[i] = cb[i];
    if (lag < BLOCKSIZE)
        for (i = 0; i < BLOCKSIZE - lag; i++)
            vect[lag + i] = cb[i];
}


/**
189
 * Search the adaptive codebook for the best entry and gain and remove its
190 191 192 193 194 195 196 197 198 199 200
 * contribution from input data
 *
 * @param adapt_cb array from which the adaptive codebook is extracted
 * @param work array used to calculate LPC-filtered vectors
 * @param coefs coefficients of the LPC filter
 * @param data input data
 * @return index of the best entry of the adaptive codebook
 */
static int adaptive_cb_search(const int16_t *adapt_cb, float *work,
                              const float *coefs, float *data)
{
201 202
    int i, av_uninit(best_vect);
    float score, gain, best_score, av_uninit(best_gain);
203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225
    float exc[BLOCKSIZE];

    gain = best_score = 0;
    for (i = BLOCKSIZE / 2; i <= BUFFERSIZE; i++) {
        create_adapt_vect(exc, adapt_cb, i);
        get_match_score(work, coefs, exc, NULL, NULL, data, &score, &gain);
        if (score > best_score) {
            best_score = score;
            best_vect = i;
            best_gain = gain;
        }
    }
    if (!best_score)
        return 0;

    /**
     * Re-calculate the filtered vector from the vector with maximum match score
     * and remove its contribution from input data.
     */
    create_adapt_vect(exc, adapt_cb, best_vect);
    ff_celp_lp_synthesis_filterf(work, coefs, exc, BLOCKSIZE, LPC_ORDER);
    for (i = 0; i < BLOCKSIZE; i++)
        data[i] -= best_gain * work[i];
226
    return best_vect - BLOCKSIZE / 2 + 1;
227 228 229 230
}


/**
231
 * Find the best vector of a fixed codebook by applying an LPC filter to
232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269
 * codebook entries, possibly othogonalizing them to up to 2 other vectors and
 * matching the results with input data
 *
 * @param work array used to calculate the filtered vectors
 * @param coefs coefficients of the LPC filter
 * @param cb fixed codebook
 * @param ortho1 first vector against which orthogonalization is performed
 * @param ortho2 second vector against which orthogonalization is performed
 * @param data input data
 * @param idx pointer to variable where the index of the best codebook entry is
 *        returned
 * @param gain pointer to variable where the gain of the best codebook entry is
 *        returned
 */
static void find_best_vect(float *work, const float *coefs,
                           const int8_t cb[][BLOCKSIZE], const float *ortho1,
                           const float *ortho2, float *data, int *idx,
                           float *gain)
{
    int i, j;
    float g, score, best_score;
    float vect[BLOCKSIZE];

    *idx = *gain = best_score = 0;
    for (i = 0; i < FIXED_CB_SIZE; i++) {
        for (j = 0; j < BLOCKSIZE; j++)
            vect[j] = cb[i][j];
        get_match_score(work, coefs, vect, ortho1, ortho2, data, &score, &g);
        if (score > best_score) {
            best_score = score;
            *idx = i;
            *gain = g;
        }
    }
}


/**
270
 * Search the two fixed codebooks for the best entry and gain
271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321
 *
 * @param work array used to calculate LPC-filtered vectors
 * @param coefs coefficients of the LPC filter
 * @param data input data
 * @param cba_idx index of the best entry of the adaptive codebook
 * @param cb1_idx pointer to variable where the index of the best entry of the
 *        first fixed codebook is returned
 * @param cb2_idx pointer to variable where the index of the best entry of the
 *        second fixed codebook is returned
 */
static void fixed_cb_search(float *work, const float *coefs, float *data,
                            int cba_idx, int *cb1_idx, int *cb2_idx)
{
    int i, ortho_cb1;
    float gain;
    float cba_vect[BLOCKSIZE], cb1_vect[BLOCKSIZE];
    float vect[BLOCKSIZE];

    /**
     * The filtered vector from the adaptive codebook can be retrieved from
     * work, because this function is called just after adaptive_cb_search().
     */
    if (cba_idx)
        memcpy(cba_vect, work, sizeof(cba_vect));

    find_best_vect(work, coefs, ff_cb1_vects, cba_idx ? cba_vect : NULL, NULL,
                   data, cb1_idx, &gain);

    /**
     * Re-calculate the filtered vector from the vector with maximum match score
     * and remove its contribution from input data.
     */
    if (gain) {
        for (i = 0; i < BLOCKSIZE; i++)
            vect[i] = ff_cb1_vects[*cb1_idx][i];
        ff_celp_lp_synthesis_filterf(work, coefs, vect, BLOCKSIZE, LPC_ORDER);
        if (cba_idx)
            orthogonalize(work, cba_vect);
        for (i = 0; i < BLOCKSIZE; i++)
            data[i] -= gain * work[i];
        memcpy(cb1_vect, work, sizeof(cb1_vect));
        ortho_cb1 = 1;
    } else
        ortho_cb1 = 0;

    find_best_vect(work, coefs, ff_cb2_vects, cba_idx ? cba_vect : NULL,
                   ortho_cb1 ? cb1_vect : NULL, data, cb2_idx, &gain);
}


/**
322
 * Encode a subblock of the current frame
323 324 325 326 327 328 329 330 331 332 333 334
 *
 * @param ractx encoder context
 * @param sblock_data input data of the subblock
 * @param lpc_coefs coefficients of the LPC filter
 * @param rms RMS of the reflection coefficients
 * @param pb pointer to PutBitContext of the current frame
 */
static void ra144_encode_subblock(RA144Context *ractx,
                                  const int16_t *sblock_data,
                                  const int16_t *lpc_coefs, unsigned int rms,
                                  PutBitContext *pb)
{
335
    float data[BLOCKSIZE] = { 0 }, work[LPC_ORDER + BLOCKSIZE];
336 337 338
    float coefs[LPC_ORDER];
    float zero[BLOCKSIZE], cba[BLOCKSIZE], cb1[BLOCKSIZE], cb2[BLOCKSIZE];
    int cba_idx, cb1_idx, cb2_idx, gain;
339 340
    int i, n;
    unsigned m[3];
341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375
    float g[3];
    float error, best_error;

    for (i = 0; i < LPC_ORDER; i++) {
        work[i] = ractx->curr_sblock[BLOCKSIZE + i];
        coefs[i] = lpc_coefs[i] * (1/4096.0);
    }

    /**
     * Calculate the zero-input response of the LPC filter and subtract it from
     * input data.
     */
    ff_celp_lp_synthesis_filterf(work + LPC_ORDER, coefs, data, BLOCKSIZE,
                                 LPC_ORDER);
    for (i = 0; i < BLOCKSIZE; i++) {
        zero[i] = work[LPC_ORDER + i];
        data[i] = sblock_data[i] - zero[i];
    }

    /**
     * Codebook search is performed without taking into account the contribution
     * of the previous subblock, since it has been just subtracted from input
     * data.
     */
    memset(work, 0, LPC_ORDER * sizeof(*work));

    cba_idx = adaptive_cb_search(ractx->adapt_cb, work + LPC_ORDER, coefs,
                                 data);
    if (cba_idx) {
        /**
         * The filtered vector from the adaptive codebook can be retrieved from
         * work, see implementation of adaptive_cb_search().
         */
        memcpy(cba, work + LPC_ORDER, sizeof(cba));

376
        ff_copy_and_dup(ractx->buffer_a, ractx->adapt_cb, cba_idx + BLOCKSIZE / 2 - 1);
377
        m[0] = (ff_irms(&ractx->adsp, ractx->buffer_a) * rms) >> 12;
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429
    }
    fixed_cb_search(work + LPC_ORDER, coefs, data, cba_idx, &cb1_idx, &cb2_idx);
    for (i = 0; i < BLOCKSIZE; i++) {
        cb1[i] = ff_cb1_vects[cb1_idx][i];
        cb2[i] = ff_cb2_vects[cb2_idx][i];
    }
    ff_celp_lp_synthesis_filterf(work + LPC_ORDER, coefs, cb1, BLOCKSIZE,
                                 LPC_ORDER);
    memcpy(cb1, work + LPC_ORDER, sizeof(cb1));
    m[1] = (ff_cb1_base[cb1_idx] * rms) >> 8;
    ff_celp_lp_synthesis_filterf(work + LPC_ORDER, coefs, cb2, BLOCKSIZE,
                                 LPC_ORDER);
    memcpy(cb2, work + LPC_ORDER, sizeof(cb2));
    m[2] = (ff_cb2_base[cb2_idx] * rms) >> 8;
    best_error = FLT_MAX;
    gain = 0;
    for (n = 0; n < 256; n++) {
        g[1] = ((ff_gain_val_tab[n][1] * m[1]) >> ff_gain_exp_tab[n]) *
               (1/4096.0);
        g[2] = ((ff_gain_val_tab[n][2] * m[2]) >> ff_gain_exp_tab[n]) *
               (1/4096.0);
        error = 0;
        if (cba_idx) {
            g[0] = ((ff_gain_val_tab[n][0] * m[0]) >> ff_gain_exp_tab[n]) *
                   (1/4096.0);
            for (i = 0; i < BLOCKSIZE; i++) {
                data[i] = zero[i] + g[0] * cba[i] + g[1] * cb1[i] +
                          g[2] * cb2[i];
                error += (data[i] - sblock_data[i]) *
                         (data[i] - sblock_data[i]);
            }
        } else {
            for (i = 0; i < BLOCKSIZE; i++) {
                data[i] = zero[i] + g[1] * cb1[i] + g[2] * cb2[i];
                error += (data[i] - sblock_data[i]) *
                         (data[i] - sblock_data[i]);
            }
        }
        if (error < best_error) {
            best_error = error;
            gain = n;
        }
    }
    put_bits(pb, 7, cba_idx);
    put_bits(pb, 8, gain);
    put_bits(pb, 7, cb1_idx);
    put_bits(pb, 7, cb2_idx);
    ff_subblock_synthesis(ractx, lpc_coefs, cba_idx, cb1_idx, cb2_idx, rms,
                          gain);
}


430 431
static int ra144_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
                              const AVFrame *frame, int *got_packet_ptr)
432 433 434
{
    static const uint8_t sizes[LPC_ORDER] = {64, 32, 32, 16, 16, 8, 8, 8, 8, 4};
    static const uint8_t bit_sizes[LPC_ORDER] = {6, 5, 5, 4, 4, 3, 3, 3, 3, 2};
435
    RA144Context *ractx = avctx->priv_data;
436 437 438 439 440 441 442
    PutBitContext pb;
    int32_t lpc_data[NBLOCKS * BLOCKSIZE];
    int32_t lpc_coefs[LPC_ORDER][MAX_LPC_ORDER];
    int shift[LPC_ORDER];
    int16_t block_coefs[NBLOCKS][LPC_ORDER];
    int lpc_refl[LPC_ORDER];    /**< reflection coefficients of the frame */
    unsigned int refl_rms[NBLOCKS]; /**< RMS of the reflection coefficients */
443
    const int16_t *samples = frame ? (const int16_t *)frame->data[0] : NULL;
444
    int energy = 0;
445
    int i, idx, ret;
446

447 448 449
    if (ractx->last_frame)
        return 0;

450
    if ((ret = ff_alloc_packet2(avctx, avpkt, FRAME_SIZE)) < 0)
451
        return ret;
452 453 454 455 456 457 458 459 460 461 462 463

    /**
     * Since the LPC coefficients are calculated on a frame centered over the
     * fourth subframe, to encode a given frame, data from the next frame is
     * needed. In each call to this function, the previous frame (whose data are
     * saved in the encoder context) is encoded, and data from the current frame
     * are saved in the encoder context to be used in the next function call.
     */
    for (i = 0; i < (2 * BLOCKSIZE + BLOCKSIZE / 2); i++) {
        lpc_data[i] = ractx->curr_block[BLOCKSIZE + BLOCKSIZE / 2 + i];
        energy += (lpc_data[i] * lpc_data[i]) >> 4;
    }
464
    if (frame) {
465
        int j;
466
        for (j = 0; j < frame->nb_samples && i < NBLOCKS * BLOCKSIZE; i++, j++) {
467 468 469
            lpc_data[i] = samples[j] >> 2;
            energy += (lpc_data[i] * lpc_data[i]) >> 4;
        }
470
    }
471 472
    if (i < NBLOCKS * BLOCKSIZE)
        memset(&lpc_data[i], 0, (NBLOCKS * BLOCKSIZE - i) * sizeof(*lpc_data));
473 474 475
    energy = ff_energy_tab[quantize(ff_t_sqrt(energy >> 5) >> 10, ff_energy_tab,
                                    32)];

476
    ff_lpc_calc_coefs(&ractx->lpc_ctx, lpc_data, NBLOCKS * BLOCKSIZE, LPC_ORDER,
477
                      LPC_ORDER, 16, lpc_coefs, shift, FF_LPC_TYPE_LEVINSON,
478
                      0, ORDER_METHOD_EST, 12, 0);
479 480 481 482 483 484 485 486 487 488 489 490 491 492
    for (i = 0; i < LPC_ORDER; i++)
        block_coefs[NBLOCKS - 1][i] = -(lpc_coefs[LPC_ORDER - 1][i] <<
                                        (12 - shift[LPC_ORDER - 1]));

    /**
     * TODO: apply perceptual weighting of the input speech through bandwidth
     * expansion of the LPC filter.
     */

    if (ff_eval_refl(lpc_refl, block_coefs[NBLOCKS - 1], avctx)) {
        /**
         * The filter is unstable: use the coefficients of the previous frame.
         */
        ff_int_to_int16(block_coefs[NBLOCKS - 1], ractx->lpc_coef[1]);
493 494 495 496
        if (ff_eval_refl(lpc_refl, block_coefs[NBLOCKS - 1], avctx)) {
            /* the filter is still unstable. set reflection coeffs to zero. */
            memset(lpc_refl, 0, sizeof(lpc_refl));
        }
497
    }
498
    init_put_bits(&pb, avpkt->data, avpkt->size);
499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520
    for (i = 0; i < LPC_ORDER; i++) {
        idx = quantize(lpc_refl[i], ff_lpc_refl_cb[i], sizes[i]);
        put_bits(&pb, bit_sizes[i], idx);
        lpc_refl[i] = ff_lpc_refl_cb[i][idx];
    }
    ractx->lpc_refl_rms[0] = ff_rms(lpc_refl);
    ff_eval_coefs(ractx->lpc_coef[0], lpc_refl);
    refl_rms[0] = ff_interp(ractx, block_coefs[0], 1, 1, ractx->old_energy);
    refl_rms[1] = ff_interp(ractx, block_coefs[1], 2,
                            energy <= ractx->old_energy,
                            ff_t_sqrt(energy * ractx->old_energy) >> 12);
    refl_rms[2] = ff_interp(ractx, block_coefs[2], 3, 0, energy);
    refl_rms[3] = ff_rescale_rms(ractx->lpc_refl_rms[0], energy);
    ff_int_to_int16(block_coefs[NBLOCKS - 1], ractx->lpc_coef[0]);
    put_bits(&pb, 5, quantize(energy, ff_energy_tab, 32));
    for (i = 0; i < NBLOCKS; i++)
        ra144_encode_subblock(ractx, ractx->curr_block + i * BLOCKSIZE,
                              block_coefs[i], refl_rms[i], &pb);
    flush_put_bits(&pb);
    ractx->old_energy = energy;
    ractx->lpc_refl_rms[1] = ractx->lpc_refl_rms[0];
    FFSWAP(unsigned int *, ractx->lpc_coef[0], ractx->lpc_coef[1]);
521 522 523

    /* copy input samples to current block for processing in next call */
    i = 0;
524 525
    if (frame) {
        for (; i < frame->nb_samples; i++)
526
            ractx->curr_block[i] = samples[i] >> 2;
527

528
        if ((ret = ff_af_queue_add(&ractx->afq, frame)) < 0)
529
            return ret;
530 531 532 533 534
    } else
        ractx->last_frame = 1;
    memset(&ractx->curr_block[i], 0,
           (NBLOCKS * BLOCKSIZE - i) * sizeof(*ractx->curr_block));

535 536 537 538
    /* Get the next frame pts/duration */
    ff_af_queue_remove(&ractx->afq, avctx->frame_size, &avpkt->pts,
                       &avpkt->duration);

539
    avpkt->size = FRAME_SIZE;
540 541
    *got_packet_ptr = 1;
    return 0;
542 543 544
}


545 546
AVCodec ff_ra_144_encoder = {
    .name           = "real_144",
547
    .long_name      = NULL_IF_CONFIG_SMALL("RealAudio 1.0 (14.4K)"),
548
    .type           = AVMEDIA_TYPE_AUDIO,
549
    .id             = AV_CODEC_ID_RA_144,
550 551
    .priv_data_size = sizeof(RA144Context),
    .init           = ra144_encode_init,
552
    .encode2        = ra144_encode_frame,
553
    .close          = ra144_encode_close,
554
    .capabilities   = CODEC_CAP_DELAY | CODEC_CAP_SMALL_LAST_FRAME,
555 556
    .sample_fmts    = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
                                                     AV_SAMPLE_FMT_NONE },
557
    .supported_samplerates = (const int[]){ 8000, 0 },
558
    .channel_layouts = (const uint64_t[]) { AV_CH_LAYOUT_MONO, 0 },
559
};