evrcdec.c 28.5 KB
Newer Older
Paul B Mahol's avatar
Paul B Mahol committed
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 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 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278
/*
 * Enhanced Variable Rate Codec, Service Option 3 decoder
 * Copyright (c) 2013 Paul B Mahol
 *
 * 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
 */

/**
 * @file
 * Enhanced Variable Rate Codec, Service Option 3 decoder
 * @author Paul B Mahol
 */

#include "libavutil/mathematics.h"
#include "avcodec.h"
#include "internal.h"
#include "get_bits.h"
#include "evrcdata.h"
#include "acelp_vectors.h"
#include "lsp.h"

#define MIN_LSP_SEP (0.05 / (2.0 * M_PI))
#define MIN_DELAY      20
#define MAX_DELAY     120
#define NB_SUBFRAMES    3
#define SUBFRAME_SIZE  54
#define FILTER_ORDER   10
#define ACB_SIZE      128

typedef enum {
    RATE_ERRS = -1,
    SILENCE,
    RATE_QUANT,
    RATE_QUARTER,
    RATE_HALF,
    RATE_FULL,
} evrc_packet_rate;

/**
 * EVRC-A unpacked data frame
 */
typedef struct EVRCAFrame {
    uint8_t  lpc_flag;        ///< spectral change indicator
    uint16_t lsp[4];          ///< index into LSP codebook
    uint8_t  pitch_delay;     ///< pitch delay for entire frame
    uint8_t  delay_diff;      ///< delay difference for entire frame
    uint8_t  acb_gain[3];     ///< adaptive codebook gain
    uint16_t fcb_shape[3][4]; ///< fixed codebook shape
    uint8_t  fcb_gain[3];     ///< fixed codebook gain index
    uint8_t  energy_gain;     ///< frame energy gain index
    uint8_t  tty;             ///< tty baud rate bit
} EVRCAFrame;

typedef struct EVRCContext {
    GetBitContext    gb;
    evrc_packet_rate bitrate;
    evrc_packet_rate last_valid_bitrate;
    EVRCAFrame       frame;

    float            lspf[FILTER_ORDER];
    float            prev_lspf[FILTER_ORDER];
    float            synthesis[FILTER_ORDER];
    float            postfilter_fir[FILTER_ORDER];
    float            postfilter_iir[FILTER_ORDER];
    float            postfilter_residual[ACB_SIZE + SUBFRAME_SIZE];
    float            pitch_delay;
    float            prev_pitch_delay;
    float            avg_acb_gain;  ///< average adaptive codebook gain
    float            avg_fcb_gain;  ///< average fixed codebook gain
    float            pitch[ACB_SIZE + FILTER_ORDER + SUBFRAME_SIZE];
    float            pitch_back[ACB_SIZE];
    float            interpolation_coeffs[136];
    float            energy_vector[NB_SUBFRAMES];
    float            fade_scale;
    float            last;

    uint8_t          prev_energy_gain;
    uint8_t          prev_error_flag;
    uint8_t          warned_buf_mismatch_bitrate;
} EVRCContext;

/**
 * Frame unpacking for RATE_FULL, RATE_HALF and RATE_QUANT
 *
 * @param e the context
 *
 * TIA/IS-127 Table 4.21-1
 */
static void unpack_frame(EVRCContext *e)
{
    EVRCAFrame *frame = &e->frame;
    GetBitContext *gb = &e->gb;

    switch (e->bitrate) {
    case RATE_FULL:
        frame->lpc_flag        = get_bits1(gb);
        frame->lsp[0]          = get_bits(gb,  6);
        frame->lsp[1]          = get_bits(gb,  6);
        frame->lsp[2]          = get_bits(gb,  9);
        frame->lsp[3]          = get_bits(gb,  7);
        frame->pitch_delay     = get_bits(gb,  7);
        frame->delay_diff      = get_bits(gb,  5);
        frame->acb_gain[0]     = get_bits(gb,  3);
        frame->fcb_shape[0][0] = get_bits(gb,  8);
        frame->fcb_shape[0][1] = get_bits(gb,  8);
        frame->fcb_shape[0][2] = get_bits(gb,  8);
        frame->fcb_shape[0][3] = get_bits(gb, 11);
        frame->fcb_gain[0]     = get_bits(gb,  5);
        frame->acb_gain[1]     = get_bits(gb,  3);
        frame->fcb_shape[1][0] = get_bits(gb,  8);
        frame->fcb_shape[1][1] = get_bits(gb,  8);
        frame->fcb_shape[1][2] = get_bits(gb,  8);
        frame->fcb_shape[1][3] = get_bits(gb, 11);
        frame->fcb_gain    [1] = get_bits(gb,  5);
        frame->acb_gain    [2] = get_bits(gb,  3);
        frame->fcb_shape[2][0] = get_bits(gb,  8);
        frame->fcb_shape[2][1] = get_bits(gb,  8);
        frame->fcb_shape[2][2] = get_bits(gb,  8);
        frame->fcb_shape[2][3] = get_bits(gb, 11);
        frame->fcb_gain    [2] = get_bits(gb,  5);
        frame->tty             = get_bits1(gb);
        break;
    case RATE_HALF:
        frame->lsp         [0] = get_bits(gb,  7);
        frame->lsp         [1] = get_bits(gb,  7);
        frame->lsp         [2] = get_bits(gb,  8);
        frame->pitch_delay     = get_bits(gb,  7);
        frame->acb_gain    [0] = get_bits(gb,  3);
        frame->fcb_shape[0][0] = get_bits(gb, 10);
        frame->fcb_gain    [0] = get_bits(gb,  4);
        frame->acb_gain    [1] = get_bits(gb,  3);
        frame->fcb_shape[1][0] = get_bits(gb, 10);
        frame->fcb_gain    [1] = get_bits(gb,  4);
        frame->acb_gain    [2] = get_bits(gb,  3);
        frame->fcb_shape[2][0] = get_bits(gb, 10);
        frame->fcb_gain    [2] = get_bits(gb,  4);
        break;
    case RATE_QUANT:
        frame->lsp         [0] = get_bits(gb, 4);
        frame->lsp         [1] = get_bits(gb, 4);
        frame->energy_gain     = get_bits(gb, 8);
        break;
    }
}

static evrc_packet_rate buf_size2bitrate(const int buf_size)
{
    switch (buf_size) {
    case 23: return RATE_FULL;
    case 11: return RATE_HALF;
    case  6: return RATE_QUARTER;
    case  3: return RATE_QUANT;
    case  1: return SILENCE;
    }

    return RATE_ERRS;
}

/**
 * Determine the bitrate from the frame size and/or the first byte of the frame.
 *
 * @param avctx the AV codec context
 * @param buf_size length of the buffer
 * @param buf the bufffer
 *
 * @return the bitrate on success,
 *         RATE_ERRS  if the bitrate cannot be satisfactorily determined
 */
static evrc_packet_rate determine_bitrate(AVCodecContext *avctx,
                                          int *buf_size,
                                          const uint8_t **buf)
{
    evrc_packet_rate bitrate;

    if ((bitrate = buf_size2bitrate(*buf_size)) >= 0) {
        if (bitrate > **buf) {
            EVRCContext *e = avctx->priv_data;
            if (!e->warned_buf_mismatch_bitrate) {
                av_log(avctx, AV_LOG_WARNING,
                       "Claimed bitrate and buffer size mismatch.\n");
                e->warned_buf_mismatch_bitrate = 1;
            }
            bitrate = **buf;
        } else if (bitrate < **buf) {
            av_log(avctx, AV_LOG_ERROR,
                   "Buffer is too small for the claimed bitrate.\n");
            return RATE_ERRS;
        }
        (*buf)++;
        *buf_size -= 1;
    } else if ((bitrate = buf_size2bitrate(*buf_size + 1)) >= 0) {
        av_log(avctx, AV_LOG_DEBUG,
               "Bitrate byte is missing, guessing the bitrate from packet size.\n");
    } else
        return RATE_ERRS;

    return bitrate;
}

static void warn_insufficient_frame_quality(AVCodecContext *avctx,
                                            const char *message)
{
    av_log(avctx, AV_LOG_WARNING, "Frame #%d, %s\n",
           avctx->frame_number, message);
}

/**
 * Initialize the speech codec according to the specification.
 *
 * TIA/IS-127 5.2
 */
static av_cold int evrc_decode_init(AVCodecContext *avctx)
{
    EVRCContext *e = avctx->priv_data;
    int i, n, idx = 0;
    float denom = 2.0 / (2.0 * 8.0 + 1.0);

    avctx->channels       = 1;
    avctx->channel_layout = AV_CH_LAYOUT_MONO;
    avctx->sample_fmt     = AV_SAMPLE_FMT_FLT;

    for (i = 0; i < FILTER_ORDER; i++) {
        e->prev_lspf[i] = (i + 1) * 0.048;
        e->synthesis[i] = 0.0;
    }

    for (i = 0; i < ACB_SIZE; i++)
        e->pitch[i] = e->pitch_back[i] = 0.0;

    e->last_valid_bitrate = RATE_QUANT;
    e->prev_pitch_delay   = 40.0;
    e->fade_scale         = 1.0;
    e->prev_error_flag    = 0;
    e->avg_acb_gain = e->avg_fcb_gain = 0.0;

    for (i = 0; i < 8; i++) {
        float tt = ((float)i - 8.0 / 2.0) / 8.0;

        for (n = -8; n <= 8; n++, idx++) {
            float arg1 = M_PI * 0.9 * (tt - n);
            float arg2 = M_PI * (tt - n);

            e->interpolation_coeffs[idx] = 0.9;
            if (arg1)
                e->interpolation_coeffs[idx] *= (0.54 + 0.46 * cos(arg2 * denom)) *
                                                 sin(arg1) / arg1;
        }
    }

    return 0;
}

/**
 * Decode the 10 vector quantized line spectral pair frequencies from the LSP
 * transmission codes of any bitrate and check for badly received packets.
 *
 * @param e the context
 *
 * @return 0 on success, -1 if the packet is badly received
 *
 * TIA/IS-127 5.2.1, 5.7.1
 */
static int decode_lspf(EVRCContext *e)
{
279
    const float * const *codebooks = evrc_lspq_codebooks[e->bitrate];
Paul B Mahol's avatar
Paul B Mahol committed
280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376
    int i, j, k = 0;

    for (i = 0; i < evrc_lspq_nb_codebooks[e->bitrate]; i++) {
        int row_size = evrc_lspq_codebooks_row_sizes[e->bitrate][i];
        const float *codebook = codebooks[i];

        for (j = 0; j < row_size; j++)
            e->lspf[k++] = codebook[e->frame.lsp[i] * row_size + j];
    }

    // check for monotonic LSPs
    for (i = 1; i < FILTER_ORDER; i++)
        if (e->lspf[i] <= e->lspf[i - 1])
            return -1;

    // check for minimum separation of LSPs at the splits
    for (i = 0, k = 0; i < evrc_lspq_nb_codebooks[e->bitrate] - 1; i++) {
        k += evrc_lspq_codebooks_row_sizes[e->bitrate][i];
        if (e->lspf[k] - e->lspf[k - 1] <= MIN_LSP_SEP)
            return -1;
    }

    return 0;
}

/*
 * Interpolation of LSP parameters.
 *
 * TIA/IS-127 5.2.3.1, 5.7.3.2
 */
static void interpolate_lsp(float *ilsp, const float *lsp,
                            const float *prev, int index)
{
    static const float lsp_interpolation_factors[] = { 0.1667, 0.5, 0.8333 };
    ff_weighted_vector_sumf(ilsp, prev, lsp,
                            1.0 - lsp_interpolation_factors[index],
                            lsp_interpolation_factors[index], FILTER_ORDER);
}

/*
 * Reconstruction of the delay contour.
 *
 * TIA/IS-127 5.2.2.3.2
 */
static void interpolate_delay(float *dst, float current, float prev, int index)
{
    static const float d_interpolation_factors[] = { 0, 0.3313, 0.6625, 1, 1 };
    dst[0] = (1.0 - d_interpolation_factors[index    ]) * prev
                  + d_interpolation_factors[index    ]  * current;
    dst[1] = (1.0 - d_interpolation_factors[index + 1]) * prev
                  + d_interpolation_factors[index + 1]  * current;
    dst[2] = (1.0 - d_interpolation_factors[index + 2]) * prev
                  + d_interpolation_factors[index + 2]  * current;
}

/*
 * Convert the quantized, interpolated line spectral frequencies,
 * to prediction coefficients.
 *
 * TIA/IS-127 5.2.3.2, 4.7.2.2
 */
static void decode_predictor_coeffs(const float *ilspf, float *ilpc)
{
    double lsp[FILTER_ORDER];
    float a[FILTER_ORDER / 2 + 1], b[FILTER_ORDER / 2 + 1];
    float a1[FILTER_ORDER / 2] = { 0 };
    float a2[FILTER_ORDER / 2] = { 0 };
    float b1[FILTER_ORDER / 2] = { 0 };
    float b2[FILTER_ORDER / 2] = { 0 };
    int i, k;

    ff_acelp_lsf2lspd(lsp, ilspf, FILTER_ORDER);

    for (k = 0; k <= FILTER_ORDER; k++) {
        a[0] = k < 2 ? 0.25 : 0;
        b[0] = k < 2 ? k < 1 ? 0.25 : -0.25 : 0;

        for (i = 0; i < FILTER_ORDER / 2; i++) {
            a[i + 1] = a[i] - 2 * lsp[i * 2    ] * a1[i] + a2[i];
            b[i + 1] = b[i] - 2 * lsp[i * 2 + 1] * b1[i] + b2[i];
            a2[i] = a1[i];
            a1[i] = a[i];
            b2[i] = b1[i];
            b1[i] = b[i];
        }

        if (k)
            ilpc[k - 1] = 2.0 * (a[FILTER_ORDER / 2] + b[FILTER_ORDER / 2]);
    }
}

static void bl_intrp(EVRCContext *e, float *ex, float delay)
{
    float *f;
    int offset, i, coef_idx;
    int16_t t;

377
    offset = lrintf(delay);
Paul B Mahol's avatar
Paul B Mahol committed
378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642

    t = (offset - delay + 0.5) * 8.0 + 0.5;
    if (t == 8) {
        t = 0;
        offset--;
    }

    f = ex - offset - 8;

    coef_idx = t * (2 * 8 + 1);

    ex[0] = 0.0;
    for (i = 0; i < 2 * 8 + 1; i++)
        ex[0] += e->interpolation_coeffs[coef_idx + i] * f[i];
}

/*
 * Adaptive codebook excitation.
 *
 * TIA/IS-127 5.2.2.3.3, 4.12.5.2
 */
static void acb_excitation(EVRCContext *e, float *excitation, float gain,
                           const float delay[3], int length)
{
    float denom, locdelay, dpr, invl;
    int i;

    invl = 1.0 / ((float) length);
    dpr = length;

    /* first at-most extra samples */
    denom = (delay[1] - delay[0]) * invl;
    for (i = 0; i < dpr; i++) {
        locdelay = delay[0] + i * denom;
        bl_intrp(e, excitation + i, locdelay);
    }

    denom = (delay[2] - delay[1]) * invl;
    /* interpolation */
    for (i = dpr; i < dpr + 10; i++) {
        locdelay = delay[1] + (i - dpr) * denom;
        bl_intrp(e, excitation + i, locdelay);
    }

    for (i = 0; i < length; i++)
        excitation[i] *= gain;
}

static void decode_8_pulses_35bits(const uint16_t *fixed_index, float *cod)
{
    int i, pos1, pos2, offset;

    offset = (fixed_index[3] >> 9) & 3;

    for (i = 0; i < 3; i++) {
        pos1 = ((fixed_index[i] & 0x7f) / 11) * 5 + ((i + offset) % 5);
        pos2 = ((fixed_index[i] & 0x7f) % 11) * 5 + ((i + offset) % 5);

        cod[pos1] = (fixed_index[i] & 0x80) ? -1.0 : 1.0;

        if (pos2 < pos1)
            cod[pos2]  = -cod[pos1];
        else
            cod[pos2] +=  cod[pos1];
    }

    pos1 = ((fixed_index[3] & 0x7f) / 11) * 5 + ((3 + offset) % 5);
    pos2 = ((fixed_index[3] & 0x7f) % 11) * 5 + ((4 + offset) % 5);

    cod[pos1] = (fixed_index[3] & 0x100) ? -1.0 : 1.0;
    cod[pos2] = (fixed_index[3] & 0x80 ) ? -1.0 : 1.0;
}

static void decode_3_pulses_10bits(uint16_t fixed_index, float *cod)
{
    float sign;
    int pos;

    sign = (fixed_index & 0x200) ? -1.0 : 1.0;

    pos = ((fixed_index        & 0x7) * 7) + 4;
    cod[pos] += sign;
    pos = (((fixed_index >> 3) & 0x7) * 7) + 2;
    cod[pos] -= sign;
    pos = (((fixed_index >> 6) & 0x7) * 7);
    cod[pos] += sign;
}

/*
 * Reconstruction of ACELP fixed codebook excitation for full and half rate.
 *
 * TIA/IS-127 5.2.3.7
 */
static void fcb_excitation(EVRCContext *e, const uint16_t *codebook,
                           float *excitation, float pitch_gain,
                           int pitch_lag, int subframe_size)
{
    int i;

    if (e->bitrate == RATE_FULL)
        decode_8_pulses_35bits(codebook, excitation);
    else
        decode_3_pulses_10bits(*codebook, excitation);

    pitch_gain = av_clipf(pitch_gain, 0.2, 0.9);

    for (i = pitch_lag; i < subframe_size; i++)
        excitation[i] += pitch_gain * excitation[i - pitch_lag];
}

/**
 * Synthesis of the decoder output signal.
 *
 * param[in]     in              input signal
 * param[in]     filter_coeffs   LPC coefficients
 * param[in/out] memory          synthesis filter memory
 * param         buffer_length   amount of data to process
 * param[out]    samples         output samples
 *
 * TIA/IS-127 5.2.3.15, 5.7.3.4
 */
static void synthesis_filter(const float *in, const float *filter_coeffs,
                             float *memory, int buffer_length, float *samples)
{
    int i, j;

    for (i = 0; i < buffer_length; i++) {
        samples[i] = in[i];
        for (j = FILTER_ORDER - 1; j > 0; j--) {
            samples[i] -= filter_coeffs[j] * memory[j];
            memory[j]   = memory[j - 1];
        }
        samples[i] -= filter_coeffs[0] * memory[0];
        memory[0]   = samples[i];
    }
}

static void bandwidth_expansion(float *coeff, const float *inbuf, float gamma)
{
    double fac = gamma;
    int i;

    for (i = 0; i < FILTER_ORDER; i++) {
        coeff[i] = inbuf[i] * fac;
        fac *= gamma;
    }
}

static void residual_filter(float *output, const float *input,
                            const float *coef, float *memory, int length)
{
    float sum;
    int i, j;

    for (i = 0; i < length; i++) {
        sum = input[i];

        for (j = FILTER_ORDER - 1; j > 0; j--) {
            sum      += coef[j] * memory[j];
            memory[j] = memory[j - 1];
        }
        sum += coef[0] * memory[0];
        memory[0] = input[i];
        output[i] = sum;
    }
}

/*
 * TIA/IS-127 Table 5.9.1-1.
 */
static const struct PfCoeff {
    float tilt;
    float ltgain;
    float p1;
    float p2;
} postfilter_coeffs[5] = {
    { 0.0 , 0.0 , 0.0 , 0.0  },
    { 0.0 , 0.0 , 0.57, 0.57 },
    { 0.0 , 0.0 , 0.0 , 0.0  },
    { 0.35, 0.50, 0.50, 0.75 },
    { 0.20, 0.50, 0.57, 0.75 },
};

/*
 * Adaptive postfilter.
 *
 * TIA/IS-127 5.9
 */
static void postfilter(EVRCContext *e, float *in, const float *coeff,
                       float *out, int idx, const struct PfCoeff *pfc,
                       int length)
{
    float wcoef1[FILTER_ORDER], wcoef2[FILTER_ORDER],
          scratch[SUBFRAME_SIZE], temp[SUBFRAME_SIZE],
          mem[SUBFRAME_SIZE];
    float sum1 = 0.0, sum2 = 0.0, gamma, gain;
    float tilt = pfc->tilt;
    int i, n, best;

    bandwidth_expansion(wcoef1, coeff, pfc->p1);
    bandwidth_expansion(wcoef2, coeff, pfc->p2);

    /* Tilt compensation filter, TIA/IS-127 5.9.1 */
    for (i = 0; i < length - 1; i++)
        sum2 += in[i] * in[i + 1];
    if (sum2 < 0.0)
        tilt = 0.0;

    for (i = 0; i < length; i++) {
        scratch[i] = in[i] - tilt * e->last;
        e->last = in[i];
    }

    /* Short term residual filter, TIA/IS-127 5.9.2 */
    residual_filter(&e->postfilter_residual[ACB_SIZE], scratch, wcoef1, e->postfilter_fir, length);

    /* Long term postfilter */
    best = idx;
    for (i = FFMIN(MIN_DELAY, idx - 3); i <= FFMAX(MAX_DELAY, idx + 3); i++) {
        for (n = ACB_SIZE, sum2 = 0; n < ACB_SIZE + length; n++)
            sum2 += e->postfilter_residual[n] * e->postfilter_residual[n - i];
        if (sum2 > sum1) {
            sum1 = sum2;
            best = i;
        }
    }

    for (i = ACB_SIZE, sum1 = 0; i < ACB_SIZE + length; i++)
        sum1 += e->postfilter_residual[i - best] * e->postfilter_residual[i - best];
    for (i = ACB_SIZE, sum2 = 0; i < ACB_SIZE + length; i++)
        sum2 += e->postfilter_residual[i] * e->postfilter_residual[i - best];

    if (sum2 * sum1 == 0 || e->bitrate == RATE_QUANT) {
        memcpy(temp, e->postfilter_residual + ACB_SIZE, length * sizeof(float));
    } else {
        gamma = sum2 / sum1;
        if (gamma < 0.5)
            memcpy(temp, e->postfilter_residual + ACB_SIZE, length * sizeof(float));
        else {
            gamma = FFMIN(gamma, 1.0);

            for (i = 0; i < length; i++) {
                temp[i] = e->postfilter_residual[ACB_SIZE + i] + gamma *
                    pfc->ltgain * e->postfilter_residual[ACB_SIZE + i - best];
            }
        }
    }

    memcpy(scratch, temp, length * sizeof(float));
    memcpy(mem, e->postfilter_iir, FILTER_ORDER * sizeof(float));
    synthesis_filter(scratch, wcoef2, mem, length, scratch);

    /* Gain computation, TIA/IS-127 5.9.4-2 */
    for (i = 0, sum1 = 0, sum2 = 0; i < length; i++) {
        sum1 += in[i] * in[i];
        sum2 += scratch[i] * scratch[i];
    }
    gain = sum2 ? sqrt(sum1 / sum2) : 1.0;

    for (i = 0; i < length; i++)
        temp[i] *= gain;

    /* Short term postfilter */
    synthesis_filter(temp, wcoef2, e->postfilter_iir, length, out);

643
    memmove(e->postfilter_residual,
Paul B Mahol's avatar
Paul B Mahol committed
644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716
           e->postfilter_residual + length, ACB_SIZE * sizeof(float));
}

static void frame_erasure(EVRCContext *e, float *samples)
{
    float ilspf[FILTER_ORDER], ilpc[FILTER_ORDER], idelay[NB_SUBFRAMES],
          tmp[SUBFRAME_SIZE + 6], f;
    int i, j;

    for (i = 0; i < FILTER_ORDER; i++) {
        if (e->bitrate != RATE_QUANT)
            e->lspf[i] = e->prev_lspf[i] * 0.875 + 0.125 * (i + 1) * 0.048;
        else
            e->lspf[i] = e->prev_lspf[i];
    }

    if (e->prev_error_flag)
        e->avg_acb_gain *= 0.75;
    if (e->bitrate == RATE_FULL)
        memcpy(e->pitch_back, e->pitch, ACB_SIZE * sizeof(float));
    if (e->last_valid_bitrate == RATE_QUANT)
        e->bitrate = RATE_QUANT;
    else
        e->bitrate = RATE_FULL;

    if (e->bitrate == RATE_FULL || e->bitrate == RATE_HALF) {
        e->pitch_delay = e->prev_pitch_delay;
    } else {
        float sum = 0;

        idelay[0] = idelay[1] = idelay[2] = MIN_DELAY;

        for (i = 0; i < NB_SUBFRAMES; i++)
            sum += evrc_energy_quant[e->prev_energy_gain][i];
        sum /= (float) NB_SUBFRAMES;
        sum  = pow(10, sum);
        for (i = 0; i < NB_SUBFRAMES; i++)
            e->energy_vector[i] = sum;
    }

    if (fabs(e->pitch_delay - e->prev_pitch_delay) > 15)
        e->prev_pitch_delay = e->pitch_delay;

    for (i = 0; i < NB_SUBFRAMES; i++) {
        int subframe_size = subframe_sizes[i];
        int pitch_lag;

        interpolate_lsp(ilspf, e->lspf, e->prev_lspf, i);

        if (e->bitrate != RATE_QUANT) {
            if (e->avg_acb_gain < 0.3) {
                idelay[0] = estimation_delay[i];
                idelay[1] = estimation_delay[i + 1];
                idelay[2] = estimation_delay[i + 2];
            } else {
                interpolate_delay(idelay, e->pitch_delay, e->prev_pitch_delay, i);
            }
        }

        pitch_lag = lrintf((idelay[1] + idelay[0]) / 2.0);
        decode_predictor_coeffs(ilspf, ilpc);

        if (e->bitrate != RATE_QUANT) {
            acb_excitation(e, e->pitch + ACB_SIZE,
                           e->avg_acb_gain, idelay, subframe_size);
            for (j = 0; j < subframe_size; j++)
                e->pitch[ACB_SIZE + j] *= e->fade_scale;
            e->fade_scale = FFMAX(e->fade_scale - 0.05, 0.0);
        } else {
            for (j = 0; j < subframe_size; j++)
                e->pitch[ACB_SIZE + j] = e->energy_vector[i];
        }

717
        memmove(e->pitch, e->pitch + subframe_size, ACB_SIZE * sizeof(float));
Paul B Mahol's avatar
Paul B Mahol committed
718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740

        if (e->bitrate != RATE_QUANT && e->avg_acb_gain < 0.4) {
            f = 0.1 * e->avg_fcb_gain;
            for (j = 0; j < subframe_size; j++)
                e->pitch[ACB_SIZE + j] += f;
        } else if (e->bitrate == RATE_QUANT) {
            for (j = 0; j < subframe_size; j++)
                e->pitch[ACB_SIZE + j] = e->energy_vector[i];
        }

        synthesis_filter(e->pitch + ACB_SIZE, ilpc,
                         e->synthesis, subframe_size, tmp);
        postfilter(e, tmp, ilpc, samples, pitch_lag,
                   &postfilter_coeffs[e->bitrate], subframe_size);

        samples += subframe_size;
    }
}

static int evrc_decode_frame(AVCodecContext *avctx, void *data,
                             int *got_frame_ptr, AVPacket *avpkt)
{
    const uint8_t *buf = avpkt->data;
741
    AVFrame *frame     = data;
Paul B Mahol's avatar
Paul B Mahol committed
742 743 744 745 746 747
    EVRCContext *e     = avctx->priv_data;
    int buf_size       = avpkt->size;
    float ilspf[FILTER_ORDER], ilpc[FILTER_ORDER], idelay[NB_SUBFRAMES];
    float *samples;
    int   i, j, ret, error_flag = 0;

748
    frame->nb_samples = 160;
749
    if ((ret = ff_get_buffer(avctx, frame, 0)) < 0)
Paul B Mahol's avatar
Paul B Mahol committed
750
        return ret;
751
    samples = (float *)frame->data[0];
Paul B Mahol's avatar
Paul B Mahol committed
752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775

    if ((e->bitrate = determine_bitrate(avctx, &buf_size, &buf)) == RATE_ERRS) {
        warn_insufficient_frame_quality(avctx, "bitrate cannot be determined.");
        goto erasure;
    }
    if (e->bitrate <= SILENCE || e->bitrate == RATE_QUARTER)
        goto erasure;
    if (e->bitrate == RATE_QUANT && e->last_valid_bitrate == RATE_FULL
                                 && !e->prev_error_flag)
        goto erasure;

    init_get_bits(&e->gb, buf, 8 * buf_size);
    memset(&e->frame, 0, sizeof(EVRCAFrame));

    unpack_frame(e);

    if (e->bitrate != RATE_QUANT) {
        uint8_t *p = (uint8_t *) &e->frame;
        for (i = 0; i < sizeof(EVRCAFrame); i++) {
            if (p[i])
                break;
        }
        if (i == sizeof(EVRCAFrame))
            goto erasure;
776 777
    } else if (e->frame.lsp[0] == 0xf &&
               e->frame.lsp[1] == 0xf &&
Paul B Mahol's avatar
Paul B Mahol committed
778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816
               e->frame.energy_gain == 0xff) {
        goto erasure;
    }

    if (decode_lspf(e) < 0)
        goto erasure;

    if (e->bitrate == RATE_FULL || e->bitrate == RATE_HALF) {
        /* Pitch delay parameter checking as per TIA/IS-127 5.1.5.1 */
        if (e->frame.pitch_delay > MAX_DELAY - MIN_DELAY)
            goto erasure;

        e->pitch_delay = e->frame.pitch_delay + MIN_DELAY;

        /* Delay diff parameter checking as per TIA/IS-127 5.1.5.2 */
        if (e->frame.delay_diff) {
            int p = e->pitch_delay - e->frame.delay_diff + 16;
            if (p < MIN_DELAY || p > MAX_DELAY)
                goto erasure;
        }

        /* Delay contour reconstruction as per TIA/IS-127 5.2.2.2 */
        if (e->frame.delay_diff &&
            e->bitrate == RATE_FULL && e->prev_error_flag) {
            float delay;

            memcpy(e->pitch, e->pitch_back, ACB_SIZE * sizeof(float));

            delay = e->prev_pitch_delay;
            e->prev_pitch_delay = delay - e->frame.delay_diff + 16.0;

            if (fabs(e->pitch_delay - delay) > 15)
                delay = e->pitch_delay;

            for (i = 0; i < NB_SUBFRAMES; i++) {
                int subframe_size = subframe_sizes[i];

                interpolate_delay(idelay, delay, e->prev_pitch_delay, i);
                acb_excitation(e, e->pitch + ACB_SIZE, e->avg_acb_gain, idelay, subframe_size);
817
                memmove(e->pitch, e->pitch + subframe_size, ACB_SIZE * sizeof(float));
Paul B Mahol's avatar
Paul B Mahol committed
818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874
            }
        }

        /* Smoothing of the decoded delay as per TIA/IS-127 5.2.2.5 */
        if (fabs(e->pitch_delay - e->prev_pitch_delay) > 15)
            e->prev_pitch_delay = e->pitch_delay;

        e->avg_acb_gain = e->avg_fcb_gain = 0.0;
    } else {
        idelay[0] = idelay[1] = idelay[2] = MIN_DELAY;

        /* Decode frame energy vectors as per TIA/IS-127 5.7.2 */
        for (i = 0; i < NB_SUBFRAMES; i++)
            e->energy_vector[i] = pow(10, evrc_energy_quant[e->frame.energy_gain][i]);
        e->prev_energy_gain = e->frame.energy_gain;
    }

    for (i = 0; i < NB_SUBFRAMES; i++) {
        float tmp[SUBFRAME_SIZE + 6] = { 0 };
        int subframe_size = subframe_sizes[i];
        int pitch_lag;

        interpolate_lsp(ilspf, e->lspf, e->prev_lspf, i);

        if (e->bitrate != RATE_QUANT)
            interpolate_delay(idelay, e->pitch_delay, e->prev_pitch_delay, i);

        pitch_lag = lrintf((idelay[1] + idelay[0]) / 2.0);
        decode_predictor_coeffs(ilspf, ilpc);

        /* Bandwidth expansion as per TIA/IS-127 5.2.3.3 */
        if (e->frame.lpc_flag && e->prev_error_flag)
            bandwidth_expansion(ilpc, ilpc, 0.75);

        if (e->bitrate != RATE_QUANT) {
            float acb_sum, f;

            f = exp((e->bitrate == RATE_HALF ? 0.5 : 0.25)
                         * (e->frame.fcb_gain[i] + 1));
            acb_sum = pitch_gain_vq[e->frame.acb_gain[i]];
            e->avg_acb_gain += acb_sum / NB_SUBFRAMES;
            e->avg_fcb_gain += f / NB_SUBFRAMES;

            acb_excitation(e, e->pitch + ACB_SIZE,
                           acb_sum, idelay, subframe_size);
            fcb_excitation(e, e->frame.fcb_shape[i], tmp,
                           acb_sum, pitch_lag, subframe_size);

            /* Total excitation generation as per TIA/IS-127 5.2.3.9 */
            for (j = 0; j < subframe_size; j++)
                e->pitch[ACB_SIZE + j] += f * tmp[j];
            e->fade_scale = FFMIN(e->fade_scale + 0.2, 1.0);
        } else {
            for (j = 0; j < subframe_size; j++)
                e->pitch[ACB_SIZE + j] = e->energy_vector[i];
        }

875
        memmove(e->pitch, e->pitch + subframe_size, ACB_SIZE * sizeof(float));
Paul B Mahol's avatar
Paul B Mahol committed
876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898

        synthesis_filter(e->pitch + ACB_SIZE, ilpc,
                         e->synthesis, subframe_size, tmp);
        postfilter(e, tmp, ilpc, samples, pitch_lag,
                   &postfilter_coeffs[e->bitrate], subframe_size);

        samples += subframe_size;
    }

    if (error_flag) {
erasure:
        error_flag = 1;
        av_log(avctx, AV_LOG_WARNING, "frame erasure\n");
        frame_erasure(e, samples);
    }

    memcpy(e->prev_lspf, e->lspf, sizeof(e->prev_lspf));
    e->prev_error_flag    = error_flag;
    e->last_valid_bitrate = e->bitrate;

    if (e->bitrate != RATE_QUANT)
        e->prev_pitch_delay = e->pitch_delay;

899
    samples = (float *)frame->data[0];
Paul B Mahol's avatar
Paul B Mahol committed
900 901 902 903 904 905 906 907 908 909
    for (i = 0; i < 160; i++)
        samples[i] /= 32768;

    *got_frame_ptr   = 1;

    return avpkt->size;
}

AVCodec ff_evrc_decoder = {
    .name           = "evrc",
910
    .long_name      = NULL_IF_CONFIG_SMALL("EVRC (Enhanced Variable Rate Codec)"),
Paul B Mahol's avatar
Paul B Mahol committed
911 912 913 914 915 916 917
    .type           = AVMEDIA_TYPE_AUDIO,
    .id             = AV_CODEC_ID_EVRC,
    .init           = evrc_decode_init,
    .decode         = evrc_decode_frame,
    .capabilities   = CODEC_CAP_DR1,
    .priv_data_size = sizeof(EVRCContext),
};