libavcodec/g729dec.c

   1 /*
   2  * G.729 decoder
   3  * Copyright (c) 2008 Vladimir Voroshilov
   4  *
   5  * This file is part of FFmpeg.
   6  *
   7  * FFmpeg 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  * FFmpeg 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 FFmpeg; if not, write to the Free Software
  19  * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
  20  */
  21 #include <stdlib.h>
  22 #include <inttypes.h>
  23 #include <limits.h>
  24 #include <stdio.h>
  25 #include <string.h>
  26 #include <math.h>
  27 #include <assert.h>
  28
  29 #include "avcodec.h"
  30 #include "libavutil/avutil.h"
  31 #include "get_bits.h"
  32
  33 #include "g729.h"
  34 #include "lsp.h"
  35 #include "celp_math.h"
  36 #include "acelp_filters.h"
  37 #include "acelp_pitch_delay.h"
  38 #include "acelp_vectors.h"
  39 #include "g729data.h"
  40
  41 /**
  42  * minimum quantized LSF value (3.2.4)
  43  * 0.005 in Q13
  44  */
  45 #define LSFQ_MIN                   40
  46
  47 /**
  48  * maximum quantized LSF value (3.2.4)
  49  * 3.135 in Q13
  50  */
  51 #define LSFQ_MAX                   25681
  52
  53 /**
  54  * minimum LSF distance (3.2.4)
  55  * 0.0391 in Q13
  56  */
  57 #define LSFQ_DIFF_MIN              321
  58
  59 /**
  60  * minimum gain pitch value (3.8, Equation 47)
  61  * 0.2 in (1.14)
  62  */
  63 #define SHARP_MIN                  3277
  64
  65 /**
  66  * maximum gain pitch value (3.8, Equation 47)
  67  * (EE) This does not comply with the specification.
  68  * Specification says about 0.8, which should be
  69  * 13107 in (1.14), but reference C code uses
  70  * 13017 (equals to 0.7945) instead of it.
  71  */
  72 #define SHARP_MAX                  13017
  73
  74 typedef struct {
  75     uint8_t ac_index_bits[2];   ///< adaptive codebook index for second subframe (size in bits)
  76     uint8_t parity_bit;         ///< parity bit for pitch delay
  77     uint8_t gc_1st_index_bits;  ///< gain codebook (first stage) index (size in bits)
  78     uint8_t gc_2nd_index_bits;  ///< gain codebook (second stage) index (size in bits)
  79     uint8_t fc_signs_bits;      ///< number of pulses in fixed-codebook vector
  80     uint8_t fc_indexes_bits;    ///< size (in bits) of fixed-codebook index entry
  81 } G729FormatDescription;
  82
  83 typedef struct {
  84     int pitch_delay_int_prev;   ///< integer part of previous subframe's pitch delay (4.1.3)
  85
  86     /// (2.13) LSP quantizer outputs
  87     int16_t  past_quantizer_output_buf[MA_NP + 1][10];
  88     int16_t* past_quantizer_outputs[MA_NP + 1];
  89
  90     int16_t lsfq[10];           ///< (2.13) quantized LSF coefficients from previous frame
  91     int16_t lsp_buf[2][10];     ///< (0.15) LSP coefficients (previous and current frames) (3.2.5)
  92     int16_t *lsp[2];            ///< pointers to lsp_buf
  93 }  G729Context;
  94
  95 static const G729FormatDescription format_g729_8k = {
  96     .ac_index_bits     = {8,5},
  97     .parity_bit        = 1,
  98     .gc_1st_index_bits = GC_1ST_IDX_BITS_8K,
  99     .gc_2nd_index_bits = GC_2ND_IDX_BITS_8K,
 100     .fc_signs_bits     = 4,
 101     .fc_indexes_bits   = 13,
 102 };
 103
 104 static const G729FormatDescription format_g729d_6k4 = {
 105     .ac_index_bits     = {8,4},
 106     .parity_bit        = 0,
 107     .gc_1st_index_bits = GC_1ST_IDX_BITS_6K4,
 108     .gc_2nd_index_bits = GC_2ND_IDX_BITS_6K4,
 109     .fc_signs_bits     = 2,
 110     .fc_indexes_bits   = 9,
 111 };
 112
 113 /**
 114  * \brief pseudo random number generator
 115  */
 116 static inline uint16_t g729_prng(uint16_t value)
 117 {
 118     return 31821 * value + 13849;
 119 }
 120
 121 /**
 122  * Get parity bit of bit 2..7
 123  */
 124 static inline int get_parity(uint8_t value)
 125 {
 126    return (0x6996966996696996ULL >> (value >> 2)) & 1;
 127 }
 128
 129 static void lsf_decode(int16_t* lsfq, int16_t* past_quantizer_outputs[MA_NP + 1],
 130                        int16_t ma_predictor,
 131                        int16_t vq_1st, int16_t vq_2nd_low, int16_t vq_2nd_high)
 132 {
 133     int i,j;
 134     static const uint8_t min_distance[2]={10, 5}; //(2.13)
 135     int16_t* quantizer_output = past_quantizer_outputs[MA_NP];
 136
 137     for (i = 0; i < 5; i++) {
 138         quantizer_output[i]     = cb_lsp_1st[vq_1st][i    ] + cb_lsp_2nd[vq_2nd_low ][i    ];
 139         quantizer_output[i + 5] = cb_lsp_1st[vq_1st][i + 5] + cb_lsp_2nd[vq_2nd_high][i + 5];
 140     }
 141
 142     for (j = 0; j < 2; j++) {
 143         for (i = 1; i < 10; i++) {
 144             int diff = (quantizer_output[i - 1] - quantizer_output[i] + min_distance[j]) >> 1;
 145             if (diff > 0) {
 146                 quantizer_output[i - 1] -= diff;
 147                 quantizer_output[i    ] += diff;
 148             }
 149         }
 150     }
 151
 152     for (i = 0; i < 10; i++) {
 153         int sum = quantizer_output[i] * cb_ma_predictor_sum[ma_predictor][i];
 154         for (j = 0; j < MA_NP; j++)
 155             sum += past_quantizer_outputs[j][i] * cb_ma_predictor[ma_predictor][j][i];
 156
 157         lsfq[i] = sum >> 15;
 158     }
 159
 160     /* Rotate past_quantizer_outputs. */
 161     memmove(past_quantizer_outputs + 1, past_quantizer_outputs, MA_NP * sizeof(int16_t*));
 162     past_quantizer_outputs[0] = quantizer_output;
 163
 164     ff_acelp_reorder_lsf(lsfq, LSFQ_DIFF_MIN, LSFQ_MIN, LSFQ_MAX, 10);
 165 }
 166
 167 static av_cold int decoder_init(AVCodecContext * avctx)
 168 {
 169     G729Context* ctx = avctx->priv_data;
 170     int i,k;
 171
 172     if (avctx->channels != 1) {
 173         av_log(avctx, AV_LOG_ERROR, "Only mono sound is supported (requested channels: %d).\n", avctx->channels);
 174         return AVERROR_NOFMT;
 175     }
 176
 177     /* Both 8kbit/s and 6.4kbit/s modes uses two subframes per frame. */
 178     avctx->frame_size = SUBFRAME_SIZE << 1;
 179
 180     for (k = 0; k < MA_NP + 1; k++) {
 181         ctx->past_quantizer_outputs[k] = ctx->past_quantizer_output_buf[k];
 182         for (i = 1; i < 11; i++)
 183             ctx->past_quantizer_outputs[k][i - 1] = (18717 * i) >> 3;
 184     }
 185
 186     ctx->lsp[0] = ctx->lsp_buf[0];
 187     ctx->lsp[1] = ctx->lsp_buf[1];
 188     memcpy(ctx->lsp[0], lsp_init, 10 * sizeof(int16_t));
 189
 190     return 0;
 191 }
 192
 193 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size,
 194                         AVPacket *avpkt)
 195 {
 196     const uint8_t *buf = avpkt->data;
 197     int buf_size       = avpkt->size;
 198     int16_t *out_frame = data;
 199     GetBitContext gb;
 200     G729FormatDescription format;
 201     int frame_erasure = 0;    ///< frame erasure detected during decoding
 202     int bad_pitch = 0;        ///< parity check failed
 203     int i;
 204     G729Context *ctx = avctx->priv_data;
 205     int16_t lp[2][11];           // (3.12)
 206     uint8_t ma_predictor;     ///< switched MA predictor of LSP quantizer
 207     uint8_t quantizer_1st;    ///< first stage vector of quantizer
 208     uint8_t quantizer_2nd_lo; ///< second stage lower vector of quantizer (size in bits)
 209     uint8_t quantizer_2nd_hi; ///< second stage higher vector of quantizer (size in bits)
 210
 211     int pitch_delay_int;         // pitch delay, integer part
 212     int pitch_delay_3x;          // pitch delay, multiplied by 3
 213
 214     if (*data_size < SUBFRAME_SIZE << 2) {
 215         av_log(avctx, AV_LOG_ERROR, "Error processing packet: output buffer too small\n");
 216         return AVERROR(EIO);
 217     }
 218
 219     if (buf_size == 10) {
 220         format = format_g729_8k;
 221         av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729 @ 8kbit/s");
 222     } else if (buf_size == 8) {
 223         format = format_g729d_6k4;
 224         av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729D @ 6.4kbit/s");
 225     } else {
 226         av_log(avctx, AV_LOG_ERROR, "Packet size %d is unknown.\n", buf_size);
 227         return (AVERROR_NOFMT);
 228     }
 229
 230     for (i=0; i < buf_size; i++)
 231         frame_erasure |= buf[i];
 232     frame_erasure = !frame_erasure;
 233
 234     init_get_bits(&gb, buf, buf_size);
 235
 236     ma_predictor     = get_bits(&gb, 1);
 237     quantizer_1st    = get_bits(&gb, VQ_1ST_BITS);
 238     quantizer_2nd_lo = get_bits(&gb, VQ_2ND_BITS);
 239     quantizer_2nd_hi = get_bits(&gb, VQ_2ND_BITS);
 240
 241     lsf_decode(ctx->lsfq, ctx->past_quantizer_outputs,
 242                ma_predictor,
 243                quantizer_1st, quantizer_2nd_lo, quantizer_2nd_hi);
 244
 245     ff_acelp_lsf2lsp(ctx->lsp[1], ctx->lsfq, 10);
 246
 247     ff_acelp_lp_decode(&lp[0][0], &lp[1][0], ctx->lsp[1], ctx->lsp[0], 10);
 248
 249     FFSWAP(int16_t*, ctx->lsp[1], ctx->lsp[0]);
 250
 251     for (i = 0; i < 2; i++) {
 252         uint8_t ac_index;      ///< adaptive codebook index
 253         uint8_t pulses_signs;  ///< fixed-codebook vector pulse signs
 254         int fc_indexes;        ///< fixed-codebook indexes
 255         uint8_t gc_1st_index;  ///< gain codebook (first stage) index
 256         uint8_t gc_2nd_index;  ///< gain codebook (second stage) index
 257
 258         ac_index      = get_bits(&gb, format.ac_index_bits[i]);
 259         if(!i && format.parity_bit)
 260             bad_pitch = get_parity(ac_index) == get_bits1(&gb);
 261         fc_indexes    = get_bits(&gb, format.fc_indexes_bits);
 262         pulses_signs  = get_bits(&gb, format.fc_signs_bits);
 263         gc_1st_index  = get_bits(&gb, format.gc_1st_index_bits);
 264         gc_2nd_index  = get_bits(&gb, format.gc_2nd_index_bits);
 265
 266         if(!i) {
 267             if (bad_pitch)
 268                 pitch_delay_3x   = 3 * ctx->pitch_delay_int_prev;
 269             else
 270                 pitch_delay_3x = ff_acelp_decode_8bit_to_1st_delay3(ac_index);
 271         } else {
 272             int pitch_delay_min = av_clip(ctx->pitch_delay_int_prev - 5,
 273                                           PITCH_DELAY_MIN, PITCH_DELAY_MAX - 9);
 274
 275             if(packet_type == FORMAT_G729D_6K4)
 276                 pitch_delay_3x = ff_acelp_decode_4bit_to_2nd_delay3(ac_index, pitch_delay_min);
 277             else
 278                 pitch_delay_3x = ff_acelp_decode_5_6_bit_to_2nd_delay3(ac_index, pitch_delay_min);
 279         }
 280
 281         /* Round pitch delay to nearest (used everywhere except ff_acelp_interpolate). */
 282         pitch_delay_int  = (pitch_delay_3x + 1) / 3;
 283
 284         ff_acelp_weighted_vector_sum(fc + pitch_delay_int,
 285                                      fc + pitch_delay_int,
 286                                      fc, 1 << 14,
 287                                      av_clip(ctx->gain_pitch, SHARP_MIN, SHARP_MAX),
 288                                      0, 14,
 289                                      SUBFRAME_SIZE - pitch_delay_int);
 290
 291         if (frame_erasure) {
 292             ctx->gain_pitch = (29491 * ctx->gain_pitch) >> 15; // 0.90 (0.15)
 293             ctx->gain_code  = ( 2007 * ctx->gain_code ) >> 11; // 0.98 (0.11)
 294
 295             gain_corr_factor = 0;
 296         } else {
 297             ctx->gain_pitch  = cb_gain_1st_8k[gc_1st_index][0] +
 298                                cb_gain_2nd_8k[gc_2nd_index][0];
 299             gain_corr_factor = cb_gain_1st_8k[gc_1st_index][1] +
 300                                cb_gain_2nd_8k[gc_2nd_index][1];
 301
 302         ff_acelp_weighted_vector_sum(ctx->exc + i * SUBFRAME_SIZE,
 303                                      ctx->exc + i * SUBFRAME_SIZE, fc,
 304                                      (!voicing && frame_erasure) ? 0 : ctx->gain_pitch,
 305                                      ( voicing && frame_erasure) ? 0 : ctx->gain_code,
 306                                      1 << 13, 14, SUBFRAME_SIZE);
 307
 308             ctx->pitch_delay_int_prev = pitch_delay_int;
 309     }
 310
 311     *data_size = SUBFRAME_SIZE << 2;
 312     return buf_size;
 313 }
 314
 315 AVCodec g729_decoder =
 316 {
 317     "g729",
 318     CODEC_TYPE_AUDIO,
 319     CODEC_ID_G729,
 320     sizeof(G729Context),
 321     decoder_init,
 322     NULL,
 323     NULL,
 324     decode_frame,
 325     .long_name = NULL_IF_CONFIG_SMALL("G.729"),
 326 };