Libav 0.7.1
|
00001 /* 00002 * G.729 decoder 00003 * Copyright (c) 2008 Vladimir Voroshilov 00004 * 00005 * This file is part of Libav. 00006 * 00007 * Libav is free software; you can redistribute it and/or 00008 * modify it under the terms of the GNU Lesser General Public 00009 * License as published by the Free Software Foundation; either 00010 * version 2.1 of the License, or (at your option) any later version. 00011 * 00012 * Libav is distributed in the hope that it will be useful, 00013 * but WITHOUT ANY WARRANTY; without even the implied warranty of 00014 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 00015 * Lesser General Public License for more details. 00016 * 00017 * You should have received a copy of the GNU Lesser General Public 00018 * License along with Libav; if not, write to the Free Software 00019 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 00020 */ 00021 #include <stdlib.h> 00022 #include <inttypes.h> 00023 #include <limits.h> 00024 #include <stdio.h> 00025 #include <string.h> 00026 #include <math.h> 00027 #include <assert.h> 00028 00029 #include "avcodec.h" 00030 #include "libavutil/avutil.h" 00031 #include "get_bits.h" 00032 00033 #include "lsp.h" 00034 #include "celp_math.h" 00035 #include "acelp_filters.h" 00036 #include "acelp_pitch_delay.h" 00037 #include "acelp_vectors.h" 00038 #include "g729data.h" 00039 00044 #define LSFQ_MIN 40 00045 00050 #define LSFQ_MAX 25681 00051 00056 #define LSFQ_DIFF_MIN 321 00057 00062 #define SHARP_MIN 3277 00063 00071 #define SHARP_MAX 13017 00072 00076 #define SUBFRAME_SIZE 40 00077 00078 00079 typedef struct { 00080 uint8_t ac_index_bits[2]; 00081 uint8_t parity_bit; 00082 uint8_t gc_1st_index_bits; 00083 uint8_t gc_2nd_index_bits; 00084 uint8_t fc_signs_bits; 00085 uint8_t fc_indexes_bits; 00086 } G729FormatDescription; 00087 00088 typedef struct { 00089 int pitch_delay_int_prev; 00090 00092 int16_t past_quantizer_output_buf[MA_NP + 1][10]; 00093 int16_t* past_quantizer_outputs[MA_NP + 1]; 00094 00095 int16_t lsfq[10]; 00096 int16_t lsp_buf[2][10]; 00097 int16_t *lsp[2]; 00098 } G729Context; 00099 00100 static const G729FormatDescription format_g729_8k = { 00101 .ac_index_bits = {8,5}, 00102 .parity_bit = 1, 00103 .gc_1st_index_bits = GC_1ST_IDX_BITS_8K, 00104 .gc_2nd_index_bits = GC_2ND_IDX_BITS_8K, 00105 .fc_signs_bits = 4, 00106 .fc_indexes_bits = 13, 00107 }; 00108 00109 static const G729FormatDescription format_g729d_6k4 = { 00110 .ac_index_bits = {8,4}, 00111 .parity_bit = 0, 00112 .gc_1st_index_bits = GC_1ST_IDX_BITS_6K4, 00113 .gc_2nd_index_bits = GC_2ND_IDX_BITS_6K4, 00114 .fc_signs_bits = 2, 00115 .fc_indexes_bits = 9, 00116 }; 00117 00121 static inline uint16_t g729_prng(uint16_t value) 00122 { 00123 return 31821 * value + 13849; 00124 } 00125 00129 static inline int get_parity(uint8_t value) 00130 { 00131 return (0x6996966996696996ULL >> (value >> 2)) & 1; 00132 } 00133 00134 static void lsf_decode(int16_t* lsfq, int16_t* past_quantizer_outputs[MA_NP + 1], 00135 int16_t ma_predictor, 00136 int16_t vq_1st, int16_t vq_2nd_low, int16_t vq_2nd_high) 00137 { 00138 int i,j; 00139 static const uint8_t min_distance[2]={10, 5}; //(2.13) 00140 int16_t* quantizer_output = past_quantizer_outputs[MA_NP]; 00141 00142 for (i = 0; i < 5; i++) { 00143 quantizer_output[i] = cb_lsp_1st[vq_1st][i ] + cb_lsp_2nd[vq_2nd_low ][i ]; 00144 quantizer_output[i + 5] = cb_lsp_1st[vq_1st][i + 5] + cb_lsp_2nd[vq_2nd_high][i + 5]; 00145 } 00146 00147 for (j = 0; j < 2; j++) { 00148 for (i = 1; i < 10; i++) { 00149 int diff = (quantizer_output[i - 1] - quantizer_output[i] + min_distance[j]) >> 1; 00150 if (diff > 0) { 00151 quantizer_output[i - 1] -= diff; 00152 quantizer_output[i ] += diff; 00153 } 00154 } 00155 } 00156 00157 for (i = 0; i < 10; i++) { 00158 int sum = quantizer_output[i] * cb_ma_predictor_sum[ma_predictor][i]; 00159 for (j = 0; j < MA_NP; j++) 00160 sum += past_quantizer_outputs[j][i] * cb_ma_predictor[ma_predictor][j][i]; 00161 00162 lsfq[i] = sum >> 15; 00163 } 00164 00165 /* Rotate past_quantizer_outputs. */ 00166 memmove(past_quantizer_outputs + 1, past_quantizer_outputs, MA_NP * sizeof(int16_t*)); 00167 past_quantizer_outputs[0] = quantizer_output; 00168 00169 ff_acelp_reorder_lsf(lsfq, LSFQ_DIFF_MIN, LSFQ_MIN, LSFQ_MAX, 10); 00170 } 00171 00172 static av_cold int decoder_init(AVCodecContext * avctx) 00173 { 00174 G729Context* ctx = avctx->priv_data; 00175 int i,k; 00176 00177 if (avctx->channels != 1) { 00178 av_log(avctx, AV_LOG_ERROR, "Only mono sound is supported (requested channels: %d).\n", avctx->channels); 00179 return AVERROR(EINVAL); 00180 } 00181 00182 /* Both 8kbit/s and 6.4kbit/s modes uses two subframes per frame. */ 00183 avctx->frame_size = SUBFRAME_SIZE << 1; 00184 00185 for (k = 0; k < MA_NP + 1; k++) { 00186 ctx->past_quantizer_outputs[k] = ctx->past_quantizer_output_buf[k]; 00187 for (i = 1; i < 11; i++) 00188 ctx->past_quantizer_outputs[k][i - 1] = (18717 * i) >> 3; 00189 } 00190 00191 ctx->lsp[0] = ctx->lsp_buf[0]; 00192 ctx->lsp[1] = ctx->lsp_buf[1]; 00193 memcpy(ctx->lsp[0], lsp_init, 10 * sizeof(int16_t)); 00194 00195 return 0; 00196 } 00197 00198 static int decode_frame(AVCodecContext *avctx, void *data, int *data_size, 00199 AVPacket *avpkt) 00200 { 00201 const uint8_t *buf = avpkt->data; 00202 int buf_size = avpkt->size; 00203 int16_t *out_frame = data; 00204 GetBitContext gb; 00205 G729FormatDescription format; 00206 int frame_erasure = 0; 00207 int bad_pitch = 0; 00208 int i; 00209 G729Context *ctx = avctx->priv_data; 00210 int16_t lp[2][11]; // (3.12) 00211 uint8_t ma_predictor; 00212 uint8_t quantizer_1st; 00213 uint8_t quantizer_2nd_lo; 00214 uint8_t quantizer_2nd_hi; 00215 00216 int pitch_delay_int; // pitch delay, integer part 00217 int pitch_delay_3x; // pitch delay, multiplied by 3 00218 00219 if (*data_size < SUBFRAME_SIZE << 2) { 00220 av_log(avctx, AV_LOG_ERROR, "Error processing packet: output buffer too small\n"); 00221 return AVERROR(EIO); 00222 } 00223 00224 if (buf_size == 10) { 00225 format = format_g729_8k; 00226 av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729 @ 8kbit/s"); 00227 } else if (buf_size == 8) { 00228 format = format_g729d_6k4; 00229 av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729D @ 6.4kbit/s"); 00230 } else { 00231 av_log(avctx, AV_LOG_ERROR, "Packet size %d is unknown.\n", buf_size); 00232 return AVERROR_INVALIDDATA; 00233 } 00234 00235 for (i=0; i < buf_size; i++) 00236 frame_erasure |= buf[i]; 00237 frame_erasure = !frame_erasure; 00238 00239 init_get_bits(&gb, buf, buf_size); 00240 00241 ma_predictor = get_bits(&gb, 1); 00242 quantizer_1st = get_bits(&gb, VQ_1ST_BITS); 00243 quantizer_2nd_lo = get_bits(&gb, VQ_2ND_BITS); 00244 quantizer_2nd_hi = get_bits(&gb, VQ_2ND_BITS); 00245 00246 lsf_decode(ctx->lsfq, ctx->past_quantizer_outputs, 00247 ma_predictor, 00248 quantizer_1st, quantizer_2nd_lo, quantizer_2nd_hi); 00249 00250 ff_acelp_lsf2lsp(ctx->lsp[1], ctx->lsfq, 10); 00251 00252 ff_acelp_lp_decode(&lp[0][0], &lp[1][0], ctx->lsp[1], ctx->lsp[0], 10); 00253 00254 FFSWAP(int16_t*, ctx->lsp[1], ctx->lsp[0]); 00255 00256 for (i = 0; i < 2; i++) { 00257 uint8_t ac_index; 00258 uint8_t pulses_signs; 00259 int fc_indexes; 00260 uint8_t gc_1st_index; 00261 uint8_t gc_2nd_index; 00262 00263 ac_index = get_bits(&gb, format.ac_index_bits[i]); 00264 if(!i && format.parity_bit) 00265 bad_pitch = get_parity(ac_index) == get_bits1(&gb); 00266 fc_indexes = get_bits(&gb, format.fc_indexes_bits); 00267 pulses_signs = get_bits(&gb, format.fc_signs_bits); 00268 gc_1st_index = get_bits(&gb, format.gc_1st_index_bits); 00269 gc_2nd_index = get_bits(&gb, format.gc_2nd_index_bits); 00270 00271 if(!i) { 00272 if (bad_pitch) 00273 pitch_delay_3x = 3 * ctx->pitch_delay_int_prev; 00274 else 00275 pitch_delay_3x = ff_acelp_decode_8bit_to_1st_delay3(ac_index); 00276 } else { 00277 int pitch_delay_min = av_clip(ctx->pitch_delay_int_prev - 5, 00278 PITCH_DELAY_MIN, PITCH_DELAY_MAX - 9); 00279 00280 if(packet_type == FORMAT_G729D_6K4) 00281 pitch_delay_3x = ff_acelp_decode_4bit_to_2nd_delay3(ac_index, pitch_delay_min); 00282 else 00283 pitch_delay_3x = ff_acelp_decode_5_6_bit_to_2nd_delay3(ac_index, pitch_delay_min); 00284 } 00285 00286 /* Round pitch delay to nearest (used everywhere except ff_acelp_interpolate). */ 00287 pitch_delay_int = (pitch_delay_3x + 1) / 3; 00288 00289 ff_acelp_weighted_vector_sum(fc + pitch_delay_int, 00290 fc + pitch_delay_int, 00291 fc, 1 << 14, 00292 av_clip(ctx->gain_pitch, SHARP_MIN, SHARP_MAX), 00293 0, 14, 00294 SUBFRAME_SIZE - pitch_delay_int); 00295 00296 if (frame_erasure) { 00297 ctx->gain_pitch = (29491 * ctx->gain_pitch) >> 15; // 0.90 (0.15) 00298 ctx->gain_code = ( 2007 * ctx->gain_code ) >> 11; // 0.98 (0.11) 00299 00300 gain_corr_factor = 0; 00301 } else { 00302 ctx->gain_pitch = cb_gain_1st_8k[gc_1st_index][0] + 00303 cb_gain_2nd_8k[gc_2nd_index][0]; 00304 gain_corr_factor = cb_gain_1st_8k[gc_1st_index][1] + 00305 cb_gain_2nd_8k[gc_2nd_index][1]; 00306 00307 ff_acelp_weighted_vector_sum(ctx->exc + i * SUBFRAME_SIZE, 00308 ctx->exc + i * SUBFRAME_SIZE, fc, 00309 (!voicing && frame_erasure) ? 0 : ctx->gain_pitch, 00310 ( voicing && frame_erasure) ? 0 : ctx->gain_code, 00311 1 << 13, 14, SUBFRAME_SIZE); 00312 00313 ctx->pitch_delay_int_prev = pitch_delay_int; 00314 } 00315 00316 *data_size = SUBFRAME_SIZE << 2; 00317 return buf_size; 00318 } 00319 00320 AVCodec ff_g729_decoder = 00321 { 00322 "g729", 00323 AVMEDIA_TYPE_AUDIO, 00324 CODEC_ID_G729, 00325 sizeof(G729Context), 00326 decoder_init, 00327 NULL, 00328 NULL, 00329 decode_frame, 00330 .long_name = NULL_IF_CONFIG_SMALL("G.729"), 00331 };