X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=libavcodec%2Fg729dec.c;h=25951716eff9c47b6bc6d018b2c9e1769d6fa20a;hb=5e546864b09379910721b35a14713982d933d9dd;hp=2e1bf18e4e49f40dc501938ee05172beac53f94d;hpb=8e789d244cc946bc350672eeb02453918b21a09f;p=ffmpeg diff --git a/libavcodec/g729dec.c b/libavcodec/g729dec.c index 2e1bf18e4e4..25951716eff 100644 --- a/libavcodec/g729dec.c +++ b/libavcodec/g729dec.c @@ -100,8 +100,6 @@ typedef struct { } G729FormatDescription; typedef struct { - AudioDSPContext adsp; - /// past excitation signal buffer int16_t exc_base[2*SUBFRAME_SIZE+PITCH_DELAY_MAX+INTERPOL_LEN]; @@ -152,7 +150,13 @@ typedef struct { /// high-pass filter data (past output) int16_t hpf_z[2]; -} G729Context; +} G729ChannelContext; + +typedef struct { + AudioDSPContext adsp; + + G729ChannelContext *channel_context; +} G729Context; static const G729FormatDescription format_g729_8k = { .ac_index_bits = {8,5}, @@ -268,8 +272,7 @@ static void g729d_get_new_exc( ff_celp_convolve_circ(fc_new, fc_cur, phase_filter[dstate], subframe_size); - for(i=0; i> 14; out[i] += (gain_code * fc_new[i] + 0x2000) >> 14; @@ -285,10 +288,10 @@ static void g729d_get_new_exc( */ static int g729d_onset_decision(int past_onset, const int16_t* past_gain_code) { - if((past_gain_code[0] >> 1) > past_gain_code[1]) + if ((past_gain_code[0] >> 1) > past_gain_code[1]) return 2; - else - return FFMAX(past_onset-1, 0); + + return FFMAX(past_onset-1, 0); } /** @@ -303,24 +306,25 @@ static int16_t g729d_voice_decision(int onset, int prev_voice_decision, const in { int i, low_gain_pitch_cnt, voice_decision; - if(past_gain_pitch[0] >= 14745) // 0.9 + if (past_gain_pitch[0] >= 14745) { // 0.9 voice_decision = DECISION_VOICE; - else if (past_gain_pitch[0] <= 9830) // 0.6 + } else if (past_gain_pitch[0] <= 9830) { // 0.6 voice_decision = DECISION_NOISE; - else + } else { voice_decision = DECISION_INTERMEDIATE; + } - for(i=0, low_gain_pitch_cnt=0; i<6; i++) - if(past_gain_pitch[i] < 9830) + for (i = 0, low_gain_pitch_cnt = 0; i < 6; i++) + if (past_gain_pitch[i] < 9830) low_gain_pitch_cnt++; - if(low_gain_pitch_cnt > 2 && !onset) + if (low_gain_pitch_cnt > 2 && !onset) voice_decision = DECISION_NOISE; - if(!onset && voice_decision > prev_voice_decision + 1) + if (!onset && voice_decision > prev_voice_decision + 1) voice_decision--; - if(onset && voice_decision < DECISION_VOICE) + if (onset && voice_decision < DECISION_VOICE) voice_decision++; return voice_decision; @@ -338,43 +342,53 @@ static int32_t scalarproduct_int16_c(const int16_t * v1, const int16_t * v2, int static av_cold int decoder_init(AVCodecContext * avctx) { - G729Context* ctx = avctx->priv_data; - int i,k; + G729Context *s = avctx->priv_data; + G729ChannelContext *ctx; + int c,i,k; - if (avctx->channels != 1) { - av_log(avctx, AV_LOG_ERROR, "Only mono sound is supported (requested channels: %d).\n", avctx->channels); + if (avctx->channels < 1 || avctx->channels > 2) { + av_log(avctx, AV_LOG_ERROR, "Only mono and stereo are supported (requested channels: %d).\n", avctx->channels); return AVERROR(EINVAL); } - avctx->sample_fmt = AV_SAMPLE_FMT_S16; + avctx->sample_fmt = AV_SAMPLE_FMT_S16P; /* Both 8kbit/s and 6.4kbit/s modes uses two subframes per frame. */ avctx->frame_size = SUBFRAME_SIZE << 1; - ctx->gain_coeff = 16384; // 1.0 in (1.14) + ctx = + s->channel_context = av_mallocz(sizeof(G729ChannelContext) * avctx->channels); + if (!ctx) + return AVERROR(ENOMEM); - for (k = 0; k < MA_NP + 1; k++) { - ctx->past_quantizer_outputs[k] = ctx->past_quantizer_output_buf[k]; - for (i = 1; i < 11; i++) - ctx->past_quantizer_outputs[k][i - 1] = (18717 * i) >> 3; - } + for (c = 0; c < avctx->channels; c++) { + ctx->gain_coeff = 16384; // 1.0 in (1.14) - ctx->lsp[0] = ctx->lsp_buf[0]; - ctx->lsp[1] = ctx->lsp_buf[1]; - memcpy(ctx->lsp[0], lsp_init, 10 * sizeof(int16_t)); + for (k = 0; k < MA_NP + 1; k++) { + ctx->past_quantizer_outputs[k] = ctx->past_quantizer_output_buf[k]; + for (i = 1; i < 11; i++) + ctx->past_quantizer_outputs[k][i - 1] = (18717 * i) >> 3; + } - ctx->exc = &ctx->exc_base[PITCH_DELAY_MAX+INTERPOL_LEN]; + ctx->lsp[0] = ctx->lsp_buf[0]; + ctx->lsp[1] = ctx->lsp_buf[1]; + memcpy(ctx->lsp[0], lsp_init, 10 * sizeof(int16_t)); - ctx->pitch_delay_int_prev = PITCH_DELAY_MIN; + ctx->exc = &ctx->exc_base[PITCH_DELAY_MAX+INTERPOL_LEN]; - /* random seed initialization */ - ctx->rand_value = 21845; + ctx->pitch_delay_int_prev = PITCH_DELAY_MIN; - /* quantized prediction error */ - for(i=0; i<4; i++) - ctx->quant_energy[i] = -14336; // -14 in (5.10) + /* random seed initialization */ + ctx->rand_value = 21845; - ff_audiodsp_init(&ctx->adsp); - ctx->adsp.scalarproduct_int16 = scalarproduct_int16_c; + /* quantized prediction error */ + for (i = 0; i < 4; i++) + ctx->quant_energy[i] = -14336; // -14 in (5.10) + + ctx++; + } + + ff_audiodsp_init(&s->adsp); + s->adsp.scalarproduct_int16 = scalarproduct_int16_c; return 0; } @@ -387,12 +401,11 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr, int16_t *out_frame; GetBitContext gb; const G729FormatDescription *format; - int frame_erasure = 0; ///< frame erasure detected during decoding - int bad_pitch = 0; ///< parity check failed - int i; + int c, i; int16_t *tmp; G729Formats packet_type; - G729Context *ctx = avctx->priv_data; + G729Context *s = avctx->priv_data; + G729ChannelContext *ctx = s->channel_context; int16_t lp[2][11]; // (3.12) uint8_t ma_predictor; ///< switched MA predictor of LSP quantizer uint8_t quantizer_1st; ///< first stage vector of quantizer @@ -405,22 +418,20 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr, int16_t synth[SUBFRAME_SIZE+10]; // fixed-codebook vector int j, ret; int gain_before, gain_after; - int is_periodic = 0; // whether one of the subframes is declared as periodic or not AVFrame *frame = data; frame->nb_samples = SUBFRAME_SIZE<<1; if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) return ret; - out_frame = (int16_t*) frame->data[0]; - if (buf_size % 10 == 0) { + if (buf_size % ((G729_8K_BLOCK_SIZE + (avctx->codec_id == AV_CODEC_ID_ACELP_KELVIN)) * avctx->channels) == 0) { packet_type = FORMAT_G729_8K; format = &format_g729_8k; //Reset voice decision ctx->onset = 0; ctx->voice_decision = DECISION_VOICE; av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729 @ 8kbit/s"); - } else if (buf_size == 8) { + } else if (buf_size == G729D_6K4_BLOCK_SIZE * avctx->channels) { packet_type = FORMAT_G729D_6K4; format = &format_g729d_6k4; av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729D @ 6.4kbit/s"); @@ -429,281 +440,307 @@ static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr, return AVERROR_INVALIDDATA; } - for (i=0; i < buf_size; i++) - frame_erasure |= buf[i]; - frame_erasure = !frame_erasure; - - init_get_bits(&gb, buf, 8*buf_size); - - ma_predictor = get_bits(&gb, 1); - quantizer_1st = get_bits(&gb, VQ_1ST_BITS); - quantizer_2nd_lo = get_bits(&gb, VQ_2ND_BITS); - quantizer_2nd_hi = get_bits(&gb, VQ_2ND_BITS); - - if(frame_erasure) - lsf_restore_from_previous(ctx->lsfq, ctx->past_quantizer_outputs, - ctx->ma_predictor_prev); - else { - lsf_decode(ctx->lsfq, ctx->past_quantizer_outputs, - ma_predictor, - quantizer_1st, quantizer_2nd_lo, quantizer_2nd_hi); - ctx->ma_predictor_prev = ma_predictor; - } + for (c = 0; c < avctx->channels; c++) { + int frame_erasure = 0; ///< frame erasure detected during decoding + int bad_pitch = 0; ///< parity check failed + int is_periodic = 0; ///< whether one of the subframes is declared as periodic or not + out_frame = (int16_t*)frame->data[c]; + if (avctx->codec_id == AV_CODEC_ID_ACELP_KELVIN) { + if (*buf != ((avctx->channels - 1 - c) * 0x80 | 2)) + avpriv_request_sample(avctx, "First byte value %x for channel %d", *buf, c); + buf++; + } - tmp = ctx->past_quantizer_outputs[MA_NP]; - memmove(ctx->past_quantizer_outputs + 1, ctx->past_quantizer_outputs, - MA_NP * sizeof(int16_t*)); - ctx->past_quantizer_outputs[0] = tmp; - - ff_acelp_lsf2lsp(ctx->lsp[1], ctx->lsfq, 10); - - ff_acelp_lp_decode(&lp[0][0], &lp[1][0], ctx->lsp[1], ctx->lsp[0], 10); - - FFSWAP(int16_t*, ctx->lsp[1], ctx->lsp[0]); - - for (i = 0; i < 2; i++) { - int gain_corr_factor; - - uint8_t ac_index; ///< adaptive codebook index - uint8_t pulses_signs; ///< fixed-codebook vector pulse signs - int fc_indexes; ///< fixed-codebook indexes - uint8_t gc_1st_index; ///< gain codebook (first stage) index - uint8_t gc_2nd_index; ///< gain codebook (second stage) index - - ac_index = get_bits(&gb, format->ac_index_bits[i]); - if(!i && format->parity_bit) - bad_pitch = av_parity(ac_index >> 2) == get_bits1(&gb); - fc_indexes = get_bits(&gb, format->fc_indexes_bits); - pulses_signs = get_bits(&gb, format->fc_signs_bits); - gc_1st_index = get_bits(&gb, format->gc_1st_index_bits); - gc_2nd_index = get_bits(&gb, format->gc_2nd_index_bits); - - if (frame_erasure) - pitch_delay_3x = 3 * ctx->pitch_delay_int_prev; - else if(!i) { - if (bad_pitch) - pitch_delay_3x = 3 * ctx->pitch_delay_int_prev; - else - pitch_delay_3x = ff_acelp_decode_8bit_to_1st_delay3(ac_index); - } else { - int pitch_delay_min = av_clip(ctx->pitch_delay_int_prev - 5, - PITCH_DELAY_MIN, PITCH_DELAY_MAX - 9); + for (i = 0; i < buf_size; i++) + frame_erasure |= buf[i]; + frame_erasure = !frame_erasure; - if(packet_type == FORMAT_G729D_6K4) - pitch_delay_3x = ff_acelp_decode_4bit_to_2nd_delay3(ac_index, pitch_delay_min); - else - pitch_delay_3x = ff_acelp_decode_5_6_bit_to_2nd_delay3(ac_index, pitch_delay_min); - } + init_get_bits(&gb, buf, 8*buf_size); - /* Round pitch delay to nearest (used everywhere except ff_acelp_interpolate). */ - pitch_delay_int[i] = (pitch_delay_3x + 1) / 3; - if (pitch_delay_int[i] > PITCH_DELAY_MAX) { - av_log(avctx, AV_LOG_WARNING, "pitch_delay_int %d is too large\n", pitch_delay_int[i]); - pitch_delay_int[i] = PITCH_DELAY_MAX; - } + ma_predictor = get_bits(&gb, 1); + quantizer_1st = get_bits(&gb, VQ_1ST_BITS); + quantizer_2nd_lo = get_bits(&gb, VQ_2ND_BITS); + quantizer_2nd_hi = get_bits(&gb, VQ_2ND_BITS); if (frame_erasure) { - ctx->rand_value = g729_prng(ctx->rand_value); - fc_indexes = av_mod_uintp2(ctx->rand_value, format->fc_indexes_bits); - - ctx->rand_value = g729_prng(ctx->rand_value); - pulses_signs = ctx->rand_value; + lsf_restore_from_previous(ctx->lsfq, ctx->past_quantizer_outputs, + ctx->ma_predictor_prev); + } else { + lsf_decode(ctx->lsfq, ctx->past_quantizer_outputs, + ma_predictor, + quantizer_1st, quantizer_2nd_lo, quantizer_2nd_hi); + ctx->ma_predictor_prev = ma_predictor; } + tmp = ctx->past_quantizer_outputs[MA_NP]; + memmove(ctx->past_quantizer_outputs + 1, ctx->past_quantizer_outputs, + MA_NP * sizeof(int16_t*)); + ctx->past_quantizer_outputs[0] = tmp; + + ff_acelp_lsf2lsp(ctx->lsp[1], ctx->lsfq, 10); + + ff_acelp_lp_decode(&lp[0][0], &lp[1][0], ctx->lsp[1], ctx->lsp[0], 10); + + FFSWAP(int16_t*, ctx->lsp[1], ctx->lsp[0]); + + for (i = 0; i < 2; i++) { + int gain_corr_factor; + + uint8_t ac_index; ///< adaptive codebook index + uint8_t pulses_signs; ///< fixed-codebook vector pulse signs + int fc_indexes; ///< fixed-codebook indexes + uint8_t gc_1st_index; ///< gain codebook (first stage) index + uint8_t gc_2nd_index; ///< gain codebook (second stage) index + + ac_index = get_bits(&gb, format->ac_index_bits[i]); + if (!i && format->parity_bit) + bad_pitch = av_parity(ac_index >> 2) == get_bits1(&gb); + fc_indexes = get_bits(&gb, format->fc_indexes_bits); + pulses_signs = get_bits(&gb, format->fc_signs_bits); + gc_1st_index = get_bits(&gb, format->gc_1st_index_bits); + gc_2nd_index = get_bits(&gb, format->gc_2nd_index_bits); + + if (frame_erasure) { + pitch_delay_3x = 3 * ctx->pitch_delay_int_prev; + } else if (!i) { + if (bad_pitch) { + pitch_delay_3x = 3 * ctx->pitch_delay_int_prev; + } else { + pitch_delay_3x = ff_acelp_decode_8bit_to_1st_delay3(ac_index); + } + } else { + int pitch_delay_min = av_clip(ctx->pitch_delay_int_prev - 5, + PITCH_DELAY_MIN, PITCH_DELAY_MAX - 9); + + if (packet_type == FORMAT_G729D_6K4) { + pitch_delay_3x = ff_acelp_decode_4bit_to_2nd_delay3(ac_index, pitch_delay_min); + } else { + pitch_delay_3x = ff_acelp_decode_5_6_bit_to_2nd_delay3(ac_index, pitch_delay_min); + } + } - memset(fc, 0, sizeof(int16_t) * SUBFRAME_SIZE); - switch (packet_type) { - case FORMAT_G729_8K: - ff_acelp_fc_pulse_per_track(fc, ff_fc_4pulses_8bits_tracks_13, - ff_fc_4pulses_8bits_track_4, - fc_indexes, pulses_signs, 3, 3); - break; - case FORMAT_G729D_6K4: - ff_acelp_fc_pulse_per_track(fc, ff_fc_2pulses_9bits_track1_gray, - ff_fc_2pulses_9bits_track2_gray, - fc_indexes, pulses_signs, 1, 4); - break; - } - - /* - This filter enhances harmonic components of the fixed-codebook vector to - improve the quality of the reconstructed speech. + /* Round pitch delay to nearest (used everywhere except ff_acelp_interpolate). */ + pitch_delay_int[i] = (pitch_delay_3x + 1) / 3; + if (pitch_delay_int[i] > PITCH_DELAY_MAX) { + av_log(avctx, AV_LOG_WARNING, "pitch_delay_int %d is too large\n", pitch_delay_int[i]); + pitch_delay_int[i] = PITCH_DELAY_MAX; + } - / fc_v[i], i < pitch_delay - fc_v[i] = < - \ fc_v[i] + gain_pitch * fc_v[i-pitch_delay], i >= pitch_delay - */ - ff_acelp_weighted_vector_sum(fc + pitch_delay_int[i], - fc + pitch_delay_int[i], - fc, 1 << 14, - av_clip(ctx->past_gain_pitch[0], SHARP_MIN, SHARP_MAX), - 0, 14, - SUBFRAME_SIZE - pitch_delay_int[i]); + if (frame_erasure) { + ctx->rand_value = g729_prng(ctx->rand_value); + fc_indexes = av_mod_uintp2(ctx->rand_value, format->fc_indexes_bits); - memmove(ctx->past_gain_pitch+1, ctx->past_gain_pitch, 5 * sizeof(int16_t)); - ctx->past_gain_code[1] = ctx->past_gain_code[0]; + ctx->rand_value = g729_prng(ctx->rand_value); + pulses_signs = ctx->rand_value; + } - if (frame_erasure) { - ctx->past_gain_pitch[0] = (29491 * ctx->past_gain_pitch[0]) >> 15; // 0.90 (0.15) - ctx->past_gain_code[0] = ( 2007 * ctx->past_gain_code[0] ) >> 11; // 0.98 (0.11) - gain_corr_factor = 0; - } else { - if (packet_type == FORMAT_G729D_6K4) { - ctx->past_gain_pitch[0] = cb_gain_1st_6k4[gc_1st_index][0] + - cb_gain_2nd_6k4[gc_2nd_index][0]; - gain_corr_factor = cb_gain_1st_6k4[gc_1st_index][1] + - cb_gain_2nd_6k4[gc_2nd_index][1]; - - /* Without check below overflow can occur in ff_acelp_update_past_gain. - It is not issue for G.729, because gain_corr_factor in it's case is always - greater than 1024, while in G.729D it can be even zero. */ - gain_corr_factor = FFMAX(gain_corr_factor, 1024); -#ifndef G729_BITEXACT - gain_corr_factor >>= 1; -#endif - } else { - ctx->past_gain_pitch[0] = cb_gain_1st_8k[gc_1st_index][0] + - cb_gain_2nd_8k[gc_2nd_index][0]; - gain_corr_factor = cb_gain_1st_8k[gc_1st_index][1] + - cb_gain_2nd_8k[gc_2nd_index][1]; + memset(fc, 0, sizeof(int16_t) * SUBFRAME_SIZE); + switch (packet_type) { + case FORMAT_G729_8K: + ff_acelp_fc_pulse_per_track(fc, ff_fc_4pulses_8bits_tracks_13, + ff_fc_4pulses_8bits_track_4, + fc_indexes, pulses_signs, 3, 3); + break; + case FORMAT_G729D_6K4: + ff_acelp_fc_pulse_per_track(fc, ff_fc_2pulses_9bits_track1_gray, + ff_fc_2pulses_9bits_track2_gray, + fc_indexes, pulses_signs, 1, 4); + break; } - /* Decode the fixed-codebook gain. */ - ctx->past_gain_code[0] = ff_acelp_decode_gain_code(&ctx->adsp, gain_corr_factor, - fc, MR_ENERGY, - ctx->quant_energy, - ma_prediction_coeff, - SUBFRAME_SIZE, 4); -#ifdef G729_BITEXACT /* - This correction required to get bit-exact result with - reference code, because gain_corr_factor in G.729D is - two times larger than in original G.729. + This filter enhances harmonic components of the fixed-codebook vector to + improve the quality of the reconstructed speech. - If bit-exact result is not issue then gain_corr_factor - can be simpler divided by 2 before call to g729_get_gain_code - instead of using correction below. + / fc_v[i], i < pitch_delay + fc_v[i] = < + \ fc_v[i] + gain_pitch * fc_v[i-pitch_delay], i >= pitch_delay */ + ff_acelp_weighted_vector_sum(fc + pitch_delay_int[i], + fc + pitch_delay_int[i], + fc, 1 << 14, + av_clip(ctx->past_gain_pitch[0], SHARP_MIN, SHARP_MAX), + 0, 14, + SUBFRAME_SIZE - pitch_delay_int[i]); + + memmove(ctx->past_gain_pitch+1, ctx->past_gain_pitch, 5 * sizeof(int16_t)); + ctx->past_gain_code[1] = ctx->past_gain_code[0]; + + if (frame_erasure) { + ctx->past_gain_pitch[0] = (29491 * ctx->past_gain_pitch[0]) >> 15; // 0.90 (0.15) + ctx->past_gain_code[0] = ( 2007 * ctx->past_gain_code[0] ) >> 11; // 0.98 (0.11) + + gain_corr_factor = 0; + } else { + if (packet_type == FORMAT_G729D_6K4) { + ctx->past_gain_pitch[0] = cb_gain_1st_6k4[gc_1st_index][0] + + cb_gain_2nd_6k4[gc_2nd_index][0]; + gain_corr_factor = cb_gain_1st_6k4[gc_1st_index][1] + + cb_gain_2nd_6k4[gc_2nd_index][1]; + + /* Without check below overflow can occur in ff_acelp_update_past_gain. + It is not issue for G.729, because gain_corr_factor in it's case is always + greater than 1024, while in G.729D it can be even zero. */ + gain_corr_factor = FFMAX(gain_corr_factor, 1024); + #ifndef G729_BITEXACT + gain_corr_factor >>= 1; + #endif + } else { + ctx->past_gain_pitch[0] = cb_gain_1st_8k[gc_1st_index][0] + + cb_gain_2nd_8k[gc_2nd_index][0]; + gain_corr_factor = cb_gain_1st_8k[gc_1st_index][1] + + cb_gain_2nd_8k[gc_2nd_index][1]; + } + + /* Decode the fixed-codebook gain. */ + ctx->past_gain_code[0] = ff_acelp_decode_gain_code(&s->adsp, gain_corr_factor, + fc, MR_ENERGY, + ctx->quant_energy, + ma_prediction_coeff, + SUBFRAME_SIZE, 4); + #ifdef G729_BITEXACT + /* + This correction required to get bit-exact result with + reference code, because gain_corr_factor in G.729D is + two times larger than in original G.729. + + If bit-exact result is not issue then gain_corr_factor + can be simpler divided by 2 before call to g729_get_gain_code + instead of using correction below. + */ + if (packet_type == FORMAT_G729D_6K4) { + gain_corr_factor >>= 1; + ctx->past_gain_code[0] >>= 1; + } + #endif + } + ff_acelp_update_past_gain(ctx->quant_energy, gain_corr_factor, 2, frame_erasure); + + /* Routine requires rounding to lowest. */ + ff_acelp_interpolate(ctx->exc + i * SUBFRAME_SIZE, + ctx->exc + i * SUBFRAME_SIZE - pitch_delay_3x / 3, + ff_acelp_interp_filter, 6, + (pitch_delay_3x % 3) << 1, + 10, SUBFRAME_SIZE); + + ff_acelp_weighted_vector_sum(ctx->exc + i * SUBFRAME_SIZE, + ctx->exc + i * SUBFRAME_SIZE, fc, + (!ctx->was_periodic && frame_erasure) ? 0 : ctx->past_gain_pitch[0], + ( ctx->was_periodic && frame_erasure) ? 0 : ctx->past_gain_code[0], + 1 << 13, 14, SUBFRAME_SIZE); + + memcpy(synth, ctx->syn_filter_data, 10 * sizeof(int16_t)); + + if (ff_celp_lp_synthesis_filter( + synth+10, + &lp[i][1], + ctx->exc + i * SUBFRAME_SIZE, + SUBFRAME_SIZE, + 10, + 1, + 0, + 0x800)) + /* Overflow occurred, downscale excitation signal... */ + for (j = 0; j < 2 * SUBFRAME_SIZE + PITCH_DELAY_MAX + INTERPOL_LEN; j++) + ctx->exc_base[j] >>= 2; + + /* ... and make synthesis again. */ if (packet_type == FORMAT_G729D_6K4) { - gain_corr_factor >>= 1; - ctx->past_gain_code[0] >>= 1; + int16_t exc_new[SUBFRAME_SIZE]; + + ctx->onset = g729d_onset_decision(ctx->onset, ctx->past_gain_code); + ctx->voice_decision = g729d_voice_decision(ctx->onset, ctx->voice_decision, ctx->past_gain_pitch); + + g729d_get_new_exc(exc_new, ctx->exc + i * SUBFRAME_SIZE, fc, ctx->voice_decision, ctx->past_gain_code[0], SUBFRAME_SIZE); + + ff_celp_lp_synthesis_filter( + synth+10, + &lp[i][1], + exc_new, + SUBFRAME_SIZE, + 10, + 0, + 0, + 0x800); + } else { + ff_celp_lp_synthesis_filter( + synth+10, + &lp[i][1], + ctx->exc + i * SUBFRAME_SIZE, + SUBFRAME_SIZE, + 10, + 0, + 0, + 0x800); } -#endif - } - ff_acelp_update_past_gain(ctx->quant_energy, gain_corr_factor, 2, frame_erasure); - - /* Routine requires rounding to lowest. */ - ff_acelp_interpolate(ctx->exc + i * SUBFRAME_SIZE, - ctx->exc + i * SUBFRAME_SIZE - pitch_delay_3x / 3, - ff_acelp_interp_filter, 6, - (pitch_delay_3x % 3) << 1, - 10, SUBFRAME_SIZE); - - ff_acelp_weighted_vector_sum(ctx->exc + i * SUBFRAME_SIZE, - ctx->exc + i * SUBFRAME_SIZE, fc, - (!ctx->was_periodic && frame_erasure) ? 0 : ctx->past_gain_pitch[0], - ( ctx->was_periodic && frame_erasure) ? 0 : ctx->past_gain_code[0], - 1 << 13, 14, SUBFRAME_SIZE); - - memcpy(synth, ctx->syn_filter_data, 10 * sizeof(int16_t)); - - if (ff_celp_lp_synthesis_filter( - synth+10, - &lp[i][1], - ctx->exc + i * SUBFRAME_SIZE, - SUBFRAME_SIZE, - 10, - 1, - 0, - 0x800)) - /* Overflow occurred, downscale excitation signal... */ - for (j = 0; j < 2 * SUBFRAME_SIZE + PITCH_DELAY_MAX + INTERPOL_LEN; j++) - ctx->exc_base[j] >>= 2; - - /* ... and make synthesis again. */ - if (packet_type == FORMAT_G729D_6K4) { - int16_t exc_new[SUBFRAME_SIZE]; - - ctx->onset = g729d_onset_decision(ctx->onset, ctx->past_gain_code); - ctx->voice_decision = g729d_voice_decision(ctx->onset, ctx->voice_decision, ctx->past_gain_pitch); - - g729d_get_new_exc(exc_new, ctx->exc + i * SUBFRAME_SIZE, fc, ctx->voice_decision, ctx->past_gain_code[0], SUBFRAME_SIZE); - - ff_celp_lp_synthesis_filter( + /* Save data (without postfilter) for use in next subframe. */ + memcpy(ctx->syn_filter_data, synth+SUBFRAME_SIZE, 10 * sizeof(int16_t)); + + /* Calculate gain of unfiltered signal for use in AGC. */ + gain_before = 0; + for (j = 0; j < SUBFRAME_SIZE; j++) + gain_before += FFABS(synth[j+10]); + + /* Call postfilter and also update voicing decision for use in next frame. */ + ff_g729_postfilter( + &s->adsp, + &ctx->ht_prev_data, + &is_periodic, + &lp[i][0], + pitch_delay_int[0], + ctx->residual, + ctx->res_filter_data, + ctx->pos_filter_data, synth+10, - &lp[i][1], - exc_new, - SUBFRAME_SIZE, - 10, - 0, - 0, - 0x800); - } else { - ff_celp_lp_synthesis_filter( + SUBFRAME_SIZE); + + /* Calculate gain of filtered signal for use in AGC. */ + gain_after = 0; + for (j = 0; j < SUBFRAME_SIZE; j++) + gain_after += FFABS(synth[j+10]); + + ctx->gain_coeff = ff_g729_adaptive_gain_control( + gain_before, + gain_after, synth+10, - &lp[i][1], - ctx->exc + i * SUBFRAME_SIZE, SUBFRAME_SIZE, - 10, - 0, - 0, - 0x800); - } - /* Save data (without postfilter) for use in next subframe. */ - memcpy(ctx->syn_filter_data, synth+SUBFRAME_SIZE, 10 * sizeof(int16_t)); - - /* Calculate gain of unfiltered signal for use in AGC. */ - gain_before = 0; - for (j = 0; j < SUBFRAME_SIZE; j++) - gain_before += FFABS(synth[j+10]); - - /* Call postfilter and also update voicing decision for use in next frame. */ - ff_g729_postfilter( - &ctx->adsp, - &ctx->ht_prev_data, - &is_periodic, - &lp[i][0], - pitch_delay_int[0], - ctx->residual, - ctx->res_filter_data, - ctx->pos_filter_data, - synth+10, - SUBFRAME_SIZE); + ctx->gain_coeff); + + if (frame_erasure) { + ctx->pitch_delay_int_prev = FFMIN(ctx->pitch_delay_int_prev + 1, PITCH_DELAY_MAX); + } else { + ctx->pitch_delay_int_prev = pitch_delay_int[i]; + } - /* Calculate gain of filtered signal for use in AGC. */ - gain_after = 0; - for(j=0; jhpf_z, 2*sizeof(int16_t)); + ff_acelp_high_pass_filter( + out_frame + i*SUBFRAME_SIZE, + ctx->hpf_f, + synth+10, + SUBFRAME_SIZE); + memcpy(ctx->hpf_z, synth+8+SUBFRAME_SIZE, 2*sizeof(int16_t)); + } - ctx->gain_coeff = ff_g729_adaptive_gain_control( - gain_before, - gain_after, - synth+10, - SUBFRAME_SIZE, - ctx->gain_coeff); + ctx->was_periodic = is_periodic; - if (frame_erasure) - ctx->pitch_delay_int_prev = FFMIN(ctx->pitch_delay_int_prev + 1, PITCH_DELAY_MAX); - else - ctx->pitch_delay_int_prev = pitch_delay_int[i]; + /* Save signal for use in next frame. */ + memmove(ctx->exc_base, ctx->exc_base + 2 * SUBFRAME_SIZE, (PITCH_DELAY_MAX+INTERPOL_LEN)*sizeof(int16_t)); - memcpy(synth+8, ctx->hpf_z, 2*sizeof(int16_t)); - ff_acelp_high_pass_filter( - out_frame + i*SUBFRAME_SIZE, - ctx->hpf_f, - synth+10, - SUBFRAME_SIZE); - memcpy(ctx->hpf_z, synth+8+SUBFRAME_SIZE, 2*sizeof(int16_t)); + buf += packet_type == FORMAT_G729_8K ? G729_8K_BLOCK_SIZE : G729D_6K4_BLOCK_SIZE; + ctx++; } - ctx->was_periodic = is_periodic; + *got_frame_ptr = 1; + return packet_type == FORMAT_G729_8K ? (G729_8K_BLOCK_SIZE + (avctx->codec_id == AV_CODEC_ID_ACELP_KELVIN)) * avctx->channels : G729D_6K4_BLOCK_SIZE * avctx->channels; +} - /* Save signal for use in next frame. */ - memmove(ctx->exc_base, ctx->exc_base + 2 * SUBFRAME_SIZE, (PITCH_DELAY_MAX+INTERPOL_LEN)*sizeof(int16_t)); +static av_cold int decode_close(AVCodecContext *avctx) +{ + G729Context *s = avctx->priv_data; + av_freep(&s->channel_context); - *got_frame_ptr = 1; - return packet_type == FORMAT_G729_8K ? 10 : 8; + return 0; } AVCodec ff_g729_decoder = { @@ -714,5 +751,18 @@ AVCodec ff_g729_decoder = { .priv_data_size = sizeof(G729Context), .init = decoder_init, .decode = decode_frame, + .close = decode_close, + .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1, +}; + +AVCodec ff_acelp_kelvin_decoder = { + .name = "acelp.kelvin", + .long_name = NULL_IF_CONFIG_SMALL("Sipro ACELP.KELVIN"), + .type = AVMEDIA_TYPE_AUDIO, + .id = AV_CODEC_ID_ACELP_KELVIN, + .priv_data_size = sizeof(G729Context), + .init = decoder_init, + .decode = decode_frame, + .close = decode_close, .capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1, };