0b61af73 |
/* |
8db3b856 |
* G.729, G729 Annex D decoders |
0b61af73 |
* Copyright (c) 2008 Vladimir Voroshilov
*
* 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
*/ |
22dd24fc |
|
0b61af73 |
#include <inttypes.h>
#include <string.h>
#include "avcodec.h"
#include "libavutil/avutil.h"
#include "get_bits.h" |
99497b46 |
#include "audiodsp.h" |
874c5b02 |
#include "internal.h"
|
0b61af73 |
|
a5e0c4dd |
#include "g729.h" |
0b61af73 |
#include "lsp.h"
#include "celp_math.h" |
b7c7fc33 |
#include "celp_filters.h" |
0b61af73 |
#include "acelp_filters.h"
#include "acelp_pitch_delay.h"
#include "acelp_vectors.h"
#include "g729data.h" |
aca516cd |
#include "g729postfilter.h" |
0b61af73 |
/**
* minimum quantized LSF value (3.2.4)
* 0.005 in Q13
*/
#define LSFQ_MIN 40
/**
* maximum quantized LSF value (3.2.4)
* 3.135 in Q13
*/
#define LSFQ_MAX 25681
/**
* minimum LSF distance (3.2.4)
* 0.0391 in Q13
*/
#define LSFQ_DIFF_MIN 321
|
b7c7fc33 |
/// interpolation filter length
#define INTERPOL_LEN 11
|
0b61af73 |
/**
* minimum gain pitch value (3.8, Equation 47)
* 0.2 in (1.14)
*/
#define SHARP_MIN 3277
/**
* maximum gain pitch value (3.8, Equation 47)
* (EE) This does not comply with the specification.
* Specification says about 0.8, which should be
* 13107 in (1.14), but reference C code uses
* 13017 (equals to 0.7945) instead of it.
*/
#define SHARP_MAX 13017
|
0a333788 |
/**
* MR_ENERGY (mean removed energy) = mean_energy + 10 * log10(2^26 * subframe_size) in (7.13)
*/
#define MR_ENERGY 1018156
|
e610c5f3 |
#define DECISION_NOISE 0
#define DECISION_INTERMEDIATE 1
#define DECISION_VOICE 2
|
70efd101 |
typedef enum {
FORMAT_G729_8K = 0,
FORMAT_G729D_6K4,
FORMAT_COUNT,
} G729Formats;
|
0b61af73 |
typedef struct {
uint8_t ac_index_bits[2]; ///< adaptive codebook index for second subframe (size in bits)
uint8_t parity_bit; ///< parity bit for pitch delay
uint8_t gc_1st_index_bits; ///< gain codebook (first stage) index (size in bits)
uint8_t gc_2nd_index_bits; ///< gain codebook (second stage) index (size in bits)
uint8_t fc_signs_bits; ///< number of pulses in fixed-codebook vector
uint8_t fc_indexes_bits; ///< size (in bits) of fixed-codebook index entry
} G729FormatDescription;
typedef struct { |
99497b46 |
AudioDSPContext adsp; |
0a333788 |
|
f830d1b7 |
/// past excitation signal buffer
int16_t exc_base[2*SUBFRAME_SIZE+PITCH_DELAY_MAX+INTERPOL_LEN];
int16_t* exc; ///< start of past excitation data in buffer |
0b61af73 |
int pitch_delay_int_prev; ///< integer part of previous subframe's pitch delay (4.1.3)
/// (2.13) LSP quantizer outputs
int16_t past_quantizer_output_buf[MA_NP + 1][10];
int16_t* past_quantizer_outputs[MA_NP + 1];
int16_t lsfq[10]; ///< (2.13) quantized LSF coefficients from previous frame
int16_t lsp_buf[2][10]; ///< (0.15) LSP coefficients (previous and current frames) (3.2.5)
int16_t *lsp[2]; ///< pointers to lsp_buf |
388f2255 |
|
0a333788 |
int16_t quant_energy[4]; ///< (5.10) past quantized energy
|
b7c7fc33 |
/// previous speech data for LP synthesis filter
int16_t syn_filter_data[10];
|
aca516cd |
/// residual signal buffer (used in long-term postfilter)
int16_t residual[SUBFRAME_SIZE + RES_PREV_DATA_SIZE];
/// previous speech data for residual calculation filter
int16_t res_filter_data[SUBFRAME_SIZE+10];
/// previous speech data for short-term postfilter
int16_t pos_filter_data[SUBFRAME_SIZE+10];
|
f7980a7b |
/// (1.14) pitch gain of current and five previous subframes
int16_t past_gain_pitch[6]; |
0a333788 |
|
f7980a7b |
/// (14.1) gain code from current and previous subframe
int16_t past_gain_code[2]; |
0a333788 |
|
e610c5f3 |
/// voice decision on previous subframe (0-noise, 1-intermediate, 2-voice), G.729D
int16_t voice_decision;
int16_t onset; ///< detected onset level (0-2) |
f830d1b7 |
int16_t was_periodic; ///< whether previous frame was declared as periodic or not (4.4) |
aca516cd |
int16_t ht_prev_data; ///< previous data for 4.2.3, equation 86 |
68233767 |
int gain_coeff; ///< (1.14) gain coefficient (4.2.4) |
7fadc015 |
uint16_t rand_value; ///< random number generator value (4.4.4) |
388f2255 |
int ma_predictor_prev; ///< switched MA predictor of LSP quantizer from last good frame |
c458bff9 |
/// (14.14) high-pass filter data (past input)
int hpf_f[2];
/// high-pass filter data (past output)
int16_t hpf_z[2]; |
0b61af73 |
} G729Context;
static const G729FormatDescription format_g729_8k = {
.ac_index_bits = {8,5},
.parity_bit = 1,
.gc_1st_index_bits = GC_1ST_IDX_BITS_8K,
.gc_2nd_index_bits = GC_2ND_IDX_BITS_8K,
.fc_signs_bits = 4,
.fc_indexes_bits = 13,
};
static const G729FormatDescription format_g729d_6k4 = {
.ac_index_bits = {8,4},
.parity_bit = 0,
.gc_1st_index_bits = GC_1ST_IDX_BITS_6K4,
.gc_2nd_index_bits = GC_2ND_IDX_BITS_6K4,
.fc_signs_bits = 2,
.fc_indexes_bits = 9,
};
/**
* @brief pseudo random number generator
*/
static inline uint16_t g729_prng(uint16_t value)
{
return 31821 * value + 13849;
}
/** |
b29e5a67 |
* Decodes LSF (Line Spectral Frequencies) from L0-L3 (3.2.4). |
bdba96e9 |
* @param[out] lsfq (2.13) quantized LSF coefficients
* @param[in,out] past_quantizer_outputs (2.13) quantizer outputs from previous frames |
b29e5a67 |
* @param ma_predictor switched MA predictor of LSP quantizer
* @param vq_1st first stage vector of quantizer
* @param vq_2nd_low second stage lower vector of LSP quantizer
* @param vq_2nd_high second stage higher vector of LSP quantizer
*/ |
0b61af73 |
static void lsf_decode(int16_t* lsfq, int16_t* past_quantizer_outputs[MA_NP + 1],
int16_t ma_predictor,
int16_t vq_1st, int16_t vq_2nd_low, int16_t vq_2nd_high)
{
int i,j;
static const uint8_t min_distance[2]={10, 5}; //(2.13)
int16_t* quantizer_output = past_quantizer_outputs[MA_NP];
for (i = 0; i < 5; i++) {
quantizer_output[i] = cb_lsp_1st[vq_1st][i ] + cb_lsp_2nd[vq_2nd_low ][i ];
quantizer_output[i + 5] = cb_lsp_1st[vq_1st][i + 5] + cb_lsp_2nd[vq_2nd_high][i + 5];
}
for (j = 0; j < 2; j++) {
for (i = 1; i < 10; i++) {
int diff = (quantizer_output[i - 1] - quantizer_output[i] + min_distance[j]) >> 1;
if (diff > 0) {
quantizer_output[i - 1] -= diff;
quantizer_output[i ] += diff;
}
}
}
for (i = 0; i < 10; i++) {
int sum = quantizer_output[i] * cb_ma_predictor_sum[ma_predictor][i];
for (j = 0; j < MA_NP; j++)
sum += past_quantizer_outputs[j][i] * cb_ma_predictor[ma_predictor][j][i];
lsfq[i] = sum >> 15;
}
ff_acelp_reorder_lsf(lsfq, LSFQ_DIFF_MIN, LSFQ_MIN, LSFQ_MAX, 10);
}
|
d1a643e7 |
/**
* Restores past LSP quantizer output using LSF from previous frame |
bdba96e9 |
* @param[in,out] lsfq (2.13) quantized LSF coefficients
* @param[in,out] past_quantizer_outputs (2.13) quantizer outputs from previous frames |
d1a643e7 |
* @param ma_predictor_prev MA predictor from previous frame
* @param lsfq_prev (2.13) quantized LSF coefficients from previous frame
*/ |
388f2255 |
static void lsf_restore_from_previous(int16_t* lsfq,
int16_t* past_quantizer_outputs[MA_NP + 1],
int ma_predictor_prev)
{
int16_t* quantizer_output = past_quantizer_outputs[MA_NP];
int i,k;
for (i = 0; i < 10; i++) {
int tmp = lsfq[i] << 15;
for (k = 0; k < MA_NP; k++)
tmp -= past_quantizer_outputs[k][i] * cb_ma_predictor[ma_predictor_prev][k][i];
quantizer_output[i] = ((tmp >> 15) * cb_ma_predictor_sum_inv[ma_predictor_prev][i]) >> 12;
}
}
|
e610c5f3 |
/**
* Constructs new excitation signal and applies phase filter to it |
bdba96e9 |
* @param[out] out constructed speech signal |
e610c5f3 |
* @param in original excitation signal
* @param fc_cur (2.13) original fixed-codebook vector
* @param gain_code (14.1) gain code
* @param subframe_size length of the subframe
*/ |
ce7c9548 |
static void g729d_get_new_exc( |
e610c5f3 |
int16_t* out,
const int16_t* in,
const int16_t* fc_cur,
int dstate,
int gain_code,
int subframe_size)
{
int i;
int16_t fc_new[SUBFRAME_SIZE];
ff_celp_convolve_circ(fc_new, fc_cur, phase_filter[dstate], subframe_size);
for(i=0; i<subframe_size; i++)
{
out[i] = in[i];
out[i] -= (gain_code * fc_cur[i] + 0x2000) >> 14;
out[i] += (gain_code * fc_new[i] + 0x2000) >> 14;
}
}
/**
* Makes decision about onset in current subframe
* @param past_onset decision result of previous subframe
* @param past_gain_code gain code of current and previous subframe
*
* @return onset decision result for current subframe
*/ |
ce7c9548 |
static int g729d_onset_decision(int past_onset, const int16_t* past_gain_code) |
e610c5f3 |
{
if((past_gain_code[0] >> 1) > past_gain_code[1])
return 2;
else
return FFMAX(past_onset-1, 0);
}
/**
* Makes decision about voice presence in current subframe
* @param onset onset level
* @param prev_voice_decision voice decision result from previous subframe
* @param past_gain_pitch pitch gain of current and previous subframes
*
* @return voice decision result for current subframe
*/
static int16_t g729d_voice_decision(int onset, int prev_voice_decision, const int16_t* past_gain_pitch)
{
int i, low_gain_pitch_cnt, voice_decision;
if(past_gain_pitch[0] >= 14745) // 0.9
voice_decision = DECISION_VOICE;
else if (past_gain_pitch[0] <= 9830) // 0.6
voice_decision = DECISION_NOISE;
else
voice_decision = DECISION_INTERMEDIATE;
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)
voice_decision = DECISION_NOISE;
if(!onset && voice_decision > prev_voice_decision + 1)
voice_decision--;
if(onset && voice_decision < DECISION_VOICE)
voice_decision++;
return voice_decision;
}
|
9ff43569 |
static int32_t scalarproduct_int16_c(const int16_t * v1, const int16_t * v2, int order) |
c3299726 |
{
int res = 0;
while (order--) |
9ff43569 |
res += *v1++ * *v2++; |
c3299726 |
return res;
}
|
0b61af73 |
static av_cold int decoder_init(AVCodecContext * avctx)
{
G729Context* ctx = avctx->priv_data;
int i,k;
if (avctx->channels != 1) {
av_log(avctx, AV_LOG_ERROR, "Only mono sound is supported (requested channels: %d).\n", avctx->channels);
return AVERROR(EINVAL);
} |
91c5f81b |
avctx->sample_fmt = AV_SAMPLE_FMT_S16; |
0b61af73 |
/* Both 8kbit/s and 6.4kbit/s modes uses two subframes per frame. */
avctx->frame_size = SUBFRAME_SIZE << 1;
|
68233767 |
ctx->gain_coeff = 16384; // 1.0 in (1.14)
|
0b61af73 |
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->lsp[0] = ctx->lsp_buf[0];
ctx->lsp[1] = ctx->lsp_buf[1];
memcpy(ctx->lsp[0], lsp_init, 10 * sizeof(int16_t));
|
f830d1b7 |
ctx->exc = &ctx->exc_base[PITCH_DELAY_MAX+INTERPOL_LEN];
|
c963189b |
ctx->pitch_delay_int_prev = PITCH_DELAY_MIN;
|
7fadc015 |
/* random seed initialization */
ctx->rand_value = 21845;
|
0a333788 |
/* quantized prediction error */
for(i=0; i<4; i++)
ctx->quant_energy[i] = -14336; // -14 in (5.10)
|
99497b46 |
ff_audiodsp_init(&ctx->adsp);
ctx->adsp.scalarproduct_int16 = scalarproduct_int16_c; |
0a333788 |
|
0b61af73 |
return 0;
}
|
7db5ff79 |
static int decode_frame(AVCodecContext *avctx, void *data, int *got_frame_ptr, |
0b61af73 |
AVPacket *avpkt)
{
const uint8_t *buf = avpkt->data;
int buf_size = avpkt->size; |
7db5ff79 |
int16_t *out_frame; |
0b61af73 |
GetBitContext gb; |
02aabd82 |
const G729FormatDescription *format; |
0b61af73 |
int frame_erasure = 0; ///< frame erasure detected during decoding
int bad_pitch = 0; ///< parity check failed
int i; |
388f2255 |
int16_t *tmp; |
70efd101 |
G729Formats packet_type; |
0b61af73 |
G729Context *ctx = avctx->priv_data;
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
uint8_t quantizer_2nd_lo; ///< second stage lower vector of quantizer (size in bits)
uint8_t quantizer_2nd_hi; ///< second stage higher vector of quantizer (size in bits)
|
16bbb8df |
int pitch_delay_int[2]; // pitch delay, integer part |
0b61af73 |
int pitch_delay_3x; // pitch delay, multiplied by 3 |
9297c782 |
int16_t fc[SUBFRAME_SIZE]; // fixed-codebook vector |
b7c7fc33 |
int16_t synth[SUBFRAME_SIZE+10]; // fixed-codebook vector |
7db5ff79 |
int j, ret; |
68233767 |
int gain_before, gain_after; |
f830d1b7 |
int is_periodic = 0; // whether one of the subframes is declared as periodic or not |
8251c053 |
AVFrame *frame = data; |
0b61af73 |
|
8251c053 |
frame->nb_samples = SUBFRAME_SIZE<<1;
if ((ret = ff_get_buffer(avctx, frame, 0)) < 0) |
7db5ff79 |
return ret; |
8251c053 |
out_frame = (int16_t*) frame->data[0]; |
0b61af73 |
|
daf73d5d |
if (buf_size % 10 == 0) { |
70efd101 |
packet_type = FORMAT_G729_8K; |
02aabd82 |
format = &format_g729_8k; |
e610c5f3 |
//Reset voice decision
ctx->onset = 0;
ctx->voice_decision = DECISION_VOICE; |
0b61af73 |
av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729 @ 8kbit/s");
} else if (buf_size == 8) { |
70efd101 |
packet_type = FORMAT_G729D_6K4; |
02aabd82 |
format = &format_g729d_6k4; |
0b61af73 |
av_log(avctx, AV_LOG_DEBUG, "Packet type: %s\n", "G.729D @ 6.4kbit/s");
} else {
av_log(avctx, AV_LOG_ERROR, "Packet size %d is unknown.\n", buf_size);
return AVERROR_INVALIDDATA;
}
for (i=0; i < buf_size; i++)
frame_erasure |= buf[i];
frame_erasure = !frame_erasure;
|
edf1a8e3 |
init_get_bits(&gb, buf, 8*buf_size); |
0b61af73 |
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);
|
388f2255 |
if(frame_erasure)
lsf_restore_from_previous(ctx->lsfq, ctx->past_quantizer_outputs,
ctx->ma_predictor_prev);
else { |
cd3e2820 |
lsf_decode(ctx->lsfq, ctx->past_quantizer_outputs,
ma_predictor,
quantizer_1st, quantizer_2nd_lo, quantizer_2nd_hi); |
388f2255 |
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; |
0b61af73 |
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++) { |
0a333788 |
int gain_corr_factor;
|
0b61af73 |
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
|
02aabd82 |
ac_index = get_bits(&gb, format->ac_index_bits[i]);
if(!i && format->parity_bit) |
28d5a3a7 |
bad_pitch = av_parity(ac_index >> 2) == get_bits1(&gb); |
02aabd82 |
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); |
0b61af73 |
|
50cad256 |
if (frame_erasure)
pitch_delay_3x = 3 * ctx->pitch_delay_int_prev;
else if(!i) { |
0b61af73 |
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);
}
/* Round pitch delay to nearest (used everywhere except ff_acelp_interpolate). */ |
16bbb8df |
pitch_delay_int[i] = (pitch_delay_3x + 1) / 3; |
a974adc3 |
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;
} |
0b61af73 |
|
7fadc015 |
if (frame_erasure) {
ctx->rand_value = g729_prng(ctx->rand_value); |
ba625dd8 |
fc_indexes = av_mod_uintp2(ctx->rand_value, format->fc_indexes_bits); |
7fadc015 |
ctx->rand_value = g729_prng(ctx->rand_value);
pulses_signs = ctx->rand_value;
}
|
9297c782 |
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.
/ fc_v[i], i < pitch_delay
fc_v[i] = <
\ fc_v[i] + gain_pitch * fc_v[i-pitch_delay], i >= pitch_delay
*/ |
16bbb8df |
ff_acelp_weighted_vector_sum(fc + pitch_delay_int[i],
fc + pitch_delay_int[i], |
0b61af73 |
fc, 1 << 14, |
f7980a7b |
av_clip(ctx->past_gain_pitch[0], SHARP_MIN, SHARP_MAX), |
0b61af73 |
0, 14, |
16bbb8df |
SUBFRAME_SIZE - pitch_delay_int[i]); |
0b61af73 |
|
f7980a7b |
memmove(ctx->past_gain_pitch+1, ctx->past_gain_pitch, 5 * sizeof(int16_t));
ctx->past_gain_code[1] = ctx->past_gain_code[0];
|
0b61af73 |
if (frame_erasure) { |
f7980a7b |
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) |
0b61af73 |
gain_corr_factor = 0;
} else { |
0b42463a |
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];
|
6851130f |
/* Without check below overflow can occur in ff_acelp_update_past_gain. |
0b42463a |
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 { |
4920a1a9 |
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]; |
0b42463a |
} |
0b61af73 |
|
0a333788 |
/* Decode the fixed-codebook gain. */ |
99497b46 |
ctx->past_gain_code[0] = ff_acelp_decode_gain_code(&ctx->adsp, gain_corr_factor, |
12081d05 |
fc, MR_ENERGY,
ctx->quant_energy,
ma_prediction_coeff,
SUBFRAME_SIZE, 4); |
0b42463a |
#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 |
6851130f |
can be simpler divided by 2 before call to g729_get_gain_code |
0b42463a |
instead of using correction below.
*/
if (packet_type == FORMAT_G729D_6K4) {
gain_corr_factor >>= 1;
ctx->past_gain_code[0] >>= 1;
}
#endif |
0a333788 |
}
ff_acelp_update_past_gain(ctx->quant_energy, gain_corr_factor, 2, frame_erasure);
|
f830d1b7 |
/* 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);
|
0b61af73 |
ff_acelp_weighted_vector_sum(ctx->exc + i * SUBFRAME_SIZE,
ctx->exc + i * SUBFRAME_SIZE, fc, |
f7980a7b |
(!ctx->was_periodic && frame_erasure) ? 0 : ctx->past_gain_pitch[0],
( ctx->was_periodic && frame_erasure) ? 0 : ctx->past_gain_code[0], |
0b61af73 |
1 << 13, 14, SUBFRAME_SIZE);
|
b7c7fc33 |
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, |
bcc67dff |
0, |
e610c5f3 |
0x800)) |
2d38081b |
/* Overflow occurred, downscale excitation signal... */ |
b7c7fc33 |
for (j = 0; j < 2 * SUBFRAME_SIZE + PITCH_DELAY_MAX + INTERPOL_LEN; j++)
ctx->exc_base[j] >>= 2;
|
e610c5f3 |
/* ... 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(
synth+10,
&lp[i][1],
exc_new,
SUBFRAME_SIZE,
10,
0, |
bcc67dff |
0, |
e610c5f3 |
0x800);
} else { |
b7c7fc33 |
ff_celp_lp_synthesis_filter(
synth+10,
&lp[i][1],
ctx->exc + i * SUBFRAME_SIZE,
SUBFRAME_SIZE,
10,
0, |
bcc67dff |
0, |
b7c7fc33 |
0x800);
}
/* Save data (without postfilter) for use in next subframe. */
memcpy(ctx->syn_filter_data, synth+SUBFRAME_SIZE, 10 * sizeof(int16_t));
|
68233767 |
/* 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]);
|
aca516cd |
/* Call postfilter and also update voicing decision for use in next frame. */ |
1c4712db |
ff_g729_postfilter( |
99497b46 |
&ctx->adsp, |
aca516cd |
&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);
|
68233767 |
/* 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]);
|
1c4712db |
ctx->gain_coeff = ff_g729_adaptive_gain_control( |
68233767 |
gain_before,
gain_after,
synth+10,
SUBFRAME_SIZE,
ctx->gain_coeff);
|
50cad256 |
if (frame_erasure)
ctx->pitch_delay_int_prev = FFMIN(ctx->pitch_delay_int_prev + 1, PITCH_DELAY_MAX);
else |
16bbb8df |
ctx->pitch_delay_int_prev = pitch_delay_int[i]; |
b7c7fc33 |
|
c458bff9 |
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)); |
0b61af73 |
}
|
f830d1b7 |
ctx->was_periodic = is_periodic;
/* 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));
|
7db5ff79 |
*got_frame_ptr = 1; |
daf73d5d |
return packet_type == FORMAT_G729_8K ? 10 : 8; |
0b61af73 |
}
|
8ce0c7d2 |
AVCodec ff_g729_decoder = { |
ba10207b |
.name = "g729", |
b46f1910 |
.long_name = NULL_IF_CONFIG_SMALL("G.729"), |
ba10207b |
.type = AVMEDIA_TYPE_AUDIO, |
7a72695c |
.id = AV_CODEC_ID_G729, |
ba10207b |
.priv_data_size = sizeof(G729Context),
.init = decoder_init,
.decode = decode_frame, |
444e9874 |
.capabilities = AV_CODEC_CAP_SUBFRAMES | AV_CODEC_CAP_DR1, |
0b61af73 |
}; |