/*
- * G.729 decoder
+ * G.729, G729 Annex D decoders
* Copyright (c) 2008 Vladimir Voroshilov
*
* This file is part of FFmpeg.
* License along with FFmpeg; if not, write to the Free Software
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
*/
-#include <stdlib.h>
+
#include <inttypes.h>
-#include <limits.h>
-#include <stdio.h>
#include <string.h>
-#include <math.h>
-#include <assert.h>
#include "avcodec.h"
#include "libavutil/avutil.h"
#include "acelp_pitch_delay.h"
#include "acelp_vectors.h"
#include "g729data.h"
+#include "g729postfilter.h"
/**
* minimum quantized LSF value (3.2.4)
/// previous speech data for LP synthesis filter
int16_t syn_filter_data[10];
+
+ /// 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];
+
/// (1.14) pitch gain of current and five previous subframes
int16_t past_gain_pitch[6];
int16_t onset; ///< detected onset level (0-2)
int16_t was_periodic; ///< whether previous frame was declared as periodic or not (4.4)
+ int16_t ht_prev_data; ///< previous data for 4.2.3, equation 86
+ int gain_coeff; ///< (1.14) gain coefficient (4.2.4)
uint16_t rand_value; ///< random number generator value (4.4.4)
int ma_predictor_prev; ///< switched MA predictor of LSP quantizer from last good frame
* @param gain_code (14.1) gain code
* @param subframe_size length of the subframe
*/
-void g729d_get_new_exc(
+static void g729d_get_new_exc(
int16_t* out,
const int16_t* in,
const int16_t* fc_cur,
*
* @return onset decision result for current subframe
*/
-int g729d_onset_decision(int past_onset, const int16_t* past_gain_code)
+static int g729d_onset_decision(int past_onset, const int16_t* past_gain_code)
{
if((past_gain_code[0] >> 1) > past_gain_code[1])
return 2;
return voice_decision;
}
+static int32_t scalarproduct_int16_c(const int16_t * v1, const int16_t * v2, int order, int shift)
+{
+ int res = 0;
+
+ while (order--)
+ res += (*v1++ * *v2++) >> shift;
+
+ return res;
+}
+
static av_cold int decoder_init(AVCodecContext * avctx)
{
G729Context* ctx = avctx->priv_data;
av_log(avctx, AV_LOG_ERROR, "Only mono sound is supported (requested channels: %d).\n", avctx->channels);
return AVERROR(EINVAL);
}
+ avctx->sample_fmt = AV_SAMPLE_FMT_S16;
/* 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)
+
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->quant_energy[i] = -14336; // -14 in (5.10)
dsputil_init(&ctx->dsp, avctx);
+ ctx->dsp.scalarproduct_int16 = scalarproduct_int16_c;
return 0;
}
int buf_size = avpkt->size;
int16_t *out_frame = data;
GetBitContext gb;
- G729FormatDescription format;
+ const G729FormatDescription *format;
int frame_erasure = 0; ///< frame erasure detected during decoding
int bad_pitch = 0; ///< parity check failed
int i;
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)
- int pitch_delay_int; // pitch delay, integer part
+ int pitch_delay_int[2]; // pitch delay, integer part
int pitch_delay_3x; // pitch delay, multiplied by 3
int16_t fc[SUBFRAME_SIZE]; // fixed-codebook vector
int16_t synth[SUBFRAME_SIZE+10]; // fixed-codebook vector
int j;
+ int gain_before, gain_after;
int is_periodic = 0; // whether one of the subframes is declared as periodic or not
if (*data_size < SUBFRAME_SIZE << 2) {
if (buf_size == 10) {
packet_type = FORMAT_G729_8K;
- format = 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) {
packet_type = FORMAT_G729D_6K4;
- format = format_g729d_6k4;
+ format = &format_g729d_6k4;
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);
frame_erasure |= buf[i];
frame_erasure = !frame_erasure;
- init_get_bits(&gb, buf, buf_size);
+ init_get_bits(&gb, buf, 8*buf_size);
ma_predictor = get_bits(&gb, 1);
quantizer_1st = get_bits(&gb, VQ_1ST_BITS);
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)
+ ac_index = get_bits(&gb, format->ac_index_bits[i]);
+ if(!i && format->parity_bit)
bad_pitch = get_parity(ac_index) == 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);
+ 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;
}
/* Round pitch delay to nearest (used everywhere except ff_acelp_interpolate). */
- pitch_delay_int = (pitch_delay_3x + 1) / 3;
+ pitch_delay_int[i] = (pitch_delay_3x + 1) / 3;
if (frame_erasure) {
ctx->rand_value = g729_prng(ctx->rand_value);
- fc_indexes = ctx->rand_value & ((1 << format.fc_indexes_bits) - 1);
+ fc_indexes = ctx->rand_value & ((1 << format->fc_indexes_bits) - 1);
ctx->rand_value = g729_prng(ctx->rand_value);
pulses_signs = ctx->rand_value;
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,
- fc + pitch_delay_int,
+ 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);
+ 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];
SUBFRAME_SIZE,
10,
1,
+ 0,
0x800))
/* Overflow occured, downscale excitation signal... */
for (j = 0; j < 2 * SUBFRAME_SIZE + PITCH_DELAY_MAX + INTERPOL_LEN; j++)
SUBFRAME_SIZE,
10,
0,
+ 0,
0x800);
} else {
ff_celp_lp_synthesis_filter(
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->dsp,
+ &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);
+
+ /* 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,
+ 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;
+ ctx->pitch_delay_int_prev = pitch_delay_int[i];
memcpy(synth+8, ctx->hpf_z, 2*sizeof(int16_t));
ff_acelp_high_pass_filter(
projects / ffmpeg / blobdiff