ffmpeg/libavcodec/aacenc.c

969 lines
36 KiB
C
Raw Normal View History

/*
* AAC encoder
* Copyright (C) 2008 Konstantin Shishkov
*
* 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
*/
/**
* @file
* AAC encoder
*/
/***********************************
* TODOs:
* add sane pulse detection
***********************************/
#include "libavutil/float_dsp.h"
#include "libavutil/opt.h"
#include "avcodec.h"
#include "put_bits.h"
2012-01-28 18:28:01 +01:00
#include "internal.h"
#include "mpeg4audio.h"
#include "kbdwin.h"
#include "sinewin.h"
#include "aac.h"
#include "aactab.h"
#include "aacenc.h"
#include "aacenctab.h"
#include "aacenc_utils.h"
#include "psymodel.h"
/**
* Make AAC audio config object.
* @see 1.6.2.1 "Syntax - AudioSpecificConfig"
*/
static void put_audio_specific_config(AVCodecContext *avctx)
{
PutBitContext pb;
AACEncContext *s = avctx->priv_data;
init_put_bits(&pb, avctx->extradata, avctx->extradata_size);
put_bits(&pb, 5, s->profile+1); //profile
put_bits(&pb, 4, s->samplerate_index); //sample rate index
put_bits(&pb, 4, s->channels);
//GASpecificConfig
put_bits(&pb, 1, 0); //frame length - 1024 samples
put_bits(&pb, 1, 0); //does not depend on core coder
put_bits(&pb, 1, 0); //is not extension
//Explicitly Mark SBR absent
put_bits(&pb, 11, 0x2b7); //sync extension
put_bits(&pb, 5, AOT_SBR);
put_bits(&pb, 1, 0);
flush_put_bits(&pb);
}
#define WINDOW_FUNC(type) \
static void apply_ ##type ##_window(AVFloatDSPContext *fdsp, \
SingleChannelElement *sce, \
const float *audio)
WINDOW_FUNC(only_long)
{
const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
float *out = sce->ret_buf;
fdsp->vector_fmul (out, audio, lwindow, 1024);
fdsp->vector_fmul_reverse(out + 1024, audio + 1024, pwindow, 1024);
}
WINDOW_FUNC(long_start)
{
const float *lwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_long_1024 : ff_sine_1024;
const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
float *out = sce->ret_buf;
fdsp->vector_fmul(out, audio, lwindow, 1024);
memcpy(out + 1024, audio + 1024, sizeof(out[0]) * 448);
fdsp->vector_fmul_reverse(out + 1024 + 448, audio + 1024 + 448, swindow, 128);
memset(out + 1024 + 576, 0, sizeof(out[0]) * 448);
}
WINDOW_FUNC(long_stop)
{
const float *lwindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_long_1024 : ff_sine_1024;
const float *swindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
float *out = sce->ret_buf;
memset(out, 0, sizeof(out[0]) * 448);
fdsp->vector_fmul(out + 448, audio + 448, swindow, 128);
memcpy(out + 576, audio + 576, sizeof(out[0]) * 448);
fdsp->vector_fmul_reverse(out + 1024, audio + 1024, lwindow, 1024);
}
WINDOW_FUNC(eight_short)
{
const float *swindow = sce->ics.use_kb_window[0] ? ff_aac_kbd_short_128 : ff_sine_128;
const float *pwindow = sce->ics.use_kb_window[1] ? ff_aac_kbd_short_128 : ff_sine_128;
const float *in = audio + 448;
float *out = sce->ret_buf;
int w;
for (w = 0; w < 8; w++) {
fdsp->vector_fmul (out, in, w ? pwindow : swindow, 128);
out += 128;
in += 128;
fdsp->vector_fmul_reverse(out, in, swindow, 128);
out += 128;
}
}
static void (*const apply_window[4])(AVFloatDSPContext *fdsp,
SingleChannelElement *sce,
const float *audio) = {
[ONLY_LONG_SEQUENCE] = apply_only_long_window,
[LONG_START_SEQUENCE] = apply_long_start_window,
[EIGHT_SHORT_SEQUENCE] = apply_eight_short_window,
[LONG_STOP_SEQUENCE] = apply_long_stop_window
};
static void apply_window_and_mdct(AACEncContext *s, SingleChannelElement *sce,
float *audio)
{
int i;
float *output = sce->ret_buf;
apply_window[sce->ics.window_sequence[0]](s->fdsp, sce, audio);
if (sce->ics.window_sequence[0] != EIGHT_SHORT_SEQUENCE)
s->mdct1024.mdct_calc(&s->mdct1024, sce->coeffs, output);
else
for (i = 0; i < 1024; i += 128)
s->mdct128.mdct_calc(&s->mdct128, &sce->coeffs[i], output + i*2);
memcpy(audio, audio + 1024, sizeof(audio[0]) * 1024);
AAC: Fix M/S stereo encoding This patch fixes a pointer arithmetic bug in adjust_frame_information that resulted in heavily corrupted audio when using M/S encoding. Also, a backup copy of untransformed coefficients has to be kept around or attempts at re-processing the frame (which happens when hevavily overspending bits during transients) will result in re-encoding of the coefficients and subsequent corruption of the resulting stream. A/B testing shows the bug as corrected, but still cannot prove that M/S coding is a win at least in numbers. Limited listening tests do show improvement on M/S encoded samples in lower bitrates, but they're hidden among the other artifacts that remain to be corrected in the encoder. Some of the regressions flagged in the report do show poor stereo image (but not buggy), so M/S encoding is clearly not good enough yet to be defaulted to auto. In numbers, Patched against Unpatched, stereo_mode auto: Files: 114 Bitrates: 6 Tests: 683 Serious Regressions: 0 (0%) Regressions: 0 (0%) Improvements: 227 (33%) Big improvements: 92 (13%) Worst regression - mybloodrusts.wv - 256k - StdDev: 28.61 pSNR: -0.43 maxdiff: 1372.00 Best improvement - 60.wv - 384k - StdDev: -369.57 pSNR: 45.02 maxdiff: -13322.00 Average - StdDev: -80.56 pSNR: 2.49 maxdiff: -8858.00 Patched against Unpatched stereo_mode ms_off shows no difference. Patched stereo_mode auto vs Unpatched stereo_mode ms_off shows a small average improvement, just not too significant: Serious Regressions: 0 (0%) Regressions: 10 (1%) Improvements: 45 (6%) Big improvements: 2 (0%) Worst regression - Illinois.wv - 256k - StdDev: 33.20 pSNR: -2.03 maxdiff: 477.00 Best improvement - song_of_circomstances.flac - 384k - StdDev: -3.97 pSNR: 7.61 maxdiff: -826.00 Average - StdDev: -10.25 pSNR: 0.20 maxdiff: -281.00 Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-03-03 07:43:06 +01:00
memcpy(sce->pcoeffs, sce->coeffs, sizeof(sce->pcoeffs));
}
/**
* Encode ics_info element.
* @see Table 4.6 (syntax of ics_info)
*/
static void put_ics_info(AACEncContext *s, IndividualChannelStream *info)
{
int w;
put_bits(&s->pb, 1, 0); // ics_reserved bit
put_bits(&s->pb, 2, info->window_sequence[0]);
put_bits(&s->pb, 1, info->use_kb_window[0]);
if (info->window_sequence[0] != EIGHT_SHORT_SEQUENCE) {
put_bits(&s->pb, 6, info->max_sfb);
put_bits(&s->pb, 1, !!info->predictor_present);
} else {
put_bits(&s->pb, 4, info->max_sfb);
for (w = 1; w < 8; w++)
put_bits(&s->pb, 1, !info->group_len[w]);
}
}
/**
* Encode MS data.
* @see 4.6.8.1 "Joint Coding - M/S Stereo"
*/
static void encode_ms_info(PutBitContext *pb, ChannelElement *cpe)
{
int i, w;
put_bits(pb, 2, cpe->ms_mode);
if (cpe->ms_mode == 1)
for (w = 0; w < cpe->ch[0].ics.num_windows; w += cpe->ch[0].ics.group_len[w])
for (i = 0; i < cpe->ch[0].ics.max_sfb; i++)
put_bits(pb, 1, cpe->ms_mask[w*16 + i]);
}
/**
* Produce integer coefficients from scalefactors provided by the model.
*/
static void adjust_frame_information(ChannelElement *cpe, int chans)
{
int i, w, w2, g, ch;
int maxsfb, cmaxsfb;
for (ch = 0; ch < chans; ch++) {
IndividualChannelStream *ics = &cpe->ch[ch].ics;
maxsfb = 0;
cpe->ch[ch].pulse.num_pulse = 0;
for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
for (w2 = 0; w2 < ics->group_len[w]; w2++) {
AAC: Fix M/S stereo encoding This patch fixes a pointer arithmetic bug in adjust_frame_information that resulted in heavily corrupted audio when using M/S encoding. Also, a backup copy of untransformed coefficients has to be kept around or attempts at re-processing the frame (which happens when hevavily overspending bits during transients) will result in re-encoding of the coefficients and subsequent corruption of the resulting stream. A/B testing shows the bug as corrected, but still cannot prove that M/S coding is a win at least in numbers. Limited listening tests do show improvement on M/S encoded samples in lower bitrates, but they're hidden among the other artifacts that remain to be corrected in the encoder. Some of the regressions flagged in the report do show poor stereo image (but not buggy), so M/S encoding is clearly not good enough yet to be defaulted to auto. In numbers, Patched against Unpatched, stereo_mode auto: Files: 114 Bitrates: 6 Tests: 683 Serious Regressions: 0 (0%) Regressions: 0 (0%) Improvements: 227 (33%) Big improvements: 92 (13%) Worst regression - mybloodrusts.wv - 256k - StdDev: 28.61 pSNR: -0.43 maxdiff: 1372.00 Best improvement - 60.wv - 384k - StdDev: -369.57 pSNR: 45.02 maxdiff: -13322.00 Average - StdDev: -80.56 pSNR: 2.49 maxdiff: -8858.00 Patched against Unpatched stereo_mode ms_off shows no difference. Patched stereo_mode auto vs Unpatched stereo_mode ms_off shows a small average improvement, just not too significant: Serious Regressions: 0 (0%) Regressions: 10 (1%) Improvements: 45 (6%) Big improvements: 2 (0%) Worst regression - Illinois.wv - 256k - StdDev: 33.20 pSNR: -2.03 maxdiff: 477.00 Best improvement - song_of_circomstances.flac - 384k - StdDev: -3.97 pSNR: 7.61 maxdiff: -826.00 Average - StdDev: -10.25 pSNR: 0.20 maxdiff: -281.00 Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-03-03 07:43:06 +01:00
for (cmaxsfb = ics->num_swb; cmaxsfb > 0 && cpe->ch[ch].zeroes[w*16+cmaxsfb-1]; cmaxsfb--)
;
maxsfb = FFMAX(maxsfb, cmaxsfb);
}
}
ics->max_sfb = maxsfb;
//adjust zero bands for window groups
for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
for (g = 0; g < ics->max_sfb; g++) {
i = 1;
for (w2 = w; w2 < w + ics->group_len[w]; w2++) {
if (!cpe->ch[ch].zeroes[w2*16 + g]) {
i = 0;
break;
}
}
cpe->ch[ch].zeroes[w*16 + g] = i;
}
}
}
if (chans > 1 && cpe->common_window) {
IndividualChannelStream *ics0 = &cpe->ch[0].ics;
IndividualChannelStream *ics1 = &cpe->ch[1].ics;
int msc = 0;
ics0->max_sfb = FFMAX(ics0->max_sfb, ics1->max_sfb);
ics1->max_sfb = ics0->max_sfb;
for (w = 0; w < ics0->num_windows*16; w += 16)
for (i = 0; i < ics0->max_sfb; i++)
if (cpe->ms_mask[w+i])
msc++;
if (msc == 0 || ics0->max_sfb == 0)
cpe->ms_mode = 0;
else
cpe->ms_mode = msc < ics0->max_sfb * ics0->num_windows ? 1 : 2;
}
}
static void apply_intensity_stereo(ChannelElement *cpe)
{
int w, w2, g, i;
IndividualChannelStream *ics = &cpe->ch[0].ics;
if (!cpe->common_window)
return;
for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
for (w2 = 0; w2 < ics->group_len[w]; w2++) {
int start = (w+w2) * 128;
for (g = 0; g < ics->num_swb; g++) {
int p = -1 + 2 * (cpe->ch[1].band_type[w*16+g] - 14);
float scale = cpe->ch[0].is_ener[w*16+g];
if (!cpe->is_mask[w*16 + g]) {
start += ics->swb_sizes[g];
continue;
}
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (cpe->ms_mask[w*16 + g])
p *= -1;
for (i = 0; i < ics->swb_sizes[g]; i++) {
float sum = (cpe->ch[0].coeffs[start+i] + p*cpe->ch[1].coeffs[start+i])*scale;
cpe->ch[0].coeffs[start+i] = sum;
cpe->ch[1].coeffs[start+i] = 0.0f;
}
start += ics->swb_sizes[g];
}
}
}
}
static void apply_mid_side_stereo(ChannelElement *cpe)
{
int w, w2, g, i;
IndividualChannelStream *ics = &cpe->ch[0].ics;
if (!cpe->common_window)
return;
for (w = 0; w < ics->num_windows; w += ics->group_len[w]) {
for (w2 = 0; w2 < ics->group_len[w]; w2++) {
int start = (w+w2) * 128;
for (g = 0; g < ics->num_swb; g++) {
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (!cpe->ms_mask[w*16 + g] && !cpe->is_mask[w*16 + g]) {
start += ics->swb_sizes[g];
continue;
}
for (i = 0; i < ics->swb_sizes[g]; i++) {
float L = (cpe->ch[0].coeffs[start+i] + cpe->ch[1].coeffs[start+i]) * 0.5f;
float R = L - cpe->ch[1].coeffs[start+i];
cpe->ch[0].coeffs[start+i] = L;
cpe->ch[1].coeffs[start+i] = R;
}
start += ics->swb_sizes[g];
}
}
}
}
/**
* Encode scalefactor band coding type.
*/
static void encode_band_info(AACEncContext *s, SingleChannelElement *sce)
{
int w;
if (s->coder->set_special_band_scalefactors)
s->coder->set_special_band_scalefactors(s, sce);
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w])
s->coder->encode_window_bands_info(s, sce, w, sce->ics.group_len[w], s->lambda);
}
/**
* Encode scalefactors.
*/
static void encode_scale_factors(AVCodecContext *avctx, AACEncContext *s,
SingleChannelElement *sce)
{
int diff, off_sf = sce->sf_idx[0], off_pns = sce->sf_idx[0] - NOISE_OFFSET;
int off_is = 0, noise_flag = 1;
int i, w;
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
for (i = 0; i < sce->ics.max_sfb; i++) {
if (!sce->zeroes[w*16 + i]) {
if (sce->band_type[w*16 + i] == NOISE_BT) {
diff = sce->sf_idx[w*16 + i] - off_pns;
off_pns = sce->sf_idx[w*16 + i];
if (noise_flag-- > 0) {
put_bits(&s->pb, NOISE_PRE_BITS, diff + NOISE_PRE);
continue;
}
} else if (sce->band_type[w*16 + i] == INTENSITY_BT ||
sce->band_type[w*16 + i] == INTENSITY_BT2) {
diff = sce->sf_idx[w*16 + i] - off_is;
off_is = sce->sf_idx[w*16 + i];
} else {
diff = sce->sf_idx[w*16 + i] - off_sf;
off_sf = sce->sf_idx[w*16 + i];
}
diff += SCALE_DIFF_ZERO;
av_assert0(diff >= 0 && diff <= 120);
put_bits(&s->pb, ff_aac_scalefactor_bits[diff], ff_aac_scalefactor_code[diff]);
}
}
}
}
/**
* Encode pulse data.
*/
static void encode_pulses(AACEncContext *s, Pulse *pulse)
{
int i;
put_bits(&s->pb, 1, !!pulse->num_pulse);
if (!pulse->num_pulse)
return;
put_bits(&s->pb, 2, pulse->num_pulse - 1);
put_bits(&s->pb, 6, pulse->start);
for (i = 0; i < pulse->num_pulse; i++) {
put_bits(&s->pb, 5, pulse->pos[i]);
put_bits(&s->pb, 4, pulse->amp[i]);
}
}
/**
* Encode spectral coefficients processed by psychoacoustic model.
*/
static void encode_spectral_coeffs(AACEncContext *s, SingleChannelElement *sce)
{
int start, i, w, w2;
for (w = 0; w < sce->ics.num_windows; w += sce->ics.group_len[w]) {
start = 0;
for (i = 0; i < sce->ics.max_sfb; i++) {
if (sce->zeroes[w*16 + i]) {
start += sce->ics.swb_sizes[i];
continue;
}
for (w2 = w; w2 < w + sce->ics.group_len[w]; w2++) {
s->coder->quantize_and_encode_band(s, &s->pb,
&sce->coeffs[start + w2*128],
NULL, sce->ics.swb_sizes[i],
sce->sf_idx[w*16 + i],
sce->band_type[w*16 + i],
s->lambda,
sce->ics.window_clipping[w]);
}
start += sce->ics.swb_sizes[i];
}
}
}
/**
* Downscale spectral coefficients for near-clipping windows to avoid artifacts
*/
static void avoid_clipping(AACEncContext *s, SingleChannelElement *sce)
{
int start, i, j, w;
if (sce->ics.clip_avoidance_factor < 1.0f) {
for (w = 0; w < sce->ics.num_windows; w++) {
start = 0;
for (i = 0; i < sce->ics.max_sfb; i++) {
float *swb_coeffs = &sce->coeffs[start + w*128];
for (j = 0; j < sce->ics.swb_sizes[i]; j++)
swb_coeffs[j] *= sce->ics.clip_avoidance_factor;
start += sce->ics.swb_sizes[i];
}
}
}
}
/**
* Encode one channel of audio data.
*/
static int encode_individual_channel(AVCodecContext *avctx, AACEncContext *s,
SingleChannelElement *sce,
int common_window)
{
put_bits(&s->pb, 8, sce->sf_idx[0]);
if (!common_window) {
put_ics_info(s, &sce->ics);
if (s->coder->encode_main_pred)
s->coder->encode_main_pred(s, sce);
}
encode_band_info(s, sce);
encode_scale_factors(avctx, s, sce);
encode_pulses(s, &sce->pulse);
aacenc_tns: rework the way coefficients are calculated This commit abandons the way the specifications state to quantize the coefficients, makes use of the new LPC float functions and is much better. The original way of converting non-normalized float samples to int32_t which out LPC system expects was wrong and it was wrong to assume the coefficients that are generated are also valid. It was essentially a full garbage-in, garbage-out system and it definitely shows when looking at spectrals and listening. The high frequencies were very overattenuated. The new LPC function performs the analysis directly. The specifications state to quantize the coefficients into four bit index values using an asin() function which of course had to have ugly ternary operators because the function turns negative if the coefficients are negative which when encoding causes invalid bitstream to get generated. This deviates from this by using the direct TNS tables, which are fairly small since you only have 4 bits at most for index values. The LPC values are directly quantized against the tables and are then used to perform filtering after the requantization, which simply fetches the array values. The end result is that TNS works much better now and doesn't attenuate anything but the actual signal, e.g. TNS removes quantization errors and does it's job correctly now. It might be enabled by default soon since it doesn't hurt and helps reduce nastyness at low bitrates. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-29 07:47:31 +02:00
put_bits(&s->pb, 1, !!sce->tns.present);
if (s->coder->encode_tns_info)
s->coder->encode_tns_info(s, sce);
put_bits(&s->pb, 1, 0); //ssr
encode_spectral_coeffs(s, sce);
return 0;
}
/**
* Write some auxiliary information about the created AAC file.
*/
static void put_bitstream_info(AACEncContext *s, const char *name)
{
int i, namelen, padbits;
namelen = strlen(name) + 2;
put_bits(&s->pb, 3, TYPE_FIL);
put_bits(&s->pb, 4, FFMIN(namelen, 15));
if (namelen >= 15)
put_bits(&s->pb, 8, namelen - 14);
put_bits(&s->pb, 4, 0); //extension type - filler
padbits = -put_bits_count(&s->pb) & 7;
avpriv_align_put_bits(&s->pb);
for (i = 0; i < namelen - 2; i++)
put_bits(&s->pb, 8, name[i]);
put_bits(&s->pb, 12 - padbits, 0);
}
/*
2012-08-25 19:04:33 +02:00
* Copy input samples.
* Channels are reordered from libavcodec's default order to AAC order.
*/
2012-08-25 19:04:33 +02:00
static void copy_input_samples(AACEncContext *s, const AVFrame *frame)
{
2012-08-25 19:04:33 +02:00
int ch;
int end = 2048 + (frame ? frame->nb_samples : 0);
const uint8_t *channel_map = aac_chan_maps[s->channels - 1];
2012-08-25 19:04:33 +02:00
/* copy and remap input samples */
for (ch = 0; ch < s->channels; ch++) {
/* copy last 1024 samples of previous frame to the start of the current frame */
memcpy(&s->planar_samples[ch][1024], &s->planar_samples[ch][2048], 1024 * sizeof(s->planar_samples[0][0]));
2012-08-25 19:04:33 +02:00
/* copy new samples and zero any remaining samples */
2012-01-28 18:28:01 +01:00
if (frame) {
2012-08-25 19:04:33 +02:00
memcpy(&s->planar_samples[ch][2048],
frame->extended_data[channel_map[ch]],
frame->nb_samples * sizeof(s->planar_samples[0][0]));
}
2012-08-25 19:04:33 +02:00
memset(&s->planar_samples[ch][end], 0,
(3072 - end) * sizeof(s->planar_samples[0][0]));
}
}
2012-01-28 18:28:01 +01:00
static int aac_encode_frame(AVCodecContext *avctx, AVPacket *avpkt,
const AVFrame *frame, int *got_packet_ptr)
{
AACEncContext *s = avctx->priv_data;
float **samples = s->planar_samples, *samples2, *la, *overlap;
ChannelElement *cpe;
SingleChannelElement *sce;
int i, its, ch, w, chans, tag, start_ch, ret, frame_bits;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
int target_bits, rate_bits, too_many_bits, too_few_bits;
int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
int chan_el_counter[4];
FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
if (s->last_frame == 2)
return 0;
2012-01-28 18:28:01 +01:00
/* add current frame to queue */
if (frame) {
if ((ret = ff_af_queue_add(&s->afq, frame)) < 0)
2012-01-28 18:28:01 +01:00
return ret;
}
2012-08-25 19:04:33 +02:00
copy_input_samples(s, frame);
if (s->psypp)
ff_psy_preprocess(s->psypp, s->planar_samples, s->channels);
if (!avctx->frame_number)
return 0;
start_ch = 0;
for (i = 0; i < s->chan_map[0]; i++) {
FFPsyWindowInfo* wi = windows + start_ch;
tag = s->chan_map[i+1];
chans = tag == TYPE_CPE ? 2 : 1;
cpe = &s->cpe[i];
for (ch = 0; ch < chans; ch++) {
IndividualChannelStream *ics = &cpe->ch[ch].ics;
int cur_channel = start_ch + ch;
float clip_avoidance_factor;
overlap = &samples[cur_channel][0];
samples2 = overlap + 1024;
la = samples2 + (448+64);
2012-01-28 18:28:01 +01:00
if (!frame)
la = NULL;
if (tag == TYPE_LFE) {
wi[ch].window_type[0] = ONLY_LONG_SEQUENCE;
wi[ch].window_shape = 0;
wi[ch].num_windows = 1;
wi[ch].grouping[0] = 1;
/* Only the lowest 12 coefficients are used in a LFE channel.
* The expression below results in only the bottom 8 coefficients
* being used for 11.025kHz to 16kHz sample rates.
*/
ics->num_swb = s->samplerate_index >= 8 ? 1 : 3;
} else {
wi[ch] = s->psy.model->window(&s->psy, samples2, la, cur_channel,
ics->window_sequence[0]);
}
ics->window_sequence[1] = ics->window_sequence[0];
ics->window_sequence[0] = wi[ch].window_type[0];
ics->use_kb_window[1] = ics->use_kb_window[0];
ics->use_kb_window[0] = wi[ch].window_shape;
ics->num_windows = wi[ch].num_windows;
ics->swb_sizes = s->psy.bands [ics->num_windows == 8];
ics->num_swb = tag == TYPE_LFE ? ics->num_swb : s->psy.num_bands[ics->num_windows == 8];
ics->swb_offset = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
ff_swb_offset_128 [s->samplerate_index]:
ff_swb_offset_1024[s->samplerate_index];
ics->tns_max_bands = wi[ch].window_type[0] == EIGHT_SHORT_SEQUENCE ?
ff_tns_max_bands_128 [s->samplerate_index]:
ff_tns_max_bands_1024[s->samplerate_index];
clip_avoidance_factor = 0.0f;
for (w = 0; w < ics->num_windows; w++)
ics->group_len[w] = wi[ch].grouping[w];
for (w = 0; w < ics->num_windows; w++) {
if (wi[ch].clipping[w] > CLIP_AVOIDANCE_FACTOR) {
ics->window_clipping[w] = 1;
clip_avoidance_factor = FFMAX(clip_avoidance_factor, wi[ch].clipping[w]);
} else {
ics->window_clipping[w] = 0;
}
}
if (clip_avoidance_factor > CLIP_AVOIDANCE_FACTOR) {
ics->clip_avoidance_factor = CLIP_AVOIDANCE_FACTOR / clip_avoidance_factor;
} else {
ics->clip_avoidance_factor = 1.0f;
}
apply_window_and_mdct(s, &cpe->ch[ch], overlap);
if (isnan(cpe->ch->coeffs[0])) {
av_log(avctx, AV_LOG_ERROR, "Input contains NaN\n");
return AVERROR(EINVAL);
}
avoid_clipping(s, &cpe->ch[ch]);
}
start_ch += chans;
}
if ((ret = ff_alloc_packet2(avctx, avpkt, 8192 * s->channels, 0)) < 0)
return ret;
frame_bits = its = 0;
do {
2012-01-28 18:28:01 +01:00
init_put_bits(&s->pb, avpkt->data, avpkt->size);
if ((avctx->frame_number & 0xFF)==1 && !(avctx->flags & AV_CODEC_FLAG_BITEXACT))
put_bitstream_info(s, LIBAVCODEC_IDENT);
start_ch = 0;
target_bits = 0;
memset(chan_el_counter, 0, sizeof(chan_el_counter));
for (i = 0; i < s->chan_map[0]; i++) {
FFPsyWindowInfo* wi = windows + start_ch;
const float *coeffs[2];
tag = s->chan_map[i+1];
chans = tag == TYPE_CPE ? 2 : 1;
cpe = &s->cpe[i];
cpe->common_window = 0;
memset(cpe->is_mask, 0, sizeof(cpe->is_mask));
memset(cpe->ms_mask, 0, sizeof(cpe->ms_mask));
put_bits(&s->pb, 3, tag);
put_bits(&s->pb, 4, chan_el_counter[tag]++);
for (ch = 0; ch < chans; ch++) {
sce = &cpe->ch[ch];
coeffs[ch] = sce->coeffs;
sce->ics.predictor_present = 0;
memset(&sce->ics.prediction_used, 0, sizeof(sce->ics.prediction_used));
memset(&sce->tns, 0, sizeof(TemporalNoiseShaping));
for (w = 0; w < 128; w++)
if (sce->band_type[w] > RESERVED_BT)
sce->band_type[w] = 0;
}
s->psy.bitres.alloc = -1;
s->psy.bitres.bits = avctx->frame_bits / s->channels;
s->psy.model->analyze(&s->psy, start_ch, coeffs, wi);
if (s->psy.bitres.alloc > 0) {
/* Lambda unused here on purpose, we need to take psy's unscaled allocation */
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
target_bits += s->psy.bitres.alloc
* (s->lambda / (avctx->global_quality ? avctx->global_quality : 120));
s->psy.bitres.alloc /= chans;
}
s->cur_type = tag;
for (ch = 0; ch < chans; ch++) {
s->cur_channel = start_ch + ch;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if (s->options.pns && s->coder->mark_pns)
s->coder->mark_pns(s, avctx, &cpe->ch[ch]);
s->coder->search_for_quantizers(avctx, s, &cpe->ch[ch], s->lambda);
}
if (chans > 1
&& wi[0].window_type[0] == wi[1].window_type[0]
&& wi[0].window_shape == wi[1].window_shape) {
cpe->common_window = 1;
for (w = 0; w < wi[0].num_windows; w++) {
if (wi[0].grouping[w] != wi[1].grouping[w]) {
cpe->common_window = 0;
break;
}
}
}
for (ch = 0; ch < chans; ch++) { /* TNS and PNS */
sce = &cpe->ch[ch];
s->cur_channel = start_ch + ch;
if (s->options.pns && s->coder->search_for_pns)
s->coder->search_for_pns(s, avctx, sce);
if (s->options.tns && s->coder->search_for_tns)
s->coder->search_for_tns(s, sce);
aacenc_tns: rework the way coefficients are calculated This commit abandons the way the specifications state to quantize the coefficients, makes use of the new LPC float functions and is much better. The original way of converting non-normalized float samples to int32_t which out LPC system expects was wrong and it was wrong to assume the coefficients that are generated are also valid. It was essentially a full garbage-in, garbage-out system and it definitely shows when looking at spectrals and listening. The high frequencies were very overattenuated. The new LPC function performs the analysis directly. The specifications state to quantize the coefficients into four bit index values using an asin() function which of course had to have ugly ternary operators because the function turns negative if the coefficients are negative which when encoding causes invalid bitstream to get generated. This deviates from this by using the direct TNS tables, which are fairly small since you only have 4 bits at most for index values. The LPC values are directly quantized against the tables and are then used to perform filtering after the requantization, which simply fetches the array values. The end result is that TNS works much better now and doesn't attenuate anything but the actual signal, e.g. TNS removes quantization errors and does it's job correctly now. It might be enabled by default soon since it doesn't hurt and helps reduce nastyness at low bitrates. Signed-off-by: Rostislav Pehlivanov <atomnuker@gmail.com>
2015-08-29 07:47:31 +02:00
if (s->options.tns && s->coder->apply_tns_filt)
s->coder->apply_tns_filt(s, sce);
if (sce->tns.present)
tns_mode = 1;
}
s->cur_channel = start_ch;
if (s->options.intensity_stereo) { /* Intensity Stereo */
if (s->coder->search_for_is)
s->coder->search_for_is(s, avctx, cpe);
aacenc: implement Intensity Stereo encoding support This commit implements intensity stereo coding support to the native aac encoder. This is a way to increase the efficiency of the encoder by zeroing the right channel's spectral coefficients (in a channel pair) and rederiving them in the decoder using information from the scalefactor indices of special band types. This commit confomrs to the official ISO 13818-7 specifications, although due to their ambiguity certain deviations have been taken to ensure maximum sound quality. This commit has been extensively tested and has shown to not result in audiable audio artifacts unless in extreme cases. This commit also adds an option, aac_is, which has the value of 0 by default. Intensity Stereo is part of the scalable aac profile and is thus non-default. The way IS coding works is that it rederives the right channel's spectral coefficients from the left channel via the scalefactor index values left in the right channel. Since an entire band's spectral coefficients do not need to be coded, the encoder's efficiency jumps up and it unzeroes some high frequency values which it previously did not have enough bits to encode. That way less information is lost than the information lost by rederiving the spectral coefficients with some error. This is why the filesize of files encoded with IS do not decrease significantly. Users wishing that IS coding should reduce filesize are expected to reduce their encoding bitrates appropriately. This is V2 of the commit. The old version did not mark ms_mask as 0 since M/S and IS coding are incompactible, which resulted in distortions with M/S coding enabled. This version also improves phase detection by measuring it for every spectral coefficient in the band and using a simple majority rule to determine whether the coefficients are in or out of phase. Also, the energy values per spectral coefficient were changed as to reflect the official specifications. Reviewed-by: Claudio Freire <klaussfreire@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-07-02 20:13:07 +02:00
if (cpe->is_mode) is_mode = 1;
apply_intensity_stereo(cpe);
aacenc: implement Intensity Stereo encoding support This commit implements intensity stereo coding support to the native aac encoder. This is a way to increase the efficiency of the encoder by zeroing the right channel's spectral coefficients (in a channel pair) and rederiving them in the decoder using information from the scalefactor indices of special band types. This commit confomrs to the official ISO 13818-7 specifications, although due to their ambiguity certain deviations have been taken to ensure maximum sound quality. This commit has been extensively tested and has shown to not result in audiable audio artifacts unless in extreme cases. This commit also adds an option, aac_is, which has the value of 0 by default. Intensity Stereo is part of the scalable aac profile and is thus non-default. The way IS coding works is that it rederives the right channel's spectral coefficients from the left channel via the scalefactor index values left in the right channel. Since an entire band's spectral coefficients do not need to be coded, the encoder's efficiency jumps up and it unzeroes some high frequency values which it previously did not have enough bits to encode. That way less information is lost than the information lost by rederiving the spectral coefficients with some error. This is why the filesize of files encoded with IS do not decrease significantly. Users wishing that IS coding should reduce filesize are expected to reduce their encoding bitrates appropriately. This is V2 of the commit. The old version did not mark ms_mask as 0 since M/S and IS coding are incompactible, which resulted in distortions with M/S coding enabled. This version also improves phase detection by measuring it for every spectral coefficient in the band and using a simple majority rule to determine whether the coefficients are in or out of phase. Also, the energy values per spectral coefficient were changed as to reflect the official specifications. Reviewed-by: Claudio Freire <klaussfreire@gmail.com> Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-07-02 20:13:07 +02:00
}
if (s->options.pred) { /* Prediction */
for (ch = 0; ch < chans; ch++) {
sce = &cpe->ch[ch];
s->cur_channel = start_ch + ch;
if (s->options.pred && s->coder->search_for_pred)
s->coder->search_for_pred(s, sce);
if (cpe->ch[ch].ics.predictor_present) pred_mode = 1;
}
if (s->coder->adjust_common_prediction)
s->coder->adjust_common_prediction(s, cpe);
for (ch = 0; ch < chans; ch++) {
sce = &cpe->ch[ch];
s->cur_channel = start_ch + ch;
if (s->options.pred && s->coder->apply_main_pred)
s->coder->apply_main_pred(s, sce);
}
s->cur_channel = start_ch;
}
if (s->options.stereo_mode) { /* Mid/Side stereo */
if (s->options.stereo_mode == -1 && s->coder->search_for_ms)
s->coder->search_for_ms(s, cpe);
else if (cpe->common_window)
memset(cpe->ms_mask, 1, sizeof(cpe->ms_mask));
apply_mid_side_stereo(cpe);
}
adjust_frame_information(cpe, chans);
if (chans == 2) {
put_bits(&s->pb, 1, cpe->common_window);
if (cpe->common_window) {
put_ics_info(s, &cpe->ch[0].ics);
if (s->coder->encode_main_pred)
s->coder->encode_main_pred(s, &cpe->ch[0]);
encode_ms_info(&s->pb, cpe);
AAC: Fix M/S stereo encoding This patch fixes a pointer arithmetic bug in adjust_frame_information that resulted in heavily corrupted audio when using M/S encoding. Also, a backup copy of untransformed coefficients has to be kept around or attempts at re-processing the frame (which happens when hevavily overspending bits during transients) will result in re-encoding of the coefficients and subsequent corruption of the resulting stream. A/B testing shows the bug as corrected, but still cannot prove that M/S coding is a win at least in numbers. Limited listening tests do show improvement on M/S encoded samples in lower bitrates, but they're hidden among the other artifacts that remain to be corrected in the encoder. Some of the regressions flagged in the report do show poor stereo image (but not buggy), so M/S encoding is clearly not good enough yet to be defaulted to auto. In numbers, Patched against Unpatched, stereo_mode auto: Files: 114 Bitrates: 6 Tests: 683 Serious Regressions: 0 (0%) Regressions: 0 (0%) Improvements: 227 (33%) Big improvements: 92 (13%) Worst regression - mybloodrusts.wv - 256k - StdDev: 28.61 pSNR: -0.43 maxdiff: 1372.00 Best improvement - 60.wv - 384k - StdDev: -369.57 pSNR: 45.02 maxdiff: -13322.00 Average - StdDev: -80.56 pSNR: 2.49 maxdiff: -8858.00 Patched against Unpatched stereo_mode ms_off shows no difference. Patched stereo_mode auto vs Unpatched stereo_mode ms_off shows a small average improvement, just not too significant: Serious Regressions: 0 (0%) Regressions: 10 (1%) Improvements: 45 (6%) Big improvements: 2 (0%) Worst regression - Illinois.wv - 256k - StdDev: 33.20 pSNR: -2.03 maxdiff: 477.00 Best improvement - song_of_circomstances.flac - 384k - StdDev: -3.97 pSNR: 7.61 maxdiff: -826.00 Average - StdDev: -10.25 pSNR: 0.20 maxdiff: -281.00 Signed-off-by: Michael Niedermayer <michaelni@gmx.at>
2015-03-03 07:43:06 +01:00
if (cpe->ms_mode) ms_mode = 1;
}
}
for (ch = 0; ch < chans; ch++) {
s->cur_channel = start_ch + ch;
encode_individual_channel(avctx, s, &cpe->ch[ch], cpe->common_window);
}
start_ch += chans;
}
if (avctx->flags & CODEC_FLAG_QSCALE) {
/* When using a constant Q-scale, don't mess with lambda */
break;
}
/* rate control stuff
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
* allow between the nominal bitrate, and what psy's bit reservoir says to target
* but drift towards the nominal bitrate always
*/
frame_bits = put_bits_count(&s->pb);
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
rate_bits = avctx->bit_rate * 1024 / avctx->sample_rate;
rate_bits = FFMIN(rate_bits, 6144 * s->channels - 3);
too_many_bits = FFMAX(target_bits, rate_bits);
too_many_bits = FFMIN(too_many_bits, 6144 * s->channels - 3);
too_few_bits = FFMIN(FFMAX(rate_bits - rate_bits/4, target_bits), too_many_bits);
/* When using ABR, be strict (but only for increasing) */
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
too_few_bits = too_few_bits - too_few_bits/8;
too_many_bits = too_many_bits + too_many_bits/2;
if ( its == 0 /* for steady-state Q-scale tracking */
|| (its < 5 && (frame_bits < too_few_bits || frame_bits > too_many_bits))
|| frame_bits >= 6144 * s->channels - 3 )
{
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
float ratio = ((float)rate_bits) / frame_bits;
if (frame_bits >= too_few_bits && frame_bits <= too_many_bits) {
/*
* This path is for steady-state Q-scale tracking
* When frame bits fall within the stable range, we still need to adjust
* lambda to maintain it like so in a stable fashion (large jumps in lambda
* create artifacts and should be avoided), but slowly
*/
ratio = sqrtf(sqrtf(ratio));
ratio = av_clipf(ratio, 0.9f, 1.1f);
} else {
/* Not so fast though */
ratio = sqrtf(ratio);
}
s->lambda = FFMIN(s->lambda * ratio, 65536.f);
/* Keep iterating if we must reduce and lambda is in the sky */
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
if ((s->lambda < 300.f || ratio > 0.9f) && (s->lambda > 10.f || ratio < 1.1f)) {
break;
} else {
if (is_mode || ms_mode || tns_mode || pred_mode) {
for (i = 0; i < s->chan_map[0]; i++) {
// Must restore coeffs
chans = tag == TYPE_CPE ? 2 : 1;
cpe = &s->cpe[i];
for (ch = 0; ch < chans; ch++)
memcpy(cpe->ch[ch].coeffs, cpe->ch[ch].pcoeffs, sizeof(cpe->ch[ch].coeffs));
}
}
its++;
}
} else {
break;
}
} while (1);
put_bits(&s->pb, 3, TYPE_END);
flush_put_bits(&s->pb);
avctx->frame_bits = put_bits_count(&s->pb);
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
s->lambda_sum += s->lambda;
s->lambda_count++;
2012-01-28 18:28:01 +01:00
if (!frame)
s->last_frame++;
2012-01-28 18:28:01 +01:00
ff_af_queue_remove(&s->afq, avctx->frame_size, &avpkt->pts,
&avpkt->duration);
avpkt->size = put_bits_count(&s->pb) >> 3;
*got_packet_ptr = 1;
return 0;
}
static av_cold int aac_encode_end(AVCodecContext *avctx)
{
AACEncContext *s = avctx->priv_data;
AAC encoder: Extensive improvements This finalizes merging of the work in the patches in ticket #2686. Improvements to twoloop and RC logic are extensive. The non-exhaustive list of twoloop improvments includes: - Tweaks to distortion limits on the RD optimization phase of twoloop - Deeper search in twoloop - PNS information marking to let twoloop decide when to use it (turned out having the decision made separately wasn't working) - Tonal band detection and priorization - Better band energy conservation rules - Strict hole avoidance For rate control: - Use psymodel's bit allocation to allow proper use of the bit reservoir. Don't work against the bit reservoir by moving lambda in the opposite direction when psymodel decides to allocate more/less bits to a frame. - Retry the encode if the effective rate lies outside a reasonable margin of psymodel's allocation or the selected ABR. - Log average lambda at the end. Useful info for everyone, but especially for tuning of the various encoder constants that relate to lambda feedback. Psy: - Do not apply lowpass with a FIR filter, instead just let the coder zero bands above the cutoff. The FIR filter induces group delay, and while zeroing bands causes ripple, it's lost in the quantization noise. - Experimental VBR bit allocation code - Tweak automatic lowpass filter threshold to maximize audio bandwidth at all bitrates while still providing acceptable, stable quality. I/S: - Phase decision fixes. Unrelated to #2686, but the bugs only surfaced when the merge was finalized. Measure I/S band energy accounting for phase, and prevent I/S and M/S from being applied both. PNS: - Avoid marking short bands with PNS when they're part of a window group in which there's a large variation of energy from one window to the next. PNS can't preserve those and the effect is extremely noticeable. M/S: - Implement BMLD protection similar to the specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Since M/S decision doesn't conform to section 6.1, a different method had to be implemented, but should provide equivalent protection. - Move the decision logic closer to the method specified in ISO-IEC/13818:7-2003, Appendix C Section 6.1. Specifically, make sure M/S needs less bits than dual stereo. - Don't apply M/S in bands that are using I/S Now, this of course needed adjustments in the compare targets and fuzz factors of the AAC encoder's fate tests, but if wondering why the targets go up (more distortion), consider the previous coder was using too many bits on LF content (far more than required by psy), and thus those signals will now be more distorted, not less. The extra distortion isn't audible though, I carried extensive ABX testing to make sure. A very similar patch was also extensively tested by Kamendo2 in the context of #2686.
2015-10-11 22:29:50 +02:00
av_log(avctx, AV_LOG_INFO, "Qavg: %.3f\n", s->lambda_sum / s->lambda_count);
ff_mdct_end(&s->mdct1024);
ff_mdct_end(&s->mdct128);
ff_psy_end(&s->psy);
ff_lpc_end(&s->lpc);
if (s->psypp)
ff_psy_preprocess_end(s->psypp);
av_freep(&s->buffer.samples);
av_freep(&s->cpe);
av_freep(&s->fdsp);
2012-01-28 18:28:01 +01:00
ff_af_queue_close(&s->afq);
return 0;
}
static av_cold int dsp_init(AVCodecContext *avctx, AACEncContext *s)
{
int ret = 0;
s->fdsp = avpriv_float_dsp_alloc(avctx->flags & AV_CODEC_FLAG_BITEXACT);
if (!s->fdsp)
return AVERROR(ENOMEM);
// window init
ff_kbd_window_init(ff_aac_kbd_long_1024, 4.0, 1024);
ff_kbd_window_init(ff_aac_kbd_short_128, 6.0, 128);
ff_init_ff_sine_windows(10);
ff_init_ff_sine_windows(7);
if ((ret = ff_mdct_init(&s->mdct1024, 11, 0, 32768.0)) < 0)
return ret;
if ((ret = ff_mdct_init(&s->mdct128, 8, 0, 32768.0)) < 0)
return ret;
return 0;
}
static av_cold int alloc_buffers(AVCodecContext *avctx, AACEncContext *s)
{
int ch;
FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->buffer.samples, s->channels, 3 * 1024 * sizeof(s->buffer.samples[0]), alloc_fail);
FF_ALLOCZ_ARRAY_OR_GOTO(avctx, s->cpe, s->chan_map[0], sizeof(ChannelElement), alloc_fail);
FF_ALLOCZ_OR_GOTO(avctx, avctx->extradata, 5 + AV_INPUT_BUFFER_PADDING_SIZE, alloc_fail);
for(ch = 0; ch < s->channels; ch++)
s->planar_samples[ch] = s->buffer.samples + 3 * 1024 * ch;
return 0;
alloc_fail:
return AVERROR(ENOMEM);
}
static av_cold int aac_encode_init(AVCodecContext *avctx)
{
AACEncContext *s = avctx->priv_data;
int i, ret = 0;
const uint8_t *sizes[2];
uint8_t grouping[AAC_MAX_CHANNELS];
int lengths[2];
avctx->frame_size = 1024;
for (i = 0; i < 16; i++)
if (avctx->sample_rate == avpriv_mpeg4audio_sample_rates[i])
break;
s->channels = avctx->channels;
ERROR_IF(i == 16 || i >= ff_aac_swb_size_1024_len || i >= ff_aac_swb_size_128_len,
"Unsupported sample rate %d\n", avctx->sample_rate);
ERROR_IF(s->channels > AAC_MAX_CHANNELS,
"Unsupported number of channels: %d\n", s->channels);
WARN_IF(1024.0 * avctx->bit_rate / avctx->sample_rate > 6144 * s->channels,
"Too many bits per frame requested, clamping to max\n");
if (avctx->profile == FF_PROFILE_AAC_MAIN) {
s->options.pred = 1;
} else if ((avctx->profile == FF_PROFILE_AAC_LOW ||
avctx->profile == FF_PROFILE_UNKNOWN) && s->options.pred) {
s->profile = 0; /* Main */
WARN_IF(1, "Prediction requested, changing profile to AAC-Main\n");
} else if (avctx->profile == FF_PROFILE_AAC_LOW ||
avctx->profile == FF_PROFILE_UNKNOWN) {
s->profile = 1; /* Low */
} else {
ERROR_IF(1, "Unsupported profile %d\n", avctx->profile);
}
if (s->options.aac_coder != AAC_CODER_TWOLOOP) {
s->options.intensity_stereo = 0;
s->options.pns = 0;
}
avctx->bit_rate = (int)FFMIN(
6144 * s->channels / 1024.0 * avctx->sample_rate,
avctx->bit_rate);
s->samplerate_index = i;
s->chan_map = aac_chan_configs[s->channels-1];
if ((ret = dsp_init(avctx, s)) < 0)
goto fail;
if ((ret = alloc_buffers(avctx, s)) < 0)
goto fail;
avctx->extradata_size = 5;
put_audio_specific_config(avctx);
sizes[0] = ff_aac_swb_size_1024[i];
sizes[1] = ff_aac_swb_size_128[i];
lengths[0] = ff_aac_num_swb_1024[i];
lengths[1] = ff_aac_num_swb_128[i];
for (i = 0; i < s->chan_map[0]; i++)
grouping[i] = s->chan_map[i + 1] == TYPE_CPE;
if ((ret = ff_psy_init(&s->psy, avctx, 2, sizes, lengths,
s->chan_map[0], grouping)) < 0)
goto fail;
s->psypp = ff_psy_preprocess_init(avctx);
s->coder = &ff_aac_coders[s->options.aac_coder];
ff_lpc_init(&s->lpc, 2*avctx->frame_size, TNS_MAX_ORDER, FF_LPC_TYPE_LEVINSON);
if (HAVE_MIPSDSPR1)
ff_aac_coder_init_mips(s);
s->lambda = avctx->global_quality > 0 ? avctx->global_quality : 120;
s->random_state = 0x1f2e3d4c;
ff_aac_tableinit();
avctx->initial_padding = 1024;
2012-01-28 18:28:01 +01:00
ff_af_queue_init(avctx, &s->afq);
return 0;
fail:
aac_encode_end(avctx);
return ret;
}
#define AACENC_FLAGS AV_OPT_FLAG_ENCODING_PARAM | AV_OPT_FLAG_AUDIO_PARAM
static const AVOption aacenc_options[] = {
{"stereo_mode", "Stereo coding method", offsetof(AACEncContext, options.stereo_mode), AV_OPT_TYPE_INT, {.i64 = 0}, -1, 1, AACENC_FLAGS, "stereo_mode"},
{"auto", "Selected by the Encoder", 0, AV_OPT_TYPE_CONST, {.i64 = -1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
{"ms_off", "Disable Mid/Side coding", 0, AV_OPT_TYPE_CONST, {.i64 = 0 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
{"ms_force", "Force Mid/Side for the whole frame if possible", 0, AV_OPT_TYPE_CONST, {.i64 = 1 }, INT_MIN, INT_MAX, AACENC_FLAGS, "stereo_mode"},
{"aac_coder", "Coding algorithm", offsetof(AACEncContext, options.aac_coder), AV_OPT_TYPE_INT, {.i64 = AAC_CODER_TWOLOOP}, 0, AAC_CODER_NB-1, AACENC_FLAGS, "aac_coder"},
{"faac", "FAAC-inspired method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAAC}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
{"anmr", "ANMR method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_ANMR}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
{"twoloop", "Two loop searching method", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_TWOLOOP}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
{"fast", "Constant quantizer", 0, AV_OPT_TYPE_CONST, {.i64 = AAC_CODER_FAST}, INT_MIN, INT_MAX, AACENC_FLAGS, "aac_coder"},
2015-09-06 19:22:23 +02:00
{"aac_pns", "Perceptual Noise Substitution", offsetof(AACEncContext, options.pns), AV_OPT_TYPE_BOOL, {.i64 = 1}, 0, 1, AACENC_FLAGS},
{"aac_is", "Intensity stereo coding", offsetof(AACEncContext, options.intensity_stereo), AV_OPT_TYPE_BOOL, {.i64 = 1}, 0, 1, AACENC_FLAGS},
{"aac_tns", "Temporal noise shaping", offsetof(AACEncContext, options.tns), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, AACENC_FLAGS},
{"aac_pred", "AAC-Main prediction", offsetof(AACEncContext, options.pred), AV_OPT_TYPE_BOOL, {.i64 = 0}, 0, 1, AACENC_FLAGS},
{NULL}
};
static const AVClass aacenc_class = {
"AAC encoder",
av_default_item_name,
aacenc_options,
LIBAVUTIL_VERSION_INT,
};
AVCodec ff_aac_encoder = {
.name = "aac",
.long_name = NULL_IF_CONFIG_SMALL("AAC (Advanced Audio Coding)"),
.type = AVMEDIA_TYPE_AUDIO,
.id = AV_CODEC_ID_AAC,
.priv_data_size = sizeof(AACEncContext),
.init = aac_encode_init,
2012-01-28 18:28:01 +01:00
.encode2 = aac_encode_frame,
.close = aac_encode_end,
.supported_samplerates = mpeg4audio_sample_rates,
.capabilities = AV_CODEC_CAP_SMALL_LAST_FRAME | AV_CODEC_CAP_DELAY |
AV_CODEC_CAP_EXPERIMENTAL,
2012-08-25 19:04:33 +02:00
.sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_FLTP,
AV_SAMPLE_FMT_NONE },
.priv_class = &aacenc_class,
};