aaccoder: move the Intensity Stereo implementation out

This commit moves the intensity stereo implementation out from aaccoder and into a separate file. This was possible using the previous commits. This commit also drastically improves the IS implementation by making it phase invariant e.g. it will always choose the best possible phase regardless of whether M/S coding is on or most of the coefficients have identical phases. This also increases the quality and reduces any distortions introduced by enablind intensity stereo. Users are encouraged to test it out using the -aac_is 1 parameter as it has always been.

Rostislav Pehlivanov committed 11 years ago

Commit d1ca7141 parent 860dbe05 files · +192 −102Browse files

Unified · Split

5 files

libavcodec/Makefile +1 −0
libavcodec/aaccoder.c +3 −99
libavcodec/aacenc.c +5 −3
libavcodec/aacenc_is.c +136 −0
libavcodec/aacenc_is.h +47 −0

libavcodec/Makefile+1−0

@@ -132,6 +132,7 @@ OBJS-$(CONFIG_AAC_FIXED_DECODER) += aacdec_fixed.o aactab.o aacsbr_fixed.o · 131 unmodified lines
132	132	sbrdsp_fixed.o aacpsdsp_fixed.o
133	133	OBJS-$(CONFIG_AAC_ENCODER) += aacenc.o aaccoder.o aacenctab.o \
134	134	aacpsy.o aactab.o \
	135	+ aacenc_is.o \
135	136	psymodel.o mpeg4audio.o kbdwin.o
136	137	OBJS-$(CONFIG_AASC_DECODER) += aasc.o msrledec.o
137	138	OBJS-$(CONFIG_AC3_DECODER) += ac3dec_float.o ac3dec_data.o ac3.o kbdwin.o

libavcodec/aaccoder.c+3−99

@@ -44,6 +44,8 @@ · 43 unmodified lines
44	44	#include "aacenc_quantization.h"
45	45	#include "aac_tablegen_decl.h"
46	46
	47	+#include "aacenc_is.h"
	48	+
47	49	/ Frequency in Hz for lower limit of noise substitution /
48	50	#define NOISE_LOW_LIMIT 4500
49	51
@@ -889,104 +891,6 @@ static void search_for_pns(AACEncContext s, AVCodecContext avctx, SingleChanne · 839 unmodified lines
889	891	}
890	892	}
891	893
892		−static void search_for_is(AACEncContext s, AVCodecContext avctx, ChannelElement *cpe)
893		−{
894		− float IS[128];
895		− float L34 = s->scoefs + 1280, R34 = s->scoefs + 1281;
896		− float I34 = s->scoefs + 1282;
897		− SingleChannelElement *sce0 = &cpe->ch[0];
898		− SingleChannelElement *sce1 = &cpe->ch[1];
899		− int start = 0, count = 0, i, w, w2, g;
900		− const float freq_mult = avctx->sample_rate/(1024.0f/sce0->ics.num_windows)/2.0f;
901		− const float lambda = s->lambda;
902		−
903		− for (w = 0; w < 128; w++)
904		− if (sce1->band_type[w] >= INTENSITY_BT2)
905		− sce1->band_type[w] = 0;
906		−
907		− if (!cpe->common_window)
908		− return;
909		− for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
910		− start = 0;
911		− for (g = 0; g < sce0->ics.num_swb; g++) {
912		− if (startfreq_mult > INT_STEREO_LOW_LIMIT(lambda/170.0f) &&
913		− cpe->ch[0].band_type[w16+g] != NOISE_BT && !cpe->ch[0].zeroes[w16+g] &&
914		− cpe->ch[1].band_type[w16+g] != NOISE_BT && !cpe->ch[1].zeroes[w16+g]) {
915		− int phase = 0;
916		− float ener0 = 0.0f, ener1 = 0.0f, ener01 = 0.0f;
917		− float dist1 = 0.0f, dist2 = 0.0f;
918		− for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
919		− for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
920		− float coef0 = sce0->pcoeffs[start+(w+w2)*128+i];
921		− float coef1 = sce1->pcoeffs[start+(w+w2)*128+i];
922		− phase += coef0*coef1 >= 0.0f ? 1 : -1;
923		− ener0 += coef0*coef0;
924		− ener1 += coef1*coef1;
925		− ener01 += (coef0 + coef1)*(coef0 + coef1);
926		− }
927		− }
928		− if (!phase) { /* Too much phase difference between channels */
929		− start += sce0->ics.swb_sizes[g];
930		− continue;
931		− }
932		− phase = av_clip(phase, -1, 1);
933		− for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
934		− FFPsyBand band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)16+g];
935		− FFPsyBand band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)16+g];
936		− int is_band_type, is_sf_idx = FFMAX(1, sce0->sf_idx[(w+w2)*16+g]-4);
937		− float e01_34 = phase*pow(sqrt(ener1/ener0), 3.0/4.0);
938		− float maxval, dist_spec_err = 0.0f;
939		− float minthr = FFMIN(band0->threshold, band1->threshold);
940		− for (i = 0; i < sce0->ics.swb_sizes[g]; i++)
941		− IS[i] = (sce0->pcoeffs[start+(w+w2)128+i] + phasesce1->pcoeffs[start+(w+w2)128+i]) sqrt(ener0/ener01);
942		− abs_pow34_v(L34, sce0->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);
943		− abs_pow34_v(R34, sce1->coeffs+start+(w+w2)*128, sce0->ics.swb_sizes[g]);
944		− abs_pow34_v(I34, IS, sce0->ics.swb_sizes[g]);
945		− maxval = find_max_val(1, sce0->ics.swb_sizes[g], I34);
946		− is_band_type = find_min_book(maxval, is_sf_idx);
947		− dist1 += quantize_band_cost(s, sce0->coeffs + start + (w+w2)*128,
948		− L34,
949		− sce0->ics.swb_sizes[g],
950		− sce0->sf_idx[(w+w2)*16+g],
951		− sce0->band_type[(w+w2)*16+g],
952		− lambda / band0->threshold, INFINITY, NULL, 0);
953		− dist1 += quantize_band_cost(s, sce1->coeffs + start + (w+w2)*128,
954		− R34,
955		− sce1->ics.swb_sizes[g],
956		− sce1->sf_idx[(w+w2)*16+g],
957		− sce1->band_type[(w+w2)*16+g],
958		− lambda / band1->threshold, INFINITY, NULL, 0);
959		− dist2 += quantize_band_cost(s, IS,
960		− I34,
961		− sce0->ics.swb_sizes[g],
962		− is_sf_idx,
963		− is_band_type,
964		− lambda / minthr, INFINITY, NULL, 0);
965		− for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
966		− dist_spec_err += (L34[i] - I34[i])*(L34[i] - I34[i]);
967		− dist_spec_err += (R34[i] - I34[i]e01_34)(R34[i] - I34[i]*e01_34);
968		− }
969		− dist_spec_err *= lambda / minthr;
970		− dist2 += dist_spec_err;
971		− }
972		− if (dist2 <= dist1) {
973		− cpe->is_mask[w*16+g] = 1;
974		− cpe->ms_mask[w*16+g] = 0;
975		− cpe->ch[0].is_ener[w*16+g] = sqrt(ener0/ener01);
976		− cpe->ch[1].is_ener[w*16+g] = ener0/ener1;
977		− if (phase)
978		− cpe->ch[1].band_type[w*16+g] = INTENSITY_BT;
979		− else
980		− cpe->ch[1].band_type[w*16+g] = INTENSITY_BT2;
981		− count++;
982		− }
983		− }
984		− start += sce0->ics.swb_sizes[g];
985		− }
986		− }
987		− cpe->is_mode = !!count;
988		−}
989		−
990	894	static void search_for_ms(AACEncContext s, ChannelElement cpe)
991	895	{
992	896	int start = 0, i, w, w2, g;
@@ -1000,7 +904,7 @@ static void search_for_ms(AACEncContext s, ChannelElement cpe) · 7 unmodified lines
1000	904	for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
1001	905	start = 0;
1002	906	for (g = 0; g < sce0->ics.num_swb; g++) {
1003		− if (!cpe->ch[0].zeroes[w16+g] && !cpe->ch[1].zeroes[w16+g] && !cpe->is_mask[w*16+g]) {
	907	+ if (!cpe->ch[0].zeroes[w16+g] && !cpe->ch[1].zeroes[w16+g]) {
1004	908	float dist1 = 0.0f, dist2 = 0.0f;
1005	909	for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
1006	910	FFPsyBand band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)16+g];

libavcodec/aacenc.c+5−3

@@ -459,7 +459,9 @@ static int aac_encode_frame(AVCodecContext avctx, AVPacket avpkt, · 458 unmodified lines
459	459	AACEncContext *s = avctx->priv_data;
460	460	float *samples = s->planar_samples, samples2, la, overlap;
461	461	ChannelElement *cpe;
462		− int i, ch, w, g, chans, tag, start_ch, ret, ms_mode = 0, is_mode = 0;
	462	+ SingleChannelElement *sce;
	463	+ int i, ch, w, g, chans, tag, start_ch, ret;
	464	+ int ms_mode = 0, is_mode = 0, tns_mode = 0, pred_mode = 0;
463	465	int chan_el_counter[4];
464	466	FFPsyWindowInfo windows[AAC_MAX_CHANNELS];
465	467
@@ -608,7 +610,7 @@ static int aac_encode_frame(AVCodecContext avctx, AVPacket avpkt, · 142 unmodified lines
608	610	s->coder->search_for_ms(s, cpe);
609	611	}
610	612	}
611		− if (chans > 1 && s->options.intensity_stereo && s->coder->search_for_is) {
	613	+ if (s->options.intensity_stereo && s->coder->search_for_is) {
612	614	s->coder->search_for_is(s, avctx, cpe);
613	615	if (cpe->is_mode) is_mode = 1;
614	616	}
@@ -636,7 +638,7 @@ static int aac_encode_frame(AVCodecContext avctx, AVPacket avpkt, · 21 unmodified lines
636	638	s->psy.bitres.bits = frame_bits / s->channels;
637	639	break;
638	640	}
639		− if (is_mode \|\| ms_mode) {
	641	+ if (is_mode \|\| ms_mode \|\| tns_mode \|\| pred_mode) {
640	642	for (i = 0; i < s->chan_map[0]; i++) {
641	643	// Must restore coeffs
642	644	chans = tag == TYPE_CPE ? 2 : 1;

libavcodec/aacenc_is.c+136−0

@@ -0,0 +1,136 @@
	1	+/*
	2	+ * AAC encoder intensity stereo
	3	+ * Copyright (C) 2015 Rostislav Pehlivanov
	4	+ *
	5	+ * This file is part of FFmpeg.
	6	+ *
	7	+ * FFmpeg is free software; you can redistribute it and/or
	8	+ * modify it under the terms of the GNU Lesser General Public
	9	+ * License as published by the Free Software Foundation; either
	10	+ * version 2.1 of the License, or (at your option) any later version.
	11	+ *
	12	+ * FFmpeg is distributed in the hope that it will be useful,
	13	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	14	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	15	+ * Lesser General Public License for more details.
	16	+ *
	17	+ * You should have received a copy of the GNU Lesser General Public
	18	+ * License along with FFmpeg; if not, write to the Free Software
	19	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	20	+ */
	21	+
	22	+/**
	23	+ * @file
	24	+ * AAC encoder Intensity Stereo
	25	+ * @author Rostislav Pehlivanov ( atomnuker gmail com )
	26	+ */
	27	+
	28	+#include "aacenc.h"
	29	+#include "aacenc_utils.h"
	30	+#include "aacenc_is.h"
	31	+#include "aacenc_quantization.h"
	32	+
	33	+struct is_error calc_is_encoding_err(AACEncContext s, ChannelElement cpe,
	34	+ int start, int w, int g, float ener0,
	35	+ float ener1, float ener01, int phase)
	36	+{
	37	+ int i, w2;
	38	+ float L34 = &s->scoefs[2560], R34 = &s->scoefs[2561];
	39	+ float IS = &s->scoefs[2562], I34 = &s->scoefs[2563];
	40	+ float dist1 = 0.0f, dist2 = 0.0f;
	41	+ struct is_error is_error = {0};
	42	+ SingleChannelElement *sce0 = &cpe->ch[0];
	43	+ SingleChannelElement *sce1 = &cpe->ch[1];
	44	+
	45	+ for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
	46	+ FFPsyBand band0 = &s->psy.ch[s->cur_channel+0].psy_bands[(w+w2)16+g];
	47	+ FFPsyBand band1 = &s->psy.ch[s->cur_channel+1].psy_bands[(w+w2)16+g];
	48	+ int is_band_type, is_sf_idx = FFMAX(1, sce0->sf_idx[(w+w2)*16+g]-4);
	49	+ float e01_34 = phase*pow(sqrt(ener1/ener0), 3.0/4.0);
	50	+ float maxval, dist_spec_err = 0.0f;
	51	+ float minthr = FFMIN(band0->threshold, band1->threshold);
	52	+ for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
	53	+ IS[i] = (sce0->pcoeffs[start+(w+w2)*128+i]+
	54	+ phasesce1->pcoeffs[start+(w+w2)128+i])*
	55	+ sqrt(ener0/ener01);
	56	+ }
	57	+ abs_pow34_v(L34, &sce0->coeffs[start+(w+w2)*128], sce0->ics.swb_sizes[g]);
	58	+ abs_pow34_v(R34, &sce1->coeffs[start+(w+w2)*128], sce0->ics.swb_sizes[g]);
	59	+ abs_pow34_v(I34, IS, sce0->ics.swb_sizes[g]);
	60	+ maxval = find_max_val(1, sce0->ics.swb_sizes[g], I34);
	61	+ is_band_type = find_min_book(maxval, is_sf_idx);
	62	+ dist1 += quantize_band_cost(s, &sce0->coeffs[start + (w+w2)*128], L34,
	63	+ sce0->ics.swb_sizes[g],
	64	+ sce0->sf_idx[(w+w2)*16+g],
	65	+ sce0->band_type[(w+w2)*16+g],
	66	+ s->lambda / band0->threshold, INFINITY, NULL, 0);
	67	+ dist1 += quantize_band_cost(s, &sce1->coeffs[start + (w+w2)*128], R34,
	68	+ sce1->ics.swb_sizes[g],
	69	+ sce1->sf_idx[(w+w2)*16+g],
	70	+ sce1->band_type[(w+w2)*16+g],
	71	+ s->lambda / band1->threshold, INFINITY, NULL, 0);
	72	+ dist2 += quantize_band_cost(s, IS, I34, sce0->ics.swb_sizes[g],
	73	+ is_sf_idx, is_band_type,
	74	+ s->lambda / minthr, INFINITY, NULL, 0);
	75	+ for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
	76	+ dist_spec_err += (L34[i] - I34[i])*(L34[i] - I34[i]);
	77	+ dist_spec_err += (R34[i] - I34[i]e01_34)(R34[i] - I34[i]*e01_34);
	78	+ }
	79	+ dist_spec_err *= s->lambda / minthr;
	80	+ dist2 += dist_spec_err;
	81	+ }
	82	+
	83	+ is_error.pass = dist2 <= dist1;
	84	+ is_error.phase = phase;
	85	+ is_error.error = fabsf(dist1 - dist2);
	86	+ is_error.dist1 = dist1;
	87	+ is_error.dist2 = dist2;
	88	+
	89	+ return is_error;
	90	+}
	91	+
	92	+void search_for_is(AACEncContext s, AVCodecContext avctx, ChannelElement *cpe)
	93	+{
	94	+ SingleChannelElement *sce0 = &cpe->ch[0];
	95	+ SingleChannelElement *sce1 = &cpe->ch[1];
	96	+ int start = 0, count = 0, w, w2, g, i;
	97	+ const float freq_mult = avctx->sample_rate/(1024.0f/sce0->ics.num_windows)/2.0f;
	98	+
	99	+ if (!cpe->common_window)
	100	+ return;
	101	+
	102	+ for (w = 0; w < sce0->ics.num_windows; w += sce0->ics.group_len[w]) {
	103	+ start = 0;
	104	+ for (g = 0; g < sce0->ics.num_swb; g++) {
	105	+ if (startfreq_mult > INT_STEREO_LOW_LIMIT(s->lambda/170.0f) &&
	106	+ cpe->ch[0].band_type[w16+g] != NOISE_BT && !cpe->ch[0].zeroes[w16+g] &&
	107	+ cpe->ch[1].band_type[w16+g] != NOISE_BT && !cpe->ch[1].zeroes[w16+g]) {
	108	+ float ener0 = 0.0f, ener1 = 0.0f, ener01 = 0.0f;
	109	+ struct is_error ph_err1, ph_err2, *erf;
	110	+ for (w2 = 0; w2 < sce0->ics.group_len[w]; w2++) {
	111	+ for (i = 0; i < sce0->ics.swb_sizes[g]; i++) {
	112	+ float coef0 = sce0->pcoeffs[start+(w+w2)*128+i];
	113	+ float coef1 = sce1->pcoeffs[start+(w+w2)*128+i];
	114	+ ener0 += coef0*coef0;
	115	+ ener1 += coef1*coef1;
	116	+ ener01 += (coef0 + coef1)*(coef0 + coef1);
	117	+ }
	118	+ }
	119	+ ph_err1 = calc_is_encoding_err(s, cpe, start, w, g,
	120	+ ener0, ener1, ener01, -1);
	121	+ ph_err2 = calc_is_encoding_err(s, cpe, start, w, g,
	122	+ ener0, ener1, ener01, +1);
	123	+ erf = ph_err1.error < ph_err2.error ? &ph_err1 : &ph_err2;
	124	+ if (erf->pass) {
	125	+ cpe->is_mask[w*16+g] = 1;
	126	+ cpe->ch[0].is_ener[w*16+g] = sqrt(ener0/ener01);
	127	+ cpe->ch[1].is_ener[w*16+g] = ener0/ener1;
	128	+ cpe->ch[1].band_type[w*16+g] = erf->phase ? INTENSITY_BT : INTENSITY_BT2;
	129	+ count++;
	130	+ }
	131	+ }
	132	+ start += sce0->ics.swb_sizes[g];
	133	+ }
	134	+ }
	135	+ cpe->is_mode = !!count;
	136	+}

libavcodec/aacenc_is.h+47−0

@@ -0,0 +1,47 @@
	1	+/*
	2	+ * AAC encoder intensity stereo
	3	+ * Copyright (C) 2015 Rostislav Pehlivanov
	4	+ *
	5	+ * This file is part of FFmpeg.
	6	+ *
	7	+ * FFmpeg is free software; you can redistribute it and/or
	8	+ * modify it under the terms of the GNU Lesser General Public
	9	+ * License as published by the Free Software Foundation; either
	10	+ * version 2.1 of the License, or (at your option) any later version.
	11	+ *
	12	+ * FFmpeg is distributed in the hope that it will be useful,
	13	+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
	14	+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
	15	+ * Lesser General Public License for more details.
	16	+ *
	17	+ * You should have received a copy of the GNU Lesser General Public
	18	+ * License along with FFmpeg; if not, write to the Free Software
	19	+ * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
	20	+ */
	21	+
	22	+/**
	23	+ * @file
	24	+ * AAC encoder Intensity Stereo
	25	+ * @author Rostislav Pehlivanov ( atomnuker gmail com )
	26	+ */
	27	+
	28	+#ifndef AVCODEC_AACENC_INTENSITY_STEREO_H
	29	+#define AVCODEC_AACENC_INTENSITY_STEREO_H
	30	+
	31	+/ Frequency in Hz for lower limit of intensity stereo /
	32	+#define INT_STEREO_LOW_LIMIT 6100
	33	+
	34	+struct is_error {
	35	+ int pass; /* 1 if dist2 <= dist1 */
	36	+ int phase; /* -1 or +1 */
	37	+ float error; /* fabs(dist1 - dist2) */
	38	+ float dist1; /* From original coeffs */
	39	+ float dist2; /* From IS'd coeffs */
	40	+};
	41	+
	42	+struct is_error calc_is_encoding_err(AACEncContext s, ChannelElement cpe,
	43	+ int start, int g, int w, float ener0,
	44	+ float ener1, float ener01, int phase);
	45	+void search_for_is(AACEncContext s, AVCodecContext avctx, ChannelElement *cpe);
	46	+
	47	+#endif /* AVCODEC_AACENC_INTENSITY_STEREO_H */

aaccoder: move the Intensity Stereo implementation out

Keyboard shortcuts

General

Issues and pull requests

Issues, pull requests and Inbox

Repository code