[FFmpeg-devel] [PATCH v2] avcodec/aacdec: add support for AAC SBR with 960 frame length
Matthias Walliczek
matthias at walliczek.de
Sat May 20 23:17:24 EEST 2023
Supports the european DAB+ digital radio coding
Fixes: https://trac.ffmpeg.org/ticket/1407 "HE-AAC (v2): 960/120 MDCT
window is not implemented"
Co-developed-by: Lynne <dev at lynne.ee>
Signed-off-by: Matthias Walliczek <matthias at walliczek.de>
---
libavcodec/aacdec_template.c | 17 ++------------
libavcodec/aacps.c | 8 ++++---
libavcodec/aacps.h | 7 +++---
libavcodec/aacps_common.c | 3 ++-
libavcodec/aacsbr_template.c | 43 +++++++++++++++++++++---------------
libavcodec/sbr.h | 2 +-
6 files changed, 38 insertions(+), 42 deletions(-)
diff --git a/libavcodec/aacdec_template.c b/libavcodec/aacdec_template.c
index 444dc4fa9d..72446537f1 100644
--- a/libavcodec/aacdec_template.c
+++ b/libavcodec/aacdec_template.c
@@ -849,12 +849,6 @@ static int decode_ga_specific_config(AACContext
*ac, AVCodecContext *avctx,
int tags = 0;
m4ac->frame_length_short = get_bits1(gb);
- if (m4ac->frame_length_short && m4ac->sbr == 1) {
- avpriv_report_missing_feature(avctx, "SBR with 960 frame length");
- if (ac) ac->warned_960_sbr = 1;
- m4ac->sbr = 0;
- m4ac->ps = 0;
- }
if (get_bits1(gb)) // dependsOnCoreCoder
skip_bits(gb, 14); // coreCoderDelay
@@ -1047,10 +1041,10 @@ static int
decode_audio_specific_config_gb(AACContext *ac,
}
ff_dlog(avctx,
- "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d\n",
+ "AOT %d chan config %d sampling index %d (%d) SBR %d PS %d
frame length %d\n",
m4ac->object_type, m4ac->chan_config, m4ac->sampling_index,
m4ac->sample_rate, m4ac->sbr,
- m4ac->ps);
+ m4ac->ps, m4ac->frame_length_short ? 960 : 1024);
return get_bits_count(gb);
}
@@ -2467,13 +2461,6 @@ static int decode_extension_payload(AACContext
*ac, GetBitContext *gb, int cnt,
if (!che) {
av_log(ac->avctx, AV_LOG_ERROR, "SBR was found before the
first channel element.\n");
return res;
- } else if (ac->oc[1].m4ac.frame_length_short) {
- if (!ac->warned_960_sbr)
- avpriv_report_missing_feature(ac->avctx,
- "SBR with 960 frame length");
- ac->warned_960_sbr = 1;
- skip_bits_long(gb, 8 * cnt - 4);
- return res;
} else if (!ac->oc[1].m4ac.sbr) {
av_log(ac->avctx, AV_LOG_ERROR, "SBR signaled to be
not-present but was found in the bitstream.\n");
skip_bits_long(gb, 8 * cnt - 4);
diff --git a/libavcodec/aacps.c b/libavcodec/aacps.c
index 655e8fe5b4..4804722e59 100644
--- a/libavcodec/aacps.c
+++ b/libavcodec/aacps.c
@@ -397,7 +397,7 @@ static void map_val_20_to_34(INTFLOAT
par[PS_MAX_NR_IIDICC])
par[ 1] = AAC_HALF_SUM(par[ 0], par[ 1]);
}
-static void decorrelation(PSContext *ps, INTFLOAT (*out)[32][2],
const INTFLOAT (*s)[32][2], int is34)
+static void decorrelation(PSContext *ps, INTFLOAT (*out)[32][2], const
INTFLOAT (*s)[32][2], int is34, int frame_length_short)
{
LOCAL_ALIGNED_16(INTFLOAT, power, [34], [PS_QMF_TIME_SLOTS]);
LOCAL_ALIGNED_16(INTFLOAT, transient_gain, [34], [PS_QMF_TIME_SLOTS]);
@@ -414,6 +414,7 @@ static void decorrelation(PSContext *ps, INTFLOAT
(*out)[32][2], const INTFLOAT
int i, k, m, n;
int n0 = 0, nL = 32;
const INTFLOAT peak_decay_factor = Q31(0.76592833836465f);
+ const int numQMFSlots = frame_length_short ? 30 : 32;
memset(power, 0, 34 * sizeof(*power));
@@ -717,7 +718,8 @@ static void stereo_processing(PSContext *ps,
INTFLOAT (*l)[32][2], INTFLOAT (*r)
}
}
-int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps,
INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top)
+int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps,
INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top,
+ int frame_length_short)
{
INTFLOAT (*Lbuf)[32][2] = ps->Lbuf;
INTFLOAT (*Rbuf)[32][2] = ps->Rbuf;
@@ -730,7 +732,7 @@ int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx,
PSContext *ps, INTFLOAT L[2][
memset(ps->ap_delay + top, 0, (NR_ALLPASS_BANDS[is34] -
top)*sizeof(ps->ap_delay[0]));
hybrid_analysis(&ps->dsp, Lbuf, ps->in_buf, L, is34, len);
- decorrelation(ps, Rbuf, (const INTFLOAT (*)[32][2]) Lbuf, is34);
+ decorrelation(ps, Rbuf, (const INTFLOAT (*)[32][2]) Lbuf, is34,
frame_length_short);
stereo_processing(ps, Lbuf, Rbuf, is34);
hybrid_synthesis(&ps->dsp, L, Lbuf, is34, len);
hybrid_synthesis(&ps->dsp, R, Rbuf, is34, len);
diff --git a/libavcodec/aacps.h b/libavcodec/aacps.h
index 3efa38ad88..c771bfb270 100644
--- a/libavcodec/aacps.h
+++ b/libavcodec/aacps.h
@@ -43,8 +43,6 @@
///< Baseline implies 10 or 20 stereo bands,
///< mixing mode A, and no ipd/opd
-#define numQMFSlots 32 //numTimeSlots * RATE
-
typedef struct PSCommonContext {
int start;
int enable_iid;
@@ -96,7 +94,8 @@ void ff_ps_init_common(void);
void AAC_RENAME(ff_ps_init)(void);
void AAC_RENAME(ff_ps_ctx_init)(PSContext *ps);
int ff_ps_read_data(AVCodecContext *avctx, GetBitContext *gb,
- PSCommonContext *ps, int bits_left);
-int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps,
INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top);
+ PSCommonContext *ps, int bits_left, int
frame_length_short);
+int AAC_RENAME(ff_ps_apply)(AVCodecContext *avctx, PSContext *ps,
INTFLOAT L[2][38][64], INTFLOAT R[2][38][64], int top,
+ int frame_length_short);
#endif /* AVCODEC_AACPS_H */
diff --git a/libavcodec/aacps_common.c b/libavcodec/aacps_common.c
index c388d5b9bc..87521c52e3 100644
--- a/libavcodec/aacps_common.c
+++ b/libavcodec/aacps_common.c
@@ -132,13 +132,14 @@ static int ps_read_extension_data(GetBitContext
*gb, PSCommonContext *ps,
}
int ff_ps_read_data(AVCodecContext *avctx, GetBitContext *gb_host,
- PSCommonContext *ps, int bits_left)
+ PSCommonContext *ps, int bits_left, int
frame_length_short)
{
int e;
int bit_count_start = get_bits_count(gb_host);
int header;
int bits_consumed;
GetBitContext gbc = *gb_host, *gb = &gbc;
+ int numQMFSlots = frame_length_short ? 30 : 32;
header = get_bits1(gb);
if (header) { //enable_ps_header
diff --git a/libavcodec/aacsbr_template.c b/libavcodec/aacsbr_template.c
index cdca402f04..fc64b1de4a 100644
--- a/libavcodec/aacsbr_template.c
+++ b/libavcodec/aacsbr_template.c
@@ -647,8 +647,7 @@ static int read_sbr_grid(AACContext *ac,
SpectralBandReplication *sbr,
{
int i;
int bs_pointer = 0;
- // frameLengthFlag ? 15 : 16; 960 sample length frames unsupported;
this value is numTimeSlots
- int abs_bord_trail = 16;
+ int abs_bord_trail = ac->oc[1].m4ac.frame_length_short ? 15 : 16;
int num_rel_lead, num_rel_trail;
unsigned bs_num_env_old = ch_data->bs_num_env;
int bs_frame_class, bs_num_env;
@@ -972,7 +971,8 @@ static void read_sbr_extension(AACContext *ac,
SpectralBandReplication *sbr,
skip_bits_long(gb, *num_bits_left); // bs_fill_bits
*num_bits_left = 0;
} else {
- *num_bits_left -= ff_ps_read_data(ac->avctx, gb,
&sbr->ps.common, *num_bits_left);
+ *num_bits_left -= ff_ps_read_data(ac->avctx, gb,
&sbr->ps.common, *num_bits_left, +
ac->oc[1].m4ac.frame_length_short);
ac->avctx->profile = FF_PROFILE_AAC_HE_V2;
// ensure the warning is not printed if PS extension is
present
ac->warned_he_aac_mono = 1;
@@ -1190,16 +1190,20 @@ static void sbr_qmf_analysis(AVFloatDSPContext
*dsp, AVTXContext *mdct,
av_tx_fn mdct_fn,
#endif /* USE_FIXED */
SBRDSPContext *sbrdsp, const INTFLOAT
*in, INTFLOAT *x,
- INTFLOAT z[320], INTFLOAT W[2][32][32][2],
int buf_idx)
+ INTFLOAT z[320], INTFLOAT W[2][32][32][2],
int buf_idx,
+ int frame_length_short)
{
int i;
+ const int frameLength = frame_length_short ? 960 : 1024;
+ const int lp = frame_length_short ? 15*2 : 16*2;
#if USE_FIXED
int j;
#endif
- memcpy(x , x+1024, (320-32)*sizeof(x[0]));
- memcpy(x+288, in, 1024*sizeof(x[0]));
- for (i = 0; i < 32; i++) { // numTimeSlots*RATE = 16*2 as 960
sample frames
- // are not supported
+ + memcpy(x , x+frameLength, (320-32)*sizeof(x[0]));
+ memcpy(x+288, in, frameLength*sizeof(x[0]));
+ + for (i = 0; i < lp; i++) {
dsp->vector_fmul_reverse(z, sbr_qmf_window_ds, x, 320);
sbrdsp->sum64x5(z);
sbrdsp->qmf_pre_shuffle(z);
@@ -1238,13 +1242,15 @@ static void sbr_qmf_synthesis(AVTXContext *mdct,
av_tx_fn mdct_fn,
#endif /* USE_FIXED */
INTFLOAT *out, INTFLOAT X[2][38][64],
INTFLOAT mdct_buf[2][64],
- INTFLOAT *v0, int *v_off, const unsigned
int div)
+ INTFLOAT *v0, int *v_off, const unsigned
int div,
+ int frame_length_short)
{
int i, n;
const INTFLOAT *sbr_qmf_window = div ? sbr_qmf_window_ds :
sbr_qmf_window_us;
const int step = 128 >> div;
+ const int lp = frame_length_short ? 15*2 : 16*2;
INTFLOAT *v;
- for (i = 0; i < 32; i++) {
+ for (i = 0; i < lp; i++) {
if (*v_off < step) {
int saved_samples = (1280 - 128) >> div;
memcpy(&v0[SBR_SYNTHESIS_BUF_SIZE - saved_samples], v0,
saved_samples * sizeof(INTFLOAT));
@@ -1288,7 +1294,7 @@ static int sbr_lf_gen(AACContext *ac,
SpectralBandReplication *sbr,
{
int i, k;
const int t_HFGen = 8;
- const int i_f = 32;
+ const int i_f = ac->oc[1].m4ac.frame_length_short ? 15*2 : 16*2;
memset(X_low, 0, 32*sizeof(*X_low));
for (k = 0; k < sbr->kx[1]; k++) {
for (i = t_HFGen; i < i_f + t_HFGen; i++) {
@@ -1344,10 +1350,10 @@ static int sbr_hf_gen(AACContext *ac,
SpectralBandReplication *sbr,
/// Generate the subband filtered lowband
static int sbr_x_gen(SpectralBandReplication *sbr, INTFLOAT X[2][38][64],
const INTFLOAT Y0[38][64][2], const INTFLOAT
Y1[38][64][2],
- const INTFLOAT X_low[32][40][2], int ch)
+ const INTFLOAT X_low[32][40][2], int ch, int
frame_length_short)
{
int k, i;
- const int i_f = 32;
+ const int i_f = frame_length_short ? 15*2 : 16*2;
const int i_Temp = FFMAX(2*sbr->data[ch].t_env_num_env_old - i_f, 0);
memset(X, 0, 2*sizeof(*X));
for (k = 0; k < sbr->kx[0]; k++) {
@@ -1531,7 +1537,7 @@ void AAC_RENAME(ff_sbr_apply)(AACContext *ac,
SpectralBandReplication *sbr, int
sbr_qmf_analysis(ac->fdsp, sbr->mdct_ana, sbr->mdct_ana_fn,
&sbr->dsp,
ch ? R : L,
sbr->data[ch].analysis_filterbank_samples,
(INTFLOAT*)sbr->qmf_filter_scratch,
- sbr->data[ch].W, sbr->data[ch].Ypos);
+ sbr->data[ch].W, sbr->data[ch].Ypos,
ac->oc[1].m4ac.frame_length_short);
sbr->c.sbr_lf_gen(ac, sbr, sbr->X_low,
(const INTFLOAT (*)[32][32][2]) sbr->data[ch].W,
sbr->data[ch].Ypos);
@@ -1564,12 +1570,13 @@ void AAC_RENAME(ff_sbr_apply)(AACContext *ac,
SpectralBandReplication *sbr, int
sbr->c.sbr_x_gen(sbr, sbr->X[ch],
(const INTFLOAT (*)[64][2])
sbr->data[ch].Y[1-sbr->data[ch].Ypos],
(const INTFLOAT (*)[64][2]) sbr->data[ch].Y[
sbr->data[ch].Ypos],
- (const INTFLOAT (*)[40][2]) sbr->X_low, ch);
+ (const INTFLOAT (*)[40][2]) sbr->X_low, ch,
ac->oc[1].m4ac.frame_length_short);
}
if (ac->oc[1].m4ac.ps == 1) {
if (sbr->ps.common.start) {
- AAC_RENAME(ff_ps_apply)(ac->avctx, &sbr->ps, sbr->X[0],
sbr->X[1], sbr->kx[1] + sbr->m[1]);
+ AAC_RENAME(ff_ps_apply)(ac->avctx, &sbr->ps, sbr->X[0],
sbr->X[1], sbr->kx[1] + sbr->m[1], +
ac->oc[1].m4ac.frame_length_short);
} else {
memcpy(sbr->X[1], sbr->X[0], sizeof(sbr->X[0]));
}
@@ -1580,13 +1587,13 @@ void AAC_RENAME(ff_sbr_apply)(AACContext *ac,
SpectralBandReplication *sbr, int
L, sbr->X[0], sbr->qmf_filter_scratch,
sbr->data[0].synthesis_filterbank_samples,
&sbr->data[0].synthesis_filterbank_samples_offset,
- downsampled);
+ downsampled, ac->oc[1].m4ac.frame_length_short);
if (nch == 2)
sbr_qmf_synthesis(sbr->mdct, sbr->mdct_fn, &sbr->dsp, ac->fdsp,
R, sbr->X[1], sbr->qmf_filter_scratch,
sbr->data[1].synthesis_filterbank_samples,
&sbr->data[1].synthesis_filterbank_samples_offset,
- downsampled);
+ downsampled, ac->oc[1].m4ac.frame_length_short);
}
static void aacsbr_func_ptr_init(AACSBRContext *c)
diff --git a/libavcodec/sbr.h b/libavcodec/sbr.h
index f949465ef5..5a03b1597f 100644
--- a/libavcodec/sbr.h
+++ b/libavcodec/sbr.h
@@ -130,7 +130,7 @@ typedef struct AACSBRContext {
const int e_a[2]);
int (*sbr_x_gen)(SpectralBandReplication *sbr, INTFLOAT X[2][38][64],
const INTFLOAT Y0[38][64][2], const INTFLOAT
Y1[38][64][2],
- const INTFLOAT X_low[32][40][2], int ch);
+ const INTFLOAT X_low[32][40][2], int ch, int
frame_length_short);
void (*sbr_hf_inverse_filter)(SBRDSPContext *dsp,
INTFLOAT (*alpha0)[2], INTFLOAT
(*alpha1)[2],
const INTFLOAT X_low[32][40][2], int
k0);
--
2.36.0.windows.1
More information about the ffmpeg-devel
mailing list