Skip to content

stereo: dynamic IS crossover and MS restriction at low bitrates #182

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
fabiangreffrath merged 1 commit into from
Aug 30, 2026
+39 −25
Merged

stereo: dynamic IS crossover and MS restriction at low bitrates #182

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 2 additions & 1 deletion libfaac/frame.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -812,7 +812,8 @@ int faacEncEncode(faacEncHandle hpEncoder,
ResetCoderSections(&coderInfo[channel]);

AACstereo(coderInfo, hEncoder->elements, hEncoder->numElements, hEncoder->freqBuff,
(float)hEncoder->aacquantCfg.quality/DEFQUAL, jointmode, hEncoder->sampleRate);
(float)hEncoder->aacquantCfg.quality/DEFQUAL, jointmode, hEncoder->sampleRate,
hEncoder->config.bandWidth);

/* AACstereo has already consumed freqBuff in place and BlocQuant
* accumulates into sf[] while reading book[], so a retry can re-run
Expand Down
57 changes: 35 additions & 22 deletions libfaac/stereo.c
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -21,25 +21,28 @@
#include "util.h"
#include "faac_internal.h"

/* Intensity stereo applies only at and above this frequency; below it the ear
* localizes from waveform detail, so panning the band would be audible. */
#define IS_START_FREQ_HZ 5500
/* Upper bound on the crossover, as a fraction of the sample rate (0.35*Fs =
* 0.7*Nyquist), so the band stays well inside the coded spectrum. */
/* Intensity stereo crossover scales with core bandwidth (3.5-7 kHz) to save low-band phase bits at low rates. */
#define IS_BW_RATIO 0.35f
#define IS_START_FREQ_MIN 3500
#define IS_START_FREQ_MAX 7000
/* Upper bound on crossover (0.35*Fs) so IS stays well inside coded spectrum. */
#define IS_FREQ_CAP_NUM 7
#define IS_FREQ_CAP_DEN 20
/* Pan, in SF_STEP_ENRG steps, beyond which the quieter channel is inaudible
* and is dropped to HCB_ZERO rather than intensity-coded. */
#define IS_PAN_LIMIT 30

/* Core bandwidth ceiling (16 kHz / <=48 kbps/ch) below which short-block M/S is disabled to prevent pre-echo. */
#define MS_SHORT_BW_CEILING 16000

/* Accumulate channel energies and cross-correlation for a scale factor band.
* Using three independent accumulators maximizes instruction-level parallelism
* by avoiding read-after-write dependencies on the FPU pipeline. */
static inline void calculate_energies(const float * restrict sl0, const float * restrict sr0,
int start, int len, int wstart, int wend,
float * restrict el_out, float * restrict er_out, float * restrict elr_out)
{
float el = 0, er = 0, elr = 0;
float el = 0.0f, er = 0.0f, elr = 0.0f;
int win, i;

for (win = wstart; win < wend; win++) {
Expand All @@ -62,9 +65,10 @@ static inline void calculate_energies(const float * restrict sl0, const float *
static inline void apply_mute(float * restrict s0, int start, int len, int wstart, int wend)
{
int win;
size_t bytes = (size_t)len * sizeof(float);
for (win = wstart; win < wend; win++) {
float * restrict s = s0 + win * BLOCK_LEN_SHORT + start;
memset(s, 0, len * sizeof(float));
memset(s, 0, bytes);
}
}

Expand All @@ -75,23 +79,23 @@ static inline void apply_ms(float * restrict sl0, float * restrict sr0,
int start, int len, int wstart, int wend, int in_phase)
{
int win, i;
size_t bytes = (size_t)len * sizeof(float);

if (in_phase) {
for (win = wstart; win < wend; win++) {
float * restrict sl = sl0 + win * BLOCK_LEN_SHORT + start;
float * restrict sr = sr0 + win * BLOCK_LEN_SHORT + start;
for (i = 0; i < len; i++) {
for (i = 0; i < len; i++)
sl[i] = 0.5f * (sl[i] + sr[i]);
sr[i] = 0.0f;
}
memset(sr, 0, bytes);
}
} else {
for (win = wstart; win < wend; win++) {
float * restrict sl = sl0 + win * BLOCK_LEN_SHORT + start;
float * restrict sr = sr0 + win * BLOCK_LEN_SHORT + start;
for (i = 0; i < len; i++) {
for (i = 0; i < len; i++)
sr[i] = 0.5f * (sl[i] - sr[i]);
sl[i] = 0.0f;
}
memset(sl, 0, bytes);
}
}
}
Expand Down Expand Up @@ -225,7 +229,7 @@ static inline int process_cpe(CoderInfo * restrict cl, CoderInfo * restrict cr,
}

void AACstereo(CoderInfo *coder, AACElement *elements, int numElements, float *s[MAX_CHANNELS],
float quality, int mode, int sampleRate)
float quality, int mode, int sampleRate, unsigned int bandWidth)
{
float inv_quality = 1.0f / quality;
float thrmid = 1.0f, isthr = 1.0f, thrside = 0.0f;
Expand Down Expand Up @@ -262,6 +266,14 @@ void AACstereo(CoderInfo *coder, AACElement *elements, int numElements, float *s
float inv_isthr = 1.0f / (isthr * isthr);
float thrside_sq = thrside * thrside;

/* Scale IS crossover with bandwidth (0.35 * bw, 3.5-7 kHz) to save phase bits. */
int cap = (sampleRate * IS_FREQ_CAP_NUM) / IS_FREQ_CAP_DEN;
int max_freq = min(IS_START_FREQ_MAX, cap);
int ifreq = (max_freq < IS_START_FREQ_MIN) ? max_freq : clamp_int((int)((float)bandWidth * IS_BW_RATIO), IS_START_FREQ_MIN, max_freq);

int target_offset_long = (ifreq * (2 * BLOCK_LEN_LONG) + sampleRate - 1) / sampleRate;
int target_offset_short = (ifreq * (2 * BLOCK_LEN_SHORT) + sampleRate - 1) / sampleRate;

for (int e = 0; e < numElements; e++) {
AACElement *elem = &elements[e];
if (elem->type != ID_CPE) continue;
Expand All @@ -283,16 +295,17 @@ void AACstereo(CoderInfo *coder, AACElement *elements, int numElements, float *s
}
if (!ok) continue;

/* Disable short-window M/S at <=16 kHz core bandwidth (<=48 kbps/ch) to prevent pre-echo. */
int cur_mode = (coder[lch].block_type == ONLY_SHORT_WINDOW && bandWidth <= MS_SHORT_BW_CEILING && mode == JOINT_MIXED) ? JOINT_IS : mode;

elem->common_window = true;
elem->msInfo.is_present = (mode == JOINT_MS);
elem->msInfo.is_present = (cur_mode == JOINT_MS);

int start = 0, sfcnt = 0, is_start_sfb = coder[lch].sfbn, msused = 0;
if (mode == JOINT_MIXED) {
int mlen = (coder[lch].block_type == ONLY_SHORT_WINDOW) ? 2*BLOCK_LEN_SHORT : 2*BLOCK_LEN_LONG;
int ifreq = IS_START_FREQ_HZ, cap = (sampleRate * IS_FREQ_CAP_NUM) / IS_FREQ_CAP_DEN;
if (ifreq > cap) ifreq = cap;
if (cur_mode == JOINT_MIXED) {
int target_offset = (coder[lch].block_type == ONLY_SHORT_WINDOW) ? target_offset_short : target_offset_long;
for (int sfb = 0; sfb < coder[lch].sfbn; sfb++) {
if ((coder[lch].sfb_offset[sfb] * sampleRate) / mlen >= ifreq) {
if (coder[lch].sfb_offset[sfb] >= target_offset) {
is_start_sfb = sfb; break;
}
}
Expand All @@ -302,9 +315,9 @@ void AACstereo(CoderInfo *coder, AACElement *elements, int numElements, float *s
int end = start + coder[lch].groups.len[g];
msused |= process_cpe(coder+lch, coder+rch, elem, s[lch], s[rch],
&sfcnt, start, end, thrmid, inv_isthr, thrside_sq,
is_start_sfb, mode);
is_start_sfb, cur_mode);
start = end;
}
if (mode == JOINT_MIXED && msused) elem->msInfo.is_present = true;
if (cur_mode == JOINT_MIXED && msused) elem->msInfo.is_present = true;
}
}
4 changes: 2 additions & 2 deletions libfaac/stereo.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -29,7 +29,7 @@ void AACstereo(CoderInfo *coder,
float *s[MAX_CHANNELS],
float quality,
int mode,
int sampleRate
);
int sampleRate,
unsigned int bandWidth);

#endif
Loading