mirror of
https://github.com/xenia-project/FFmpeg.git
synced 2024-11-28 14:01:27 +00:00
c8af852b97
This is a new library for audio sample format, channel layout, and sample rate conversion.
481 lines
16 KiB
C
481 lines
16 KiB
C
/*
|
|
* Copyright (c) 2004 Michael Niedermayer <michaelni@gmx.at>
|
|
* Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
|
|
*
|
|
* This file is part of Libav.
|
|
*
|
|
* Libav 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.
|
|
*
|
|
* Libav 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 Libav; if not, write to the Free Software
|
|
* Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
|
|
*/
|
|
|
|
#include "libavutil/libm.h"
|
|
#include "libavutil/log.h"
|
|
#include "internal.h"
|
|
#include "audio_data.h"
|
|
|
|
#ifdef CONFIG_RESAMPLE_FLT
|
|
/* float template */
|
|
#define FILTER_SHIFT 0
|
|
#define FELEM float
|
|
#define FELEM2 float
|
|
#define FELEML float
|
|
#define WINDOW_TYPE 24
|
|
#elifdef CONFIG_RESAMPLE_S32
|
|
/* s32 template */
|
|
#define FILTER_SHIFT 30
|
|
#define FELEM int32_t
|
|
#define FELEM2 int64_t
|
|
#define FELEML int64_t
|
|
#define FELEM_MAX INT32_MAX
|
|
#define FELEM_MIN INT32_MIN
|
|
#define WINDOW_TYPE 12
|
|
#else
|
|
/* s16 template */
|
|
#define FILTER_SHIFT 15
|
|
#define FELEM int16_t
|
|
#define FELEM2 int32_t
|
|
#define FELEML int64_t
|
|
#define FELEM_MAX INT16_MAX
|
|
#define FELEM_MIN INT16_MIN
|
|
#define WINDOW_TYPE 9
|
|
#endif
|
|
|
|
struct ResampleContext {
|
|
AVAudioResampleContext *avr;
|
|
AudioData *buffer;
|
|
FELEM *filter_bank;
|
|
int filter_length;
|
|
int ideal_dst_incr;
|
|
int dst_incr;
|
|
int index;
|
|
int frac;
|
|
int src_incr;
|
|
int compensation_distance;
|
|
int phase_shift;
|
|
int phase_mask;
|
|
int linear;
|
|
double factor;
|
|
};
|
|
|
|
/**
|
|
* 0th order modified bessel function of the first kind.
|
|
*/
|
|
static double bessel(double x)
|
|
{
|
|
double v = 1;
|
|
double lastv = 0;
|
|
double t = 1;
|
|
int i;
|
|
|
|
x = x * x / 4;
|
|
for (i = 1; v != lastv; i++) {
|
|
lastv = v;
|
|
t *= x / (i * i);
|
|
v += t;
|
|
}
|
|
return v;
|
|
}
|
|
|
|
/**
|
|
* Build a polyphase filterbank.
|
|
*
|
|
* @param[out] filter filter coefficients
|
|
* @param factor resampling factor
|
|
* @param tap_count tap count
|
|
* @param phase_count phase count
|
|
* @param scale wanted sum of coefficients for each filter
|
|
* @param type 0->cubic
|
|
* 1->blackman nuttall windowed sinc
|
|
* 2..16->kaiser windowed sinc beta=2..16
|
|
* @return 0 on success, negative AVERROR code on failure
|
|
*/
|
|
static int build_filter(FELEM *filter, double factor, int tap_count,
|
|
int phase_count, int scale, int type)
|
|
{
|
|
int ph, i;
|
|
double x, y, w;
|
|
double *tab;
|
|
const int center = (tap_count - 1) / 2;
|
|
|
|
tab = av_malloc(tap_count * sizeof(*tab));
|
|
if (!tab)
|
|
return AVERROR(ENOMEM);
|
|
|
|
/* if upsampling, only need to interpolate, no filter */
|
|
if (factor > 1.0)
|
|
factor = 1.0;
|
|
|
|
for (ph = 0; ph < phase_count; ph++) {
|
|
double norm = 0;
|
|
for (i = 0; i < tap_count; i++) {
|
|
x = M_PI * ((double)(i - center) - (double)ph / phase_count) * factor;
|
|
if (x == 0) y = 1.0;
|
|
else y = sin(x) / x;
|
|
switch (type) {
|
|
case 0: {
|
|
const float d = -0.5; //first order derivative = -0.5
|
|
x = fabs(((double)(i - center) - (double)ph / phase_count) * factor);
|
|
if (x < 1.0) y = 1 - 3 * x*x + 2 * x*x*x + d * ( -x*x + x*x*x);
|
|
else y = d * (-4 + 8 * x - 5 * x*x + x*x*x);
|
|
break;
|
|
}
|
|
case 1:
|
|
w = 2.0 * x / (factor * tap_count) + M_PI;
|
|
y *= 0.3635819 - 0.4891775 * cos( w) +
|
|
0.1365995 * cos(2 * w) -
|
|
0.0106411 * cos(3 * w);
|
|
break;
|
|
default:
|
|
w = 2.0 * x / (factor * tap_count * M_PI);
|
|
y *= bessel(type * sqrt(FFMAX(1 - w * w, 0)));
|
|
break;
|
|
}
|
|
|
|
tab[i] = y;
|
|
norm += y;
|
|
}
|
|
|
|
/* normalize so that an uniform color remains the same */
|
|
for (i = 0; i < tap_count; i++) {
|
|
#ifdef CONFIG_RESAMPLE_FLT
|
|
filter[ph * tap_count + i] = tab[i] / norm;
|
|
#else
|
|
filter[ph * tap_count + i] = av_clip(lrintf(tab[i] * scale / norm),
|
|
FELEM_MIN, FELEM_MAX);
|
|
#endif
|
|
}
|
|
}
|
|
|
|
av_free(tab);
|
|
return 0;
|
|
}
|
|
|
|
ResampleContext *ff_audio_resample_init(AVAudioResampleContext *avr)
|
|
{
|
|
ResampleContext *c;
|
|
int out_rate = avr->out_sample_rate;
|
|
int in_rate = avr->in_sample_rate;
|
|
double factor = FFMIN(out_rate * avr->cutoff / in_rate, 1.0);
|
|
int phase_count = 1 << avr->phase_shift;
|
|
|
|
/* TODO: add support for s32 and float internal formats */
|
|
if (avr->internal_sample_fmt != AV_SAMPLE_FMT_S16P) {
|
|
av_log(avr, AV_LOG_ERROR, "Unsupported internal format for "
|
|
"resampling: %s\n",
|
|
av_get_sample_fmt_name(avr->internal_sample_fmt));
|
|
return NULL;
|
|
}
|
|
c = av_mallocz(sizeof(*c));
|
|
if (!c)
|
|
return NULL;
|
|
|
|
c->avr = avr;
|
|
c->phase_shift = avr->phase_shift;
|
|
c->phase_mask = phase_count - 1;
|
|
c->linear = avr->linear_interp;
|
|
c->factor = factor;
|
|
c->filter_length = FFMAX((int)ceil(avr->filter_size / factor), 1);
|
|
|
|
c->filter_bank = av_mallocz(c->filter_length * (phase_count + 1) * sizeof(FELEM));
|
|
if (!c->filter_bank)
|
|
goto error;
|
|
|
|
if (build_filter(c->filter_bank, factor, c->filter_length, phase_count,
|
|
1 << FILTER_SHIFT, WINDOW_TYPE) < 0)
|
|
goto error;
|
|
|
|
memcpy(&c->filter_bank[c->filter_length * phase_count + 1],
|
|
c->filter_bank, (c->filter_length - 1) * sizeof(FELEM));
|
|
c->filter_bank[c->filter_length * phase_count] = c->filter_bank[c->filter_length - 1];
|
|
|
|
c->compensation_distance = 0;
|
|
if (!av_reduce(&c->src_incr, &c->dst_incr, out_rate,
|
|
in_rate * (int64_t)phase_count, INT32_MAX / 2))
|
|
goto error;
|
|
c->ideal_dst_incr = c->dst_incr;
|
|
|
|
c->index = -phase_count * ((c->filter_length - 1) / 2);
|
|
c->frac = 0;
|
|
|
|
/* allocate internal buffer */
|
|
c->buffer = ff_audio_data_alloc(avr->resample_channels, 0,
|
|
avr->internal_sample_fmt,
|
|
"resample buffer");
|
|
if (!c->buffer)
|
|
goto error;
|
|
|
|
av_log(avr, AV_LOG_DEBUG, "resample: %s from %d Hz to %d Hz\n",
|
|
av_get_sample_fmt_name(avr->internal_sample_fmt),
|
|
avr->in_sample_rate, avr->out_sample_rate);
|
|
|
|
return c;
|
|
|
|
error:
|
|
ff_audio_data_free(&c->buffer);
|
|
av_free(c->filter_bank);
|
|
av_free(c);
|
|
return NULL;
|
|
}
|
|
|
|
void ff_audio_resample_free(ResampleContext **c)
|
|
{
|
|
if (!*c)
|
|
return;
|
|
ff_audio_data_free(&(*c)->buffer);
|
|
av_free((*c)->filter_bank);
|
|
av_freep(c);
|
|
}
|
|
|
|
int avresample_set_compensation(AVAudioResampleContext *avr, int sample_delta,
|
|
int compensation_distance)
|
|
{
|
|
ResampleContext *c;
|
|
AudioData *fifo_buf = NULL;
|
|
int ret = 0;
|
|
|
|
if (compensation_distance < 0)
|
|
return AVERROR(EINVAL);
|
|
if (!compensation_distance && sample_delta)
|
|
return AVERROR(EINVAL);
|
|
|
|
/* if resampling was not enabled previously, re-initialize the
|
|
AVAudioResampleContext and force resampling */
|
|
if (!avr->resample_needed) {
|
|
int fifo_samples;
|
|
double matrix[AVRESAMPLE_MAX_CHANNELS * AVRESAMPLE_MAX_CHANNELS] = { 0 };
|
|
|
|
/* buffer any remaining samples in the output FIFO before closing */
|
|
fifo_samples = av_audio_fifo_size(avr->out_fifo);
|
|
if (fifo_samples > 0) {
|
|
fifo_buf = ff_audio_data_alloc(avr->out_channels, fifo_samples,
|
|
avr->out_sample_fmt, NULL);
|
|
if (!fifo_buf)
|
|
return AVERROR(EINVAL);
|
|
ret = ff_audio_data_read_from_fifo(avr->out_fifo, fifo_buf,
|
|
fifo_samples);
|
|
if (ret < 0)
|
|
goto reinit_fail;
|
|
}
|
|
/* save the channel mixing matrix */
|
|
ret = avresample_get_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS);
|
|
if (ret < 0)
|
|
goto reinit_fail;
|
|
|
|
/* close the AVAudioResampleContext */
|
|
avresample_close(avr);
|
|
|
|
avr->force_resampling = 1;
|
|
|
|
/* restore the channel mixing matrix */
|
|
ret = avresample_set_matrix(avr, matrix, AVRESAMPLE_MAX_CHANNELS);
|
|
if (ret < 0)
|
|
goto reinit_fail;
|
|
|
|
/* re-open the AVAudioResampleContext */
|
|
ret = avresample_open(avr);
|
|
if (ret < 0)
|
|
goto reinit_fail;
|
|
|
|
/* restore buffered samples to the output FIFO */
|
|
if (fifo_samples > 0) {
|
|
ret = ff_audio_data_add_to_fifo(avr->out_fifo, fifo_buf, 0,
|
|
fifo_samples);
|
|
if (ret < 0)
|
|
goto reinit_fail;
|
|
ff_audio_data_free(&fifo_buf);
|
|
}
|
|
}
|
|
c = avr->resample;
|
|
c->compensation_distance = compensation_distance;
|
|
if (compensation_distance) {
|
|
c->dst_incr = c->ideal_dst_incr - c->ideal_dst_incr *
|
|
(int64_t)sample_delta / compensation_distance;
|
|
} else {
|
|
c->dst_incr = c->ideal_dst_incr;
|
|
}
|
|
return 0;
|
|
|
|
reinit_fail:
|
|
ff_audio_data_free(&fifo_buf);
|
|
return ret;
|
|
}
|
|
|
|
static int resample(ResampleContext *c, int16_t *dst, const int16_t *src,
|
|
int *consumed, int src_size, int dst_size, int update_ctx)
|
|
{
|
|
int dst_index, i;
|
|
int index = c->index;
|
|
int frac = c->frac;
|
|
int dst_incr_frac = c->dst_incr % c->src_incr;
|
|
int dst_incr = c->dst_incr / c->src_incr;
|
|
int compensation_distance = c->compensation_distance;
|
|
|
|
if (!dst != !src)
|
|
return AVERROR(EINVAL);
|
|
|
|
if (compensation_distance == 0 && c->filter_length == 1 &&
|
|
c->phase_shift == 0) {
|
|
int64_t index2 = ((int64_t)index) << 32;
|
|
int64_t incr = (1LL << 32) * c->dst_incr / c->src_incr;
|
|
dst_size = FFMIN(dst_size,
|
|
(src_size-1-index) * (int64_t)c->src_incr /
|
|
c->dst_incr);
|
|
|
|
if (dst) {
|
|
for(dst_index = 0; dst_index < dst_size; dst_index++) {
|
|
dst[dst_index] = src[index2 >> 32];
|
|
index2 += incr;
|
|
}
|
|
} else {
|
|
dst_index = dst_size;
|
|
}
|
|
index += dst_index * dst_incr;
|
|
index += (frac + dst_index * (int64_t)dst_incr_frac) / c->src_incr;
|
|
frac = (frac + dst_index * (int64_t)dst_incr_frac) % c->src_incr;
|
|
} else {
|
|
for (dst_index = 0; dst_index < dst_size; dst_index++) {
|
|
FELEM *filter = c->filter_bank +
|
|
c->filter_length * (index & c->phase_mask);
|
|
int sample_index = index >> c->phase_shift;
|
|
|
|
if (!dst && (sample_index + c->filter_length > src_size ||
|
|
-sample_index >= src_size))
|
|
break;
|
|
|
|
if (dst) {
|
|
FELEM2 val = 0;
|
|
|
|
if (sample_index < 0) {
|
|
for (i = 0; i < c->filter_length; i++)
|
|
val += src[FFABS(sample_index + i) % src_size] *
|
|
(FELEM2)filter[i];
|
|
} else if (sample_index + c->filter_length > src_size) {
|
|
break;
|
|
} else if (c->linear) {
|
|
FELEM2 v2 = 0;
|
|
for (i = 0; i < c->filter_length; i++) {
|
|
val += src[abs(sample_index + i)] * (FELEM2)filter[i];
|
|
v2 += src[abs(sample_index + i)] * (FELEM2)filter[i + c->filter_length];
|
|
}
|
|
val += (v2 - val) * (FELEML)frac / c->src_incr;
|
|
} else {
|
|
for (i = 0; i < c->filter_length; i++)
|
|
val += src[sample_index + i] * (FELEM2)filter[i];
|
|
}
|
|
|
|
#ifdef CONFIG_RESAMPLE_FLT
|
|
dst[dst_index] = av_clip_int16(lrintf(val));
|
|
#else
|
|
val = (val + (1<<(FILTER_SHIFT-1)))>>FILTER_SHIFT;
|
|
dst[dst_index] = av_clip_int16(val);
|
|
#endif
|
|
}
|
|
|
|
frac += dst_incr_frac;
|
|
index += dst_incr;
|
|
if (frac >= c->src_incr) {
|
|
frac -= c->src_incr;
|
|
index++;
|
|
}
|
|
if (dst_index + 1 == compensation_distance) {
|
|
compensation_distance = 0;
|
|
dst_incr_frac = c->ideal_dst_incr % c->src_incr;
|
|
dst_incr = c->ideal_dst_incr / c->src_incr;
|
|
}
|
|
}
|
|
}
|
|
if (consumed)
|
|
*consumed = FFMAX(index, 0) >> c->phase_shift;
|
|
|
|
if (update_ctx) {
|
|
if (index >= 0)
|
|
index &= c->phase_mask;
|
|
|
|
if (compensation_distance) {
|
|
compensation_distance -= dst_index;
|
|
if (compensation_distance <= 0)
|
|
return AVERROR_BUG;
|
|
}
|
|
c->frac = frac;
|
|
c->index = index;
|
|
c->dst_incr = dst_incr_frac + c->src_incr*dst_incr;
|
|
c->compensation_distance = compensation_distance;
|
|
}
|
|
|
|
return dst_index;
|
|
}
|
|
|
|
int ff_audio_resample(ResampleContext *c, AudioData *dst, AudioData *src,
|
|
int *consumed)
|
|
{
|
|
int ch, in_samples, in_leftover, out_samples = 0;
|
|
int ret = AVERROR(EINVAL);
|
|
|
|
in_samples = src ? src->nb_samples : 0;
|
|
in_leftover = c->buffer->nb_samples;
|
|
|
|
/* add input samples to the internal buffer */
|
|
if (src) {
|
|
ret = ff_audio_data_combine(c->buffer, in_leftover, src, 0, in_samples);
|
|
if (ret < 0)
|
|
return ret;
|
|
} else if (!in_leftover) {
|
|
/* no remaining samples to flush */
|
|
return 0;
|
|
} else {
|
|
/* TODO: pad buffer to flush completely */
|
|
}
|
|
|
|
/* calculate output size and reallocate output buffer if needed */
|
|
/* TODO: try to calculate this without the dummy resample() run */
|
|
if (!dst->read_only && dst->allow_realloc) {
|
|
out_samples = resample(c, NULL, NULL, NULL, c->buffer->nb_samples,
|
|
INT_MAX, 0);
|
|
ret = ff_audio_data_realloc(dst, out_samples);
|
|
if (ret < 0) {
|
|
av_log(c->avr, AV_LOG_ERROR, "error reallocating output\n");
|
|
return ret;
|
|
}
|
|
}
|
|
|
|
/* resample each channel plane */
|
|
for (ch = 0; ch < c->buffer->channels; ch++) {
|
|
out_samples = resample(c, (int16_t *)dst->data[ch],
|
|
(const int16_t *)c->buffer->data[ch], consumed,
|
|
c->buffer->nb_samples, dst->allocated_samples,
|
|
ch + 1 == c->buffer->channels);
|
|
}
|
|
if (out_samples < 0) {
|
|
av_log(c->avr, AV_LOG_ERROR, "error during resampling\n");
|
|
return out_samples;
|
|
}
|
|
|
|
/* drain consumed samples from the internal buffer */
|
|
ff_audio_data_drain(c->buffer, *consumed);
|
|
|
|
av_dlog(c->avr, "resampled %d in + %d leftover to %d out + %d leftover\n",
|
|
in_samples, in_leftover, out_samples, c->buffer->nb_samples);
|
|
|
|
dst->nb_samples = out_samples;
|
|
return 0;
|
|
}
|
|
|
|
int avresample_get_delay(AVAudioResampleContext *avr)
|
|
{
|
|
if (!avr->resample_needed || !avr->resample)
|
|
return 0;
|
|
|
|
return avr->resample->buffer->nb_samples;
|
|
}
|