[FFmpeg-devel] [PATCH] libavfilter: add atempo filter (revised patch v3)

Stefano Sabatini stefasab at gmail.com
Sun Jun 10 21:17:42 CEST 2012


On date Friday 2012-06-08 09:16:36 -0600, Pavel Koshevoy encoded:
> Add atempo audio filter for adjusting audio tempo without affecting
> pitch. This filter implements WSOLA algorithm with fast cross
> correlation calculation in frequency domain.
> 
> Signed-off-by: Pavel Koshevoy <pavel at homestead.aragog.com>
> ---
>  Changelog                |    1 +
>  configure                |    1 +
>  doc/filters.texi         |   18 +
>  libavfilter/Makefile     |    2 +
>  libavfilter/af_atempo.c  | 1218 ++++++++++++++++++++++++++++++++++++++++++++++
>  libavfilter/allfilters.c |    1 +
>  6 files changed, 1241 insertions(+), 0 deletions(-)
>  create mode 100644 libavfilter/af_atempo.c
> 
> diff --git a/Changelog b/Changelog
> index 41b0bdc..a639c71 100644
> --- a/Changelog
> +++ b/Changelog
> @@ -5,6 +5,7 @@ version next:
>  - INI and flat output in ffprobe
>  - Scene detection in libavfilter
>  - Indeo Audio decoder
> +- atempo filter
>  
>  
>  version 0.11:
> diff --git a/configure b/configure
> index 7a60f29..2eac85f 100755
> --- a/configure
> +++ b/configure
> @@ -1690,6 +1690,7 @@ amovie_filter_deps="avcodec avformat"
>  aresample_filter_deps="swresample"
>  ass_filter_deps="libass"
>  asyncts_filter_deps="avresample"
> +atempo_filter_deps="avcodec"
>  blackframe_filter_deps="gpl"
>  boxblur_filter_deps="gpl"
>  colormatrix_filter_deps="gpl"
> diff --git a/doc/filters.texi b/doc/filters.texi
> index 150bde3..c10d133 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -273,6 +273,24 @@ For example, to resample the input audio to 44100Hz:
>  aresample=44100
>  @end example
>  
> + at section atempo
> +
> +Adjust audio tempo.
> +
> +The filter accepts exactly one parameter, the audio tempo.  If not
> +specified then the filter will assume nominal 1.0 tempo.  Tempo must
> +be in the [0.5, 2.0] range.
> +
> +For example, to slow down audio to 80% tempo:
> + at example
> +atempo=0.8
> + at end example
> +
> +For example, to speed up audio to 125% tempo:
> + at example
> +atempo=1.25
> + at end example
> +
>  @section ashowinfo
>  
>  Show a line containing various information for each input audio frame.
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index 29345fc..a1ced51 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -8,6 +8,7 @@ FFLIBS-$(CONFIG_RESAMPLE_FILTER) += avresample
>  FFLIBS-$(CONFIG_ACONVERT_FILTER)             += swresample
>  FFLIBS-$(CONFIG_AMOVIE_FILTER)               += avformat avcodec
>  FFLIBS-$(CONFIG_ARESAMPLE_FILTER)            += swresample
> +FFLIBS-$(CONFIG_ATEMPO_FILTER)               += avcodec
>  FFLIBS-$(CONFIG_MOVIE_FILTER)                += avformat avcodec
>  FFLIBS-$(CONFIG_PAN_FILTER)                  += swresample
>  FFLIBS-$(CONFIG_REMOVELOGO_FILTER)           += avformat avcodec
> @@ -54,6 +55,7 @@ OBJS-$(CONFIG_ASHOWINFO_FILTER)              += af_ashowinfo.o
>  OBJS-$(CONFIG_ASPLIT_FILTER)                 += split.o
>  OBJS-$(CONFIG_ASTREAMSYNC_FILTER)            += af_astreamsync.o
>  OBJS-$(CONFIG_ASYNCTS_FILTER)                += af_asyncts.o
> +OBJS-$(CONFIG_ATEMPO_FILTER)                 += af_atempo.o
>  OBJS-$(CONFIG_EARWAX_FILTER)                 += af_earwax.o
>  OBJS-$(CONFIG_PAN_FILTER)                    += af_pan.o
>  OBJS-$(CONFIG_RESAMPLE_FILTER)               += af_resample.o
> diff --git a/libavfilter/af_atempo.c b/libavfilter/af_atempo.c
> new file mode 100644
> index 0000000..4c59b34
> --- /dev/null
> +++ b/libavfilter/af_atempo.c
> @@ -0,0 +1,1218 @@
> +/*
> + * Copyright (c) 2012 Pavel Koshevoy <pkoshevoy at gmail.com>
> + *
> + * 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
> + * tempo scaling audio filter -- an implementation of WSOLA algorithm
> + *
> + * Based on MIT licensed yaeAudioTempoFilter.h and yaeAudioFragment.h
> + * from Apprentice Video player by Pavel Koshevoy.
> + * https://sourceforge.net/projects/apprenticevideo/
> + *
> + * An explanation of SOLA algorithm is available at
> + * http://www.surina.net/article/time-and-pitch-scaling.html
> + *
> + * WSOLA is very similar to SOLA, only one major difference exists between
> + * these algorithms.  SOLA shifts audio fragments along the output stream,
> + * where as WSOLA shifts audio fragments along the input stream.
> + *
> + * The advantage of WSOLA algorithm is that the overlap region size is
> + * always the same, therefore the blending function is constant and
> + * can be precomputed.
> + */
> +
> +#include <float.h>
> +#include "libavcodec/avfft.h"
> +#include "libavutil/avassert.h"
> +#include "libavutil/avstring.h"
> +#include "libavutil/eval.h"
> +#include "libavutil/opt.h"
> +#include "libavutil/samplefmt.h"
> +#include "avfilter.h"
> +#include "audio.h"
> +#include "internal.h"
> +
> +/**
> + * A fragment of audio waveform
> + */
> +typedef struct {
> +    // index of the first sample of this fragment in the overall waveform;
> +    // 0: input sample position
> +    // 1: output sample position
> +    int64_t position[2];
> +
> +    // original packed multi-channel samples:
> +    uint8_t *data;
> +
> +    // number of samples in this fragment:
> +    int nsamples;
> +
> +    // FFT transform of the downmixed mono fragment, used for
> +    // fast waveform alignment via correlation in frequency domain:
> +    FFTComplex *xdat;
> +
> +} AudioFragment;
> +

> +/**
> + * Filter state machine states
> + */
> +typedef enum {
> +    YAE_LOAD_FRAGMENT      = 0,
> +    YAE_ADJUST_POSITION    = 1,
> +    YAE_RELOAD_FRAGMENT    = 2,
> +    YAE_OUTPUT_OVERLAP_ADD = 3,
> +    YAE_FLUSH_OUTPUT       = 4,
> +
> +} FilterState;

Nit: no need to explicitely enumerate, C compiler will deal with it
starting at 0. YAE_ prefixes still look weird but I won't insist if
you want to keep them.

> +/**
> + * Filter state machine
> + */
> +typedef struct {
> +    // ring-buffer of input samples, necessary because some times
> +    // input fragment position may be adjusted backwards:
> +    uint8_t *buffer;
> +
> +    // ring-buffer maximum capacity,
> +    // expressed as number of multi-channel sample units;
> +    //
> +    // for example, given stereo data 1 multi-channel sample unit
> +    // refers to 2 samples for left/right channels:
> +    int ring;
> +
> +    // ring-buffer house keeping:
> +    int size;
> +    int head;
> +    int tail;

Also note that we have several buffer implementations, in particular
libavutil/fifo.h and audio_fifo.h, don't know if they could be used in
this particular case.

> +    // 0: input sample position corresponding to the ring buffer tail
> +    // 1: output sample position
> +    int64_t position[2];
> +
> +    // sample format:
> +    enum AVSampleFormat format;
> +
> +    // number of channels:
> +    int channels;
> +
> +    // row of bytes to skip from one sample to next, across multple channels;
> +    // stride = (number-of-channels * bits-per-sample-per-channel) / 8
> +    int stride;
> +
> +    // fragment window size, power-of-two integer:
> +    int window;
> +
> +    // Hann window coefficients, for feathering
> +    // (blending) the overlapping fragment region:
> +    float *hann;
> +
> +    // tempo scaling factor:
> +    double tempo;
> +
> +    // cumulative alignment drift:
> +    int drift;
> +
> +    // current/previous fragment ring-buffer:
> +    AudioFragment frag[2];
> +
> +    // current fragment index:
> +    uint64_t nfrag;
> +
> +    // current state:
> +    FilterState state;
> +
> +    // for fast correlation calculation in frequency domain:
> +    FFTContext *fft_forward;
> +    FFTContext *fft_inverse;
> +    FFTComplex *correlation;
> +
> +    // for managing AVFilterPad.request_frame and AVFilterPad.filter_samples
> +    int request_fulfilled;
> +    AVFilterBufferRef *dst_buffer;
> +    uint8_t *dst;
> +    uint8_t *dst_end;
> +    uint64_t nsamples_in;
> +    uint64_t nsamples_out;
> +
> +} ATempoContext;
> +
> +/**
> + * This resets filter to initial state, does not deallocate existing buffers.
> + */

Nit++: Reset filter to initial state, do not deallocate ...

> +static void yae_clear(ATempoContext *atempo)
> +{
> +    atempo->size = 0;
> +    atempo->head = 0;
> +    atempo->tail = 0;
> +
> +    atempo->drift = 0;
> +    atempo->nfrag = 0;
> +    atempo->state = YAE_LOAD_FRAGMENT;
> +
> +    atempo->position[0] = 0;
> +    atempo->position[1] = 0;
> +
> +    atempo->frag[0].position[0] = 0;
> +    atempo->frag[0].position[1] = 0;
> +    atempo->frag[0].nsamples    = 0;
> +
> +    atempo->frag[1].position[0] = 0;
> +    atempo->frag[1].position[1] = 0;
> +    atempo->frag[1].nsamples    = 0;
> +
>
> +    // shift left position of 1st fragment by half a window
> +    // so that no re-normalization would be required for
> +    // the left half of the 1st fragment:
> +    atempo->frag[0].position[0] = -(int64_t)(atempo->window / 2);
> +    atempo->frag[0].position[1] = -(int64_t)(atempo->window / 2);
> +
> +    avfilter_unref_bufferp(&atempo->dst_buffer);
>
> +    atempo->dst     = NULL;
> +    atempo->dst_end = NULL;
> +
> +    atempo->request_fulfilled = 0;
> +    atempo->nsamples_in       = 0;
> +    atempo->nsamples_out      = 0;
> +}
> +
> +/**
> + * Prepare filter for processing audio data of given format,
> + * sample rate and number of channels.
> + */
> +static int yae_reset(ATempoContext *atempo,
> +                     enum AVSampleFormat format,
> +                     int sample_rate,
> +                     int channels)
> +{
> +    const int sample_size = av_get_bytes_per_sample(format);
> +    uint32_t nlevels  = 0;
> +    uint32_t pot;
> +    int i;
> +
> +    atempo->format   = format;
> +    atempo->channels = channels;
> +    atempo->stride   = sample_size * channels;
> +
> +    // pick a segment window size:
> +    atempo->window = sample_rate / 24;
> +
> +    // adjust window size to be a power-of-two integer:
> +    nlevels = av_log2(atempo->window);
> +    pot = 1 << nlevels;
> +    av_assert0(pot <= atempo->window);
> +
> +    if (pot < atempo->window) {
> +        atempo->window = pot * 2;
> +        nlevels++;
> +    }
> +
> +    // initialize audio fragment buffers:
> +    atempo->frag[0].data = av_realloc(atempo->frag[0].data,
> +                                      atempo->window * atempo->stride);
> +    if (!atempo->frag[0].data) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    atempo->frag[1].data = av_realloc(atempo->frag[1].data,
> +                                      atempo->window * atempo->stride);
> +    if (!atempo->frag[1].data) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    atempo->frag[0].xdat = av_realloc(atempo->frag[0].xdat,
> +                                      atempo->window * 2 *
> +                                      sizeof(FFTComplex));
> +    if (!atempo->frag[0].xdat) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    atempo->frag[1].xdat = av_realloc(atempo->frag[1].xdat,
> +                                      atempo->window * 2 *
> +                                      sizeof(FFTComplex));
> +    if (!atempo->frag[1].xdat) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    // initialize FFT contexts:
> +    av_fft_end(atempo->fft_forward);
> +    av_fft_end(atempo->fft_inverse);
> +
> +    atempo->fft_forward = av_fft_init(nlevels + 1, 0);
> +    if (!atempo->fft_forward) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    atempo->fft_inverse = av_fft_init(nlevels + 1, 1);
> +    if (!atempo->fft_inverse) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    atempo->correlation = av_realloc(atempo->correlation,
> +                                     atempo->window * 2 *
> +                                     sizeof(FFTComplex));
> +    if (!atempo->correlation) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    atempo->ring = atempo->window * 3;
> +    atempo->buffer = av_realloc(atempo->buffer, atempo->ring * atempo->stride);
> +    if (!atempo->buffer) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    // initialize the Hann window function:
> +    atempo->hann = av_realloc(atempo->hann, atempo->window * sizeof(float));
> +    if (!atempo->hann) {
> +        return AVERROR(ENOMEM);
> +    }
> +
> +    for (i = 0; i < atempo->window; i++) {
> +        double t = (double)i / (double)(atempo->window - 1);
> +        double h = 0.5 * (1.0 - cos(2.0 * M_PI * t));
> +        atempo->hann[i] = (float)h;
> +    }
> +
> +    yae_clear(atempo);
> +    return 0;
> +}
> +
> +static int yae_set_tempo(ATempoContext *atempo,
> +                         double tempo,
> +                         AVFilterContext *ctx)
> +{
> +    if (tempo < 0.5 || tempo > 2.0) {
> +        av_log(ctx, AV_LOG_ERROR, "tempo value %f exceeds [0.5, 2.0] range\n",
> +               tempo);
> +        return AVERROR(EINVAL);
> +    }
> +
> +    atempo->tempo = tempo;
> +    return 0;
> +}
> +
> +inline static AudioFragment *yae_curr_frag(ATempoContext *atempo)
> +{
> +    return &atempo->frag[atempo->nfrag % 2];
> +}
> +
> +inline static AudioFragment *yae_prev_frag(ATempoContext *atempo)
> +{
> +    return &atempo->frag[(atempo->nfrag + 1) % 2];
> +}
> +
> +inline static void yae_transform(FFTComplex *xdat, FFTContext *fft)
> +{
> +    av_fft_permute(fft, xdat);
> +    av_fft_calc(fft, xdat);
> +}
> +
> +/**
> + * A helper macro for initializing complex data buffer with scalar data
> + * of a given type.
> + */
> +#define yae_init_xdat(TScalar, scalar_max)                              \

nit+: TScaler -> scalar_type

> +    do {                                                                \
> +        const uint8_t *src_end =                                        \
> +            src + frag->nsamples * atempo->channels * sizeof(TScalar);  \
> +                                                                        \
> +        FFTComplex *xdat = frag->xdat;                                  \
> +        TScalar tmp;                                                    \
> +                                                                        \
> +        if (atempo->channels == 1) {                                    \
> +            for (; src < src_end; blend++) {                            \
> +                tmp = *(const TScalar *)src;                            \
> +                src += sizeof(TScalar);                                 \
> +                                                                        \
> +                xdat->re = (FFTSample)tmp;                              \
> +                xdat->im = 0;                                           \
> +                xdat++;                                                 \
> +            }                                                           \
> +        } else {                                                        \
> +            FFTSample s;                                                \
> +            FFTSample max;                                              \
> +            FFTSample ti;                                               \
> +            FFTSample si;                                               \

nit++: this can be put on a single line, save some vertical space

> +            int i;                                                      \
> +                                                                        \
> +            for (; src < src_end; blend++) {                            \
> +                tmp = *(const TScalar *)src;                            \
> +                src += sizeof(TScalar);                                 \
> +                                                                        \
> +                max = (FFTSample)tmp;                                   \
> +                s = FFMIN((FFTSample)scalar_max,                        \
> +                          (FFTSample)fabsf(max));                       \
> +                                                                        \
> +                for (i = 1; i < atempo->channels; i++) {                \
> +                    tmp = *(const TScalar *)src;                        \
> +                    src += sizeof(TScalar);                             \
> +                                                                        \
> +                    ti = (FFTSample)tmp;                                \
> +                    si = FFMIN((FFTSample)scalar_max,                   \
> +                               (FFTSample)fabsf(ti));                   \
> +                                                                        \
> +                    if (s < si) {                                       \
> +                        s   = si;                                       \
> +                        max = ti;                                       \
> +                    }                                                   \
> +                }                                                       \
> +                                                                        \
> +                xdat->re = max;                                         \
> +                xdat->im = 0;                                           \
> +                xdat++;                                                 \
> +            }                                                           \
> +        }                                                               \
> +    } while (0)
> +
> +/**
> + * Initialize complex data buffer of a given audio fragment
> + * with down-mixed mono data of appropriate scalar type.
> + */
> +static void yae_downmix(ATempoContext *atempo, AudioFragment *frag)
> +{
> +    // shortcuts:
> +    const uint8_t *src = frag->data;
> +    const float *blend = atempo->hann;
> +
> +    // init complex data buffer used for FFT and Correlation:
> +    memset(frag->xdat, 0, sizeof(FFTComplex) * atempo->window * 2);
> +
> +    if (atempo->format == AV_SAMPLE_FMT_U8) {
> +        yae_init_xdat(uint8_t, 127);
> +    } else if (atempo->format == AV_SAMPLE_FMT_S16) {
> +        yae_init_xdat(int16_t, 32767);
> +    } else if (atempo->format == AV_SAMPLE_FMT_S32) {
> +        yae_init_xdat(int, 2147483647);
> +    } else if (atempo->format == AV_SAMPLE_FMT_FLT) {
> +        yae_init_xdat(float, 1);
> +    } else if (atempo->format == AV_SAMPLE_FMT_DBL) {
> +        yae_init_xdat(double, 1);
> +    }
> +}
> +
> +/**
> + * Populate the internal data buffer on as-needed basis.
> + *
> + * @return 0 if requested data was already available or was loaded.
> + * @return AVERROR(EAGAIN) if more input data is required.

nit+++: avoid double @return, a single paragraph is fine (not that it
gains much since this is internal)

> + */
> +static int yae_load_data(ATempoContext *atempo,
> +                         const uint8_t **src_ref,
> +                         const uint8_t *src_end,
> +                         int64_t stop_here)
> +{
> +    // shortcut:
> +    const uint8_t *src = *src_ref;
> +    const int read_size = stop_here - atempo->position[0];
> +
> +    if (stop_here <= atempo->position[0]) {
> +        return 0;
> +    }
> +
> +    // samples are not expected to be skipped:
> +    av_assert0(read_size <= atempo->ring);
> +
> +    while (atempo->position[0] < stop_here && src < src_end) {
> +        int src_samples = (src_end - src) / atempo->stride;
> +
> +        // load data piece-wise, in order to avoid complicating the logic:
> +        int nsamples = FFMIN(read_size, src_samples);
> +        int na;
> +        int nb;
> +
> +        nsamples = FFMIN(nsamples, atempo->ring);
> +        na = FFMIN(nsamples, atempo->ring - atempo->tail);
> +        nb = FFMIN(nsamples - na, atempo->ring);
> +
> +        if (na) {
> +            uint8_t *a = atempo->buffer + atempo->tail * atempo->stride;
> +            memcpy(a, src, na * atempo->stride);
> +
> +            src += na * atempo->stride;
> +            atempo->position[0] += na;
> +
> +            atempo->size = FFMIN(atempo->size + na, atempo->ring);
> +            atempo->tail = (atempo->tail + na) % atempo->ring;
> +            atempo->head =
> +                atempo->size < atempo->ring ?
> +                atempo->tail - atempo->size :
> +                atempo->tail;
> +        }
> +
> +        if (nb) {
> +            uint8_t *b = atempo->buffer;
> +            memcpy(b, src, nb * atempo->stride);
> +
> +            src += nb * atempo->stride;
> +            atempo->position[0] += nb;
> +
> +            atempo->size = FFMIN(atempo->size + nb, atempo->ring);
> +            atempo->tail = (atempo->tail + nb) % atempo->ring;
> +            atempo->head =
> +                atempo->size < atempo->ring ?
> +                atempo->tail - atempo->size :
> +                atempo->tail;
> +        }
> +    }
> +
> +    // pass back the updated source buffer pointer:
> +    *src_ref = src;
> +
> +    // sanity check:
> +    av_assert0(atempo->position[0] <= stop_here);
> +
> +    return atempo->position[0] == stop_here ? 0 : AVERROR(EAGAIN);
> +}
> +

> +/**
> + * Populate current audio fragment data buffer.
> + *
> + * @return 0 when the fragment is ready.
> + * @return AVERROR(EAGAIN) if more input data is required.

ditto

> + */
> +static int yae_load_frag(ATempoContext *atempo,
> +                         const uint8_t **src_ref,
> +                         const uint8_t *src_end)
> +{

> +    // shortcuts:
> +    AudioFragment *frag = yae_curr_frag(atempo);
> +    uint8_t *dst;
> +    int64_t missing;
> +    uint32_t nsamples;
> +    int64_t start;
> +    int64_t zeros;
> +    int na;
> +    int nb;
> +    const uint8_t *a;
> +    const uint8_t *b;
> +    int i0;
> +    int i1;
> +    int n0;
> +    int n1;

nit++: try to group variables together, so we get a shorter
declaration list (easier to read)

[...]
> +static av_cold int init(AVFilterContext *ctx, const char *args, void *opaque)
> +{
> +    ATempoContext *atempo = ctx->priv;
> +    double tempo = 1.0;
> +
> +    // NOTE: this assumes that the caller has memset ctx->priv to 0:
> +    atempo->format = AV_SAMPLE_FMT_NONE;
> +    atempo->tempo  = 1.0;
> +    atempo->state  = YAE_LOAD_FRAGMENT;
> +
> +    if (args) {
> +        char *tail = NULL;
> +        tempo = av_strtod(args, &tail);
> +
> +        if (tail && *tail) {
> +            av_log(ctx, AV_LOG_ERROR, "invalid tempo value '%s'\n", args);

Nit+++: "Invalid..."

> +            return AVERROR(EINVAL);
> +        }
> +    }
> +
> +    return yae_set_tempo(atempo, tempo, ctx);
> +}
> +
> +static av_cold void uninit(AVFilterContext *ctx)
> +{
> +    ATempoContext *atempo = ctx->priv;
> +    yae_clear(atempo);
> +
> +    av_freep(&atempo->frag[0].data);
> +    av_freep(&atempo->frag[1].data);
> +    av_freep(&atempo->frag[0].xdat);
> +    av_freep(&atempo->frag[1].xdat);
> +
> +    av_freep(&atempo->buffer);
> +    av_freep(&atempo->hann);
> +    av_freep(&atempo->correlation);
> +
> +    av_fft_end(atempo->fft_forward);
> +    atempo->fft_forward = NULL;
> +
> +    av_fft_end(atempo->fft_inverse);
> +    atempo->fft_inverse = NULL;
> +}
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> +    AVFilterChannelLayouts *layouts = NULL;
> +    AVFilterFormats        *formats = NULL;
> +
> +    // WSOLA necessitates an internal sliding window ring buffer
> +    // for incoming audio stream.
> +    //
> +    // Planar sample formats are too cumbersome to store in a ring buffer,
> +    // therefore planar sample formats are not supported.
> +    //
> +    enum AVSampleFormat sample_fmts[] = {
> +        AV_SAMPLE_FMT_U8,
> +        AV_SAMPLE_FMT_S16,
> +        AV_SAMPLE_FMT_S32,
> +        AV_SAMPLE_FMT_FLT,
> +        AV_SAMPLE_FMT_DBL,
> +        AV_SAMPLE_FMT_NONE
> +    };
> +
> +    layouts = ff_all_channel_layouts();
> +    if (!layouts) {
> +        return AVERROR(ENOMEM);
> +    }
> +    ff_set_common_channel_layouts(ctx, layouts);
> +
> +    formats = avfilter_make_format_list(sample_fmts);
> +    if (!formats) {
> +        return AVERROR(ENOMEM);
> +    }
> +    avfilter_set_common_sample_formats(ctx, formats);
> +

> +    formats = ff_all_samplerates();
> +    if (!formats) {
> +        return AVERROR(ENOMEM);
> +    }

> +
> +    ff_set_common_samplerates(ctx, formats);

Nit++++: remove the previous empty line, so you group all the instructions

> +    return 0;
> +}
> +

> +/**
> + * Check configuration properties of the input link
> + * and update filters internal state as necessary.
> + *
> + * Return zero on success, AVERRROR(..) on failure.
> + */

feel free to drop this doxy

> +static int
> +config_props(AVFilterLink *inlink)
> +{
> +    AVFilterContext  *ctx = inlink->dst;
> +    ATempoContext *atempo = ctx->priv;
> +
> +    enum AVSampleFormat format = inlink->format;
> +    int sample_rate = (int)inlink->sample_rate;
> +    int channels = av_get_channel_layout_nb_channels(inlink->channel_layout);
> +
> +    return yae_reset(atempo, format, sample_rate, channels);
> +}
> +
> +static void filter_samples(AVFilterLink *inlink,
> +                           AVFilterBufferRef *src_buffer)
> +{
> +    AVFilterContext  *ctx = inlink->dst;
> +    ATempoContext *atempo = ctx->priv;
> +    AVFilterLink *outlink = ctx->outputs[0];
> +
> +    int n_in = src_buffer->audio->nb_samples;
> +    int n_out = (int)(0.5 + ((double)n_in) / atempo->tempo);
> +
> +    const uint8_t *src = src_buffer->data[0];
> +    const uint8_t *src_end = src + n_in * atempo->stride;
> +
> +    while (src < src_end) {
> +        if (!atempo->dst_buffer) {
> +            atempo->dst_buffer = ff_get_audio_buffer(outlink,
> +                                                     AV_PERM_WRITE,
> +                                                     n_out);
> +            avfilter_copy_buffer_ref_props(atempo->dst_buffer, src_buffer);
> +
> +            atempo->dst = atempo->dst_buffer->data[0];
> +            atempo->dst_end = atempo->dst + n_out * atempo->stride;
> +        }
> +
> +        yae_apply(atempo, &src, src_end, &atempo->dst, atempo->dst_end);
> +
> +        if (atempo->dst == atempo->dst_end) {
> +            atempo->dst_buffer->audio->sample_rate = outlink->sample_rate;
> +            atempo->dst_buffer->audio->nb_samples  = n_out;
> +
> +            // adjust the PTS:
> +            atempo->dst_buffer->pts =
> +                av_rescale(outlink->time_base.den,
> +                           atempo->nsamples_out,
> +                           outlink->time_base.num * outlink->sample_rate);

not sure but maybe av_rescale_q(atempo->nsamples_out, (AVRational){1, outlink->sample_rate}, outlink->time_base)

may be safer, as it avoids a possible int*int -> int64_t overflow:
outlink->time_base.num * outlink->sample_rate

> +
> +            ff_filter_samples(outlink, atempo->dst_buffer);
> +            atempo->dst_buffer = NULL;
> +            atempo->dst        = NULL;
> +            atempo->dst_end    = NULL;
> +
> +            atempo->nsamples_out += n_out;
> +            atempo->request_fulfilled = 1;
> +        }
> +    }
> +
> +    atempo->nsamples_in += n_in;

> +    avfilter_unref_buffer(src_buffer);

avfilter_unref_bufferp should be safer

> +}
> +
> +/**
> + * Frame request callback.
> + *
> + * A call to this should result in at least one frame
> + * being output over the given link.
> + *
> + * @return 0 on success.
> + */
> +static int request_frame(AVFilterLink *outlink)
> +{
> +    AVFilterContext  *ctx = outlink->src;
> +    ATempoContext *atempo = ctx->priv;
> +    int ret;
> +
> +    atempo->request_fulfilled = 0;
> +    do {
> +        ret = avfilter_request_frame(ctx->inputs[0]);
> +    }
> +    while (!atempo->request_fulfilled && ret >= 0);
> +
> +    if (ret == AVERROR_EOF) {
> +        // flush the filter:
> +        int n_max = atempo->ring;
> +        int n_out;
> +        int err = AVERROR(EAGAIN);
> +
> +        while (err == AVERROR(EAGAIN)) {
> +            if (!atempo->dst_buffer) {
> +                atempo->dst_buffer = ff_get_audio_buffer(outlink,
> +                                                         AV_PERM_WRITE,
> +                                                         n_max);
> +
> +                atempo->dst = atempo->dst_buffer->data[0];
> +                atempo->dst_end = atempo->dst + n_max * atempo->stride;
> +            }
> +
> +            err = yae_flush(atempo, &atempo->dst, atempo->dst_end);
> +
> +            n_out = ((atempo->dst - atempo->dst_buffer->data[0]) /
> +                     atempo->stride);
> +
> +            if (n_out) {
> +                atempo->dst_buffer->audio->sample_rate = outlink->sample_rate;
> +                atempo->dst_buffer->audio->nb_samples  = n_out;
> +
> +                // adjust the PTS:
> +                atempo->dst_buffer->pts =
> +                    av_rescale(outlink->time_base.den,
> +                               atempo->nsamples_out,
> +                               outlink->time_base.num * outlink->sample_rate);
> +
> +                ff_filter_samples(outlink, atempo->dst_buffer);
> +                atempo->dst_buffer = NULL;
> +                atempo->dst        = NULL;
> +                atempo->dst_end    = NULL;
> +
> +                atempo->nsamples_out += n_out;
> +            }
> +        }
> +
> +        avfilter_unref_bufferp(&atempo->dst_buffer);
> +        atempo->dst     = NULL;
> +        atempo->dst_end = NULL;
> +
> +        return AVERROR_EOF;
> +    }
> +
> +    return ret;
> +}
> +

> +/**
> + * Make the filter instance process a command.
> + *
> + * @param cmd   -- an alphanumeric command to process
> + * @param arg   -- the argument for the command
> + * @param res   -- a buffer with size res_size for filter response
> + * @param flags -- command flags
> + *
> + * When AVFILTER_CMD_FLAG_FAST flag is set time consuming commands
> + * should be treated as unsupported.
> + *
> + * @return 0 on success, otherwise an error code.
> + * @return AVERROR(ENOSYS) on unsupported commands.
> + */

same I'd prefer to skip this doxy, so I won't have to udpate this when
the API changes...

> +static int process_command(AVFilterContext *ctx,
> +                           const char *cmd,
> +                           const char *arg,
> +                           char *res,
> +                           int res_len,
> +                           int flags)
> +{
> +    ATempoContext *atempo = ctx->priv;
> +

> +    if (strcmp(cmd, "tempo") == 0) {
> +        char   *tail = NULL;
> +        double tempo = av_strtod(arg, &tail);
> +
> +        if (tail && *tail) {
> +            av_log(ctx, AV_LOG_ERROR, "invalid tempo value '%s'\n", arg);
> +            return AVERROR(EINVAL);
> +        }
> +
> +        return yae_set_tempo(atempo, tempo, ctx);
> +    }

you could change the signature of yae_set_tempo to
static int yae_set_tempo(AVFilterContext *ctx, const char *tempo)
and factorize some code in init().


> +
> +    return AVERROR(ENOSYS);
> +}
> +
> +AVFilter avfilter_af_atempo = {
> +    .name            = "atempo",
> +    .description     = NULL_IF_CONFIG_SMALL("Adjust audio tempo."),

> +    .init            = &init,
> +    .uninit          = &uninit,
> +    .query_formats   = &query_formats,
> +    .process_command = &process_command,

nit: "&" can be skipped

[...]

I'd love to be able to comment on the more mathematical/algorithmical
part of the patch, but if noone will do I suppose we can assume that
your code has been sufficiently tested.

BTW add your name as maintainer for this filter in MAINTAINERS if you
feel like so.
-- 
FFmpeg = Fierce and Fantastic Meaningful Political Explosive God


More information about the ffmpeg-devel mailing list