[FFmpeg-devel] [PATCH] lavfi: add compand filter
Stefano Sabatini
stefasab at gmail.com
Sat Jul 27 16:00:44 CEST 2013
On date Thursday 2013-07-25 13:24:58 +0000, Paul B Mahol encoded:
> Signed-off-by: Paul B Mahol <onemda at gmail.com>
> ---
> doc/filters.texi | 48 +++++
> libavfilter/Makefile | 1 +
> libavfilter/af_compand.c | 514 +++++++++++++++++++++++++++++++++++++++++++++++
> libavfilter/allfilters.c | 1 +
> 4 files changed, 564 insertions(+)
> create mode 100644 libavfilter/af_compand.c
>
> diff --git a/doc/filters.texi b/doc/filters.texi
> index 0c18446..080c598 100644
> --- a/doc/filters.texi
> +++ b/doc/filters.texi
> @@ -1176,6 +1176,54 @@ front_center.wav -map '[LFE]' lfe.wav -map '[SL]' side_left.wav -map '[SR]'
> side_right.wav
> @end example
>
> + at section compand
> +
> +Compress or expand the dynamic range of the audio.
> +
> +A description of the accepted parameters follows.
s/parameters/options?
> +
> + at table @option
> + at item attacks
> + at item decays
> +Set list of times in seconds for each channel over which the instantaneous level
> +of the input signal is averaged to determine its volume.
> + at option{attacks} refer to increase of volume and @option{decays} to decrease of
> +volume.
refers
> +For most situations, tha attack time (response to the music getting louder)
the attack
overall s/music/sound/
> +should be shorter than the decay time because the human ear is more sensitive
> +to sudden loud music than sudden soft music.
> +Typical value for attack is 0.3 seconds and for decay 0.8 seconds.
> +
> + at item points
> +Set list of points for tranfer function, specified in dB relative to maximum possible
> +signal amplitued.
> +The input values must be in strictly increasing order but the transfer function does
> +not have to me monotonically rising. The point 0/0 is assumed but may be overriden
> +(by 0\out-dBn). Typical values for the transfer function are @code{-70\-60|-20\0}.
Syntax is not specified.
> +
> + at item soft-knee
> +Set amount for which the points at where adjacent line segments on the transfer function meet will be rounded.
> +
> + at item gain
> +Set additional gain in dB to be applied at all points on the transfer function
> +and allows easy adjustment of the overall gain.
> +Default is @code{0}.
> +
> + at item volume
> +Set initial volume in dB to be assumed for each channel when filtering starts.
> +This permits the user to supply a nominal level initially, so that, for example,
> +a very large gain is not applied to initial signal levels before the companding
> +has begun to operates. A typical value, for audio which is initially quiet is -90 dB.
> +Default is @code{0}.
> +
> + at item delay
> +Set delay in seconds. Default is @code{0}. The input audio
> +is analysed immediately, but is delayed before being fed to the
what is delayed?
> +volume adjuster. Specifying a delay approximately equal to the attack/decay
> +times allow the filter to effectively operate in predictive rather than
allows
> +reactive mode.
> + at end table
A few examples would be useful.
> +
> @section earwax
>
> Make audio easier to listen to on headphones.
> diff --git a/libavfilter/Makefile b/libavfilter/Makefile
> index f54e100..3751d54 100644
> --- a/libavfilter/Makefile
> +++ b/libavfilter/Makefile
> @@ -84,6 +84,7 @@ OBJS-$(CONFIG_BASS_FILTER) += af_biquads.o
> OBJS-$(CONFIG_BIQUAD_FILTER) += af_biquads.o
> OBJS-$(CONFIG_CHANNELMAP_FILTER) += af_channelmap.o
> OBJS-$(CONFIG_CHANNELSPLIT_FILTER) += af_channelsplit.o
> +OBJS-$(CONFIG_COMPAND_FILTER) += af_compand.o
> OBJS-$(CONFIG_EARWAX_FILTER) += af_earwax.o
> OBJS-$(CONFIG_EBUR128_FILTER) += f_ebur128.o
> OBJS-$(CONFIG_EQUALIZER_FILTER) += af_biquads.o
> diff --git a/libavfilter/af_compand.c b/libavfilter/af_compand.c
> new file mode 100644
> index 0000000..e76b3fd
> --- /dev/null
> +++ b/libavfilter/af_compand.c
> @@ -0,0 +1,514 @@
> +/*
> + * Copyright (c) 1999 Chris Bagwell
> + * Copyright (c) 1999 Nick Bailey
> + * Copyright (c) 2007 Rob Sykes <robs at users.sourceforge.net>
> + * Copyright (c) 2013 Paul B Mahol
> + *
> + * 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
> + *
> + */
> +
> +#include "libavutil/avstring.h"
> +#include "libavutil/opt.h"
> +#include "libavutil/samplefmt.h"
> +#include "avfilter.h"
> +#include "audio.h"
> +#include "internal.h"
> +
> +typedef struct ChanParam {
> + double attack;
> + double decay;
> + double volume;
> +} ChanParam;
> +
> +typedef struct CompandSegment {
> + double x, y;
> + double a, b;
> +} CompandSegment;
> +
> +typedef struct CompandContext {
> + const AVClass *class;
> + char *attacks, *decays, *points;
> + CompandSegment *segments;
> + ChanParam *channels;
> + double in_min_lin;
> + double out_min_lin;
> + double curve_dB;
> + double gain_dB;
> + double initial_volume;
> + double delay;
> + uint8_t **delayptrs;
> + int delay_samples;
> + int delay_count;
> + int delay_index;
> + int64_t pts;
> +
> + int (*compand)(AVFilterContext *ctx, AVFrame *frame);
> +} CompandContext;
> +
> +#define OFFSET(x) offsetof(CompandContext, x)
> +#define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
> +
> +static const AVOption compand_options[] = {
> + { "attacks", "set time over which increase of volume is determined", OFFSET(attacks), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, A },
> + { "decays", "set time over which decrease of volume is determined", OFFSET(decays), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, A },
> + { "points", "set points of transfer function", OFFSET(points), AV_OPT_TYPE_STRING, {.str=NULL}, 0, 0, A },
> + { "soft-knee", "set soft-knee", OFFSET(curve_dB), AV_OPT_TYPE_DOUBLE, {.dbl=0.01}, 0.01, 900, A },
> + { "gain", "set output gain", OFFSET(gain_dB), AV_OPT_TYPE_DOUBLE, {.dbl=0}, -900, 900, A },
> + { "volume", "set initial volume", OFFSET(initial_volume), AV_OPT_TYPE_DOUBLE, {.dbl=0}, -900, 0, A },
> + { "delay", "set delay for samples before sending them to volume adjuster", OFFSET(delay), AV_OPT_TYPE_DOUBLE, {.dbl=0}, 0, 20, A },
> + { NULL },
> +};
> +
> +AVFILTER_DEFINE_CLASS(compand);
> +
> +static av_cold int init(AVFilterContext *ctx)
> +{
> + CompandContext *s = ctx->priv;
> +
> + if (!s->attacks || !s->decays || !s->points) {
> + av_log(ctx, AV_LOG_ERROR, "Missing attacks and/or decays and/or points.\n");
Nit: skip the ending dot
> + return AVERROR(EINVAL);
> + }
> +
> + return 0;
> +}
> +
> +static av_cold void uninit(AVFilterContext *ctx)
> +{
> + CompandContext *s = ctx->priv;
> +
> + av_freep(&s->channels);
> + av_freep(&s->segments);
> + if (s->delayptrs)
> + av_freep(&s->delayptrs[0]);
> + av_freep(&s->delayptrs);
> +}
> +
> +static int query_formats(AVFilterContext *ctx)
> +{
> + AVFilterChannelLayouts *layouts;
> + AVFilterFormats *formats;
> + static const enum AVSampleFormat sample_fmts[] = {
> + AV_SAMPLE_FMT_DBLP,
> + AV_SAMPLE_FMT_NONE
> + };
> +
> + layouts = ff_all_channel_layouts();
> + if (!layouts)
> + return AVERROR(ENOMEM);
> + ff_set_common_channel_layouts(ctx, layouts);
> +
> + formats = ff_make_format_list(sample_fmts);
> + if (!formats)
> + return AVERROR(ENOMEM);
> + ff_set_common_formats(ctx, formats);
> +
> + formats = ff_all_samplerates();
> + if (!formats)
> + return AVERROR(ENOMEM);
> + ff_set_common_samplerates(ctx, formats);
> +
> + return 0;
> +}
> +
> +static void count_items(char *item_str, int *nb_items)
> +{
> + char *p;
> +
> + *nb_items = 1;
> + for (p = item_str; *p; p++) {
> + if (*p == '|')
> + (*nb_items)++;
> + }
> +
> +}
> +
> +static void update_volume(ChanParam *cp, double in)
> +{
> + double delta = in - cp->volume;
> +
> + if (delta > 0.0)
> + cp->volume += delta * cp->attack;
> + else
> + cp->volume += delta * cp->decay;
> +}
> +
> +static double get_volume(CompandContext *s, double in_lin)
> +{
nit: s -> compand, so the following is more readable (or it's just
me?)
> + CompandSegment *cs;
> + double in_log, out_log;
> + int i;
> +
> + if (in_lin < s->in_min_lin)
> + return s->out_min_lin;
> +
> + in_log = log(in_lin);
> +
> + for (i = 1;; i++)
> + if (in_log <= s->segments[i + 1].x)
> + break;
can't overflow?
> + cs = &s->segments[i];
> + in_log -= cs->x;
> + out_log = cs->y + in_log * (cs->a * in_log + cs->b);
> +
> + return exp(out_log);
> +}
> +
> +static int compand_nodelay(AVFilterContext *ctx, AVFrame *frame)
> +{
> + CompandContext *s = ctx->priv;
> + AVFilterLink *inlink = ctx->inputs[0];
> + const int channels = inlink->channels;
> + const int nb_samples = frame->nb_samples;
> + AVFrame *out_frame;
> + int chan, i;
> +
> + if (av_frame_is_writable(frame)) {
> + out_frame = frame;
> + } else {
> + out_frame = ff_get_audio_buffer(inlink, nb_samples);
> + if (!out_frame)
> + return AVERROR(ENOMEM);
> + av_frame_copy_props(out_frame, frame);
> + }
> +
> + for (chan = 0; chan < channels; chan++) {
> + const double *src = (double *)frame->data[chan];
> + double *dst = (double *)out_frame->data[chan];
> + ChanParam *cp = &s->channels[chan];
> + double volume = get_volume(s, cp->volume);
> +
> + for (i = 0; i < nb_samples; i++) {
> + update_volume(cp, fabs(src[i]));
> +
> + dst[i] = av_clipd(src[i] * volume, -1, 1);
> + }
> + }
> +
> + if (frame != out_frame)
> + av_frame_free(&frame);
> +
> + return ff_filter_frame(ctx->outputs[0], out_frame);
> +}
> +
> +#define MOD(a, b) (((a) >= (b)) ? (a) - (b) : (a))
> +
> +static int compand_delay(AVFilterContext *ctx, AVFrame *frame)
> +{
> + CompandContext *s = ctx->priv;
> + AVFilterLink *inlink = ctx->inputs[0];
> + const int channels = inlink->channels;
> + const int nb_samples = frame->nb_samples;
> + int chan, i, dindex, oindex, count;
> + AVFrame *out_frame = NULL;
> +
> + for (chan = 0; chan < channels; chan++) {
> + const double *src = (double *)frame->data[chan];
> + double *dbuf = (double *)s->delayptrs[chan];
> + double *dst;
> + ChanParam *cp = &s->channels[chan];
> +
> + count = s->delay_count;
> + dindex = s->delay_index;
> + for (i = 0, oindex = 0; i < nb_samples; i++) {
> + const double in = src[i];
> + update_volume(cp, fabs(in));
> +
> + if (count >= s->delay_samples) {
> + if (!out_frame) {
> + out_frame = ff_get_audio_buffer(inlink, nb_samples - i);
> + if (!out_frame)
> + return AVERROR(ENOMEM);
> + av_frame_copy_props(out_frame, frame);
> + out_frame->pts = s->pts;
> + s->pts += av_rescale_q(nb_samples - i, (AVRational){1, inlink->sample_rate}, inlink->time_base);
> + }
> +
> + dst = (double *)out_frame->data[chan];
> + dst[oindex++] = av_clipd(dbuf[dindex] * get_volume(s, cp->volume), -1, 1);
> + } else {
> + count++;
> + }
> +
> + dbuf[dindex] = in;
> + dindex = MOD(dindex + 1, s->delay_samples);
> + }
> + }
> +
> + s->delay_count = count;
> + s->delay_index = dindex;
> +
> + av_frame_free(&frame);
> + return out_frame ? ff_filter_frame(ctx->outputs[0], out_frame) : 0;
> +}
> +
> +static int compand_drain(AVFilterLink *outlink)
> +{
> + AVFilterContext *ctx = outlink->src;
> + CompandContext *s = ctx->priv;
> + const int channels = outlink->channels;
> + int chan, i, dindex;
> + AVFrame *frame = NULL;
> +
> + frame = ff_get_audio_buffer(outlink, FFMIN(2048, s->delay_count));
> + if (!frame)
> + return AVERROR(ENOMEM);
> + frame->pts = s->pts;
> + s->pts += av_rescale_q(frame->nb_samples, (AVRational){1, outlink->sample_rate}, outlink->time_base);
> +
> + for (chan = 0; chan < channels; chan++) {
> + double *dbuf = (double *)s->delayptrs[chan];
> + double *dst = (double *)frame->data[chan];
> + ChanParam *cp = &s->channels[chan];
> +
> + dindex = s->delay_index;
> + for (i = 0; i < frame->nb_samples; i++) {
> + dst[i] = av_clipd(dbuf[dindex] * get_volume(s, cp->volume), -1, 1);
> + dindex = MOD(dindex + 1, s->delay_samples);
> + }
> + }
> + s->delay_count -= frame->nb_samples;
> + s->delay_index = dindex;
> +
> + return ff_filter_frame(outlink, frame);
> +}
> +
> +static int config_output(AVFilterLink *outlink)
> +{
> + AVFilterContext *ctx = outlink->src;
> + CompandContext *s = ctx->priv;
> + const int sample_rate = outlink->sample_rate;
> + double radius = s->curve_dB * M_LN10 / 20;
> + int nb_attacks, nb_decays, nb_points;
> + char *p, *saveptr = NULL;
> + int new_nb_items, num;
> + int i;
> +
> + count_items(s->attacks, &nb_attacks);
> + count_items(s->decays, &nb_decays);
> + count_items(s->points, &nb_points);
> +
> + if ((nb_attacks > outlink->channels) || (nb_decays > outlink->channels))
> + return AVERROR(EINVAL);
No feedback?
> +
> + uninit(ctx);
why?
> +
> + s->channels = av_mallocz_array(outlink->channels, sizeof(*s->channels));
> + s->segments = av_mallocz_array((nb_points + 4) * 2, sizeof(*s->segments));
> +
> + if (!s->channels || !s->segments)
> + return AVERROR(ENOMEM);
> +
> + p = s->attacks;
> + for (i = 0, new_nb_items = 0; i < nb_attacks; i++) {
> + char *tstr = av_strtok(p, "|", &saveptr);
> + p = NULL;
> + new_nb_items += sscanf(tstr, "%lf", &s->channels[i].attack) == 1;
> + if (s->channels[i].attack < 0)
> + return AVERROR(EINVAL);
> + }
> + nb_attacks = new_nb_items;
> +
> + p = s->decays;
> + for (i = 0, new_nb_items = 0; i < nb_decays; i++) {
> + char *tstr = av_strtok(p, "|", &saveptr);
> + p = NULL;
> + new_nb_items += sscanf(tstr, "%lf", &s->channels[i].decay) == 1;
> + if (s->channels[i].decay < 0)
> + return AVERROR(EINVAL);
> + }
> + nb_decays = new_nb_items;
> +
Nit: could be factorized
> + if (nb_attacks != nb_decays) {
> + av_log(ctx, AV_LOG_ERROR, "Number of attacks %d differs from number of decays %d.\n", nb_attacks, nb_decays);
Nit: skip ending dot, here and in the messages below
> + return AVERROR(EINVAL);
> + }
> +
> +#define S(x) s->segments[2 * ((x) + 1)]
> + p = s->points;
> + for (i = 0, new_nb_items = 0; i < nb_points; i++) {
> + char *tstr = av_strtok(p, "|", &saveptr);
> + p = NULL;
> + if (sscanf(tstr, "%lf/%lf", &S(i).x, &S(i).y) != 2) {
> + av_log(ctx, AV_LOG_ERROR, "Invalid and/or missing input/output value.\n");
> + return AVERROR(EINVAL);
> + }
> + if (i && S(i - 1).x >= S(i).x) {
> + av_log(ctx, AV_LOG_ERROR, "Transfer function input values must be increasing.\n");
> + return AVERROR(EINVAL);
> + }
> + S(i).y -= S(i).x;
> + av_log(ctx, AV_LOG_DEBUG, "%d: x=%lf y=%lf\n", i, S(i).x, S(i).y);
> + new_nb_items++;
> + }
> + num = new_nb_items;
> +
> + /* Add 0,0 if necessary */
> + if (num == 0 || S(num - 1).x)
> + num++;
> +
> +#undef S
> +#define S(x) s->segments[2 * (x)]
> + /* Add a tail off segment at the start */
> + S(0).x = S(1).x - 2 * s->curve_dB;
> + S(0).y = S(1).y;
> + num++;
> +
> + /* Join adjacent colinear segments */
> + for (i = 2; i < num; i++) {
> + double g1 = (S(i - 1).y - S(i - 2).y) * (S(i - 0).x - S(i - 1).x);
> + double g2 = (S(i - 0).y - S(i - 1).y) * (S(i - 1).x - S(i - 2).x);
> + int j;
> +
> + if (fabs(g1 - g2))
> + continue;
> + num--;
> + for (j = --i; j < num; j++)
> + S(j) = S(j + 1);
> + }
> +
> + for (i = 0; !i || s->segments[i - 2].x; i += 2) {
> + s->segments[i].y += s->gain_dB;
> + s->segments[i].x *= M_LN10 / 20;
> + s->segments[i].y *= M_LN10 / 20;
> + }
> +
> +#define L(x) s->segments[i - (x)]
> + for (i = 4; s->segments[i - 2].x; i += 2) {
> + double x, y, cx, cy, in1, in2, out1, out2, theta, len, r;
> +
> + L(4).a = 0;
> + L(4).b = (L(2).y - L(4).y) / (L(2).x - L(4).x);
> +
> + L(2).a = 0;
> + L(2).b = (L(0).y - L(2).y) / (L(0).x - L(2).x);
> +
> + theta = atan2(L(2).y - L(4).y, L(2).x - L(4).x);
> + len = sqrt(pow(L(2).x - L(4).x, 2.) + pow(L(2).y - L(4).y, 2.));
> + r = FFMIN(radius, len);
> + L(3).x = L(2).x - r * cos(theta);
> + L(3).y = L(2).y - r * sin(theta);
> +
> + theta = atan2(L(0).y - L(2).y, L(0).x - L(2).x);
> + len = sqrt(pow(L(0).x - L(2).x, 2.) + pow(L(0).y - L(2).y, 2.));
> + r = FFMIN(radius, len / 2);
> + x = L(2).x + r * cos(theta);
> + y = L(2).y + r * sin(theta);
> +
> + cx = (L(3).x + L(2).x + x) / 3;
> + cy = (L(3).y + L(2).y + y) / 3;
> +
> + L(2).x = x;
> + L(2).y = y;
> +
> + in1 = cx - L(3).x;
> + out1 = cy - L(3).y;
> + in2 = L(2).x - L(3).x;
> + out2 = L(2).y - L(3).y;
> + L(3).a = (out2 / in2 - out1 / in1) / (in2-in1);
> + L(3).b = out1 / in1 - L(3).a * in1;
> + }
> + L(3).x = 0;
> + L(3).y = L(2).y;
> +
> + s->in_min_lin = exp(s->segments[1].x);
> + s->out_min_lin = exp(s->segments[1].y);
> +
> + for (i = 0; i < outlink->channels; i++) {
> + ChanParam *cp = &s->channels[i];
> +
> + if (cp->attack > 1.0 / sample_rate)
> + cp->attack = 1.0 - exp(-1.0 / (sample_rate * cp->attack));
> + else
> + cp->attack = 1.0;
> + if (cp->decay > 1.0 / sample_rate)
> + cp->decay = 1.0 - exp(-1.0 / (sample_rate * cp->decay));
> + else
> + cp->decay = 1.0;
> + cp->volume = pow(10.0, s->initial_volume / 20);
> + }
> +
> + s->delay_samples = s->delay * sample_rate;
> + if (s->delay_samples > 0) {
> + int ret;
> + if ((ret = av_samples_alloc_array_and_samples(&s->delayptrs, NULL,
> + outlink->channels,
> + s->delay_samples,
> + outlink->format, 0)) < 0)
> + return ret;
> + s->compand = compand_delay;
> + outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
> + } else {
> + s->compand = compand_nodelay;
> + }
> + return 0;
> +}
> +
> +static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
> +{
> + AVFilterContext *ctx = inlink->dst;
> + CompandContext *s = ctx->priv;
> +
> + return s->compand(ctx, frame);
> +}
> +
> +static int request_frame(AVFilterLink *outlink)
> +{
> + AVFilterContext *ctx = outlink->src;
> + CompandContext *s = ctx->priv;
> + int ret;
> +
> + ret = ff_request_frame(ctx->inputs[0]);
> +
> + if (ret == AVERROR_EOF && !ctx->is_disabled && s->delay_count)
> + ret = compand_drain(outlink);
> +
> + return ret;
> +}
> +
> +static const AVFilterPad compand_inputs[] = {
> + {
> + .name = "default",
> + .type = AVMEDIA_TYPE_AUDIO,
> + .filter_frame = filter_frame,
> + },
> + { NULL },
> +};
> +
> +static const AVFilterPad compand_outputs[] = {
> + {
> + .name = "default",
> + .request_frame = request_frame,
> + .config_props = config_output,
> + .type = AVMEDIA_TYPE_AUDIO,
> + },
> + { NULL },
> +};
> +
> +AVFilter avfilter_af_compand = {
> + .name = "compand",
> + .description = NULL_IF_CONFIG_SMALL("Compress or expand the dynamic range of the audio."),
> + .query_formats = query_formats,
> + .priv_size = sizeof(CompandContext),
> + .priv_class = &compand_class,
> + .init = init,
> + .uninit = uninit,
> + .inputs = compand_inputs,
> + .outputs = compand_outputs,
> +};
No timeline?
[...]
--
FFmpeg = Fanciful and Freak Merciless Patchable Ecumenical Game
More information about the ffmpeg-devel
mailing list