| Message ID | 20170501083020.5250-1-onemda@gmail.com |
|---|---|
| State | Superseded |
| Headers | show |
On Mon, May 1, 2017 at 3:30 PM, Paul B Mahol <onemda@gmail.com> wrote: > Signed-off-by: Paul B Mahol <onemda@gmail.com> > --- > configure | 2 + > doc/filters.texi | 10 ++ > libavfilter/Makefile | 1 + > libavfilter/af_afirfilter.c | 409 ++++++++++++++++++++++++++++++++++++++++++++ > libavfilter/allfilters.c | 1 + > 5 files changed, 423 insertions(+) > create mode 100644 libavfilter/af_afirfilter.c > > diff --git a/configure b/configure > index b3cb5b0..7fc7af4 100755 > --- a/configure > +++ b/configure > @@ -3078,6 +3078,8 @@ unix_protocol_select="network" > # filters > afftfilt_filter_deps="avcodec" > afftfilt_filter_select="fft" > +afirfilter_filter_deps="avcodec" > +afirfilter_filter_select="fft" > amovie_filter_deps="avcodec avformat" > aresample_filter_deps="swresample" > ass_filter_deps="libass" > diff --git a/doc/filters.texi b/doc/filters.texi > index 119e747..ea343d1 100644 > --- a/doc/filters.texi > +++ b/doc/filters.texi > @@ -878,6 +878,16 @@ afftfilt="1-clip((b/nb)*b,0,1)" > @end example > @end itemize > > +@section afirfilter > + > +Apply an Arbitary Frequency Impulse Response filter. > + > +This filter uses second stream as FIR coefficients. > +If second stream holds single channel, it will be used > +for all input channels in first stream, otherwise > +number of channels in second stream must be same as > +number of channels in first stream. > + > @anchor{aformat} > @section aformat > > diff --git a/libavfilter/Makefile b/libavfilter/Makefile > index 66c36e4..1a0f24b 100644 > --- a/libavfilter/Makefile > +++ b/libavfilter/Makefile > @@ -38,6 +38,7 @@ OBJS-$(CONFIG_AEMPHASIS_FILTER) += af_aemphasis.o > OBJS-$(CONFIG_AEVAL_FILTER) += aeval.o > OBJS-$(CONFIG_AFADE_FILTER) += af_afade.o > OBJS-$(CONFIG_AFFTFILT_FILTER) += af_afftfilt.o window_func.o > +OBJS-$(CONFIG_AFIRFILTER_FILTER) += af_afirfilter.o > OBJS-$(CONFIG_AFORMAT_FILTER) += af_aformat.o > OBJS-$(CONFIG_AGATE_FILTER) += af_agate.o > OBJS-$(CONFIG_AINTERLEAVE_FILTER) += f_interleave.o > diff --git a/libavfilter/af_afirfilter.c b/libavfilter/af_afirfilter.c > new file mode 100644 > index 0000000..ef2488a > --- /dev/null > +++ b/libavfilter/af_afirfilter.c > @@ -0,0 +1,409 @@ > +/* > + * Copyright (c) 2017 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 > + */ > + > +/** > + * @file > + * An arbitrary audio FIR filter > + */ > + > +#include "libavutil/audio_fifo.h" > +#include "libavutil/avassert.h" > +#include "libavutil/channel_layout.h" > +#include "libavutil/common.h" > +#include "libavutil/opt.h" > +#include "libavcodec/avfft.h" > + > +#include "audio.h" > +#include "avfilter.h" > +#include "formats.h" > +#include "internal.h" > + > +typedef struct FIRContext { > + const AVClass *class; > + > + int n; > + int eof_coeffs; > + int have_coeffs; > + int nb_taps; > + int fft_length; > + int nb_channels; > + int one2many; > + > + FFTContext *fft, *ifft; > + FFTComplex **fft_data; > + FFTComplex **fft_coef; Probably you may use rdft for performance reason. > + > + AVAudioFifo *fifo[2]; > + AVFrame *in[2]; > + AVFrame *buffer; > + int64_t pts; > + int hop_size; > + int start, end; > +} FIRContext; > + > +static int fir_filter(FIRContext *s, AVFilterLink *outlink) > +{ > + AVFilterContext *ctx = outlink->src; > + int start = s->start, end = s->end; > + int ret = 0, n, ch, j, k; > + int nb_samples; > + AVFrame *out; > + > + nb_samples = FFMIN(s->fft_length, av_audio_fifo_size(s->fifo[0])); > + > + s->in[0] = ff_get_audio_buffer(ctx->inputs[0], nb_samples); > + if (!s->in[0]) > + return AVERROR(ENOMEM); > + > + av_audio_fifo_peek(s->fifo[0], (void **)s->in[0]->extended_data, nb_samples); > + > + for (ch = 0; ch < outlink->channels; ch++) { > + const float *src = (float *)s->in[0]->extended_data[ch]; > + float *buf = (float *)s->buffer->extended_data[ch]; > + FFTComplex *fft_data = s->fft_data[ch]; > + FFTComplex *fft_coef = s->fft_coef[ch]; > + > + memset(fft_data, 0, sizeof(*fft_data) * s->fft_length); > + for (n = 0; n < nb_samples; n++) { > + fft_data[n].re = src[n]; > + fft_data[n].im = 0; > + } > + > + av_fft_permute(s->fft, fft_data); > + av_fft_calc(s->fft, fft_data); > + > + fft_data[0].re *= fft_coef[0].re; > + fft_data[0].im *= fft_coef[0].im; > + for (n = 1; n < s->fft_length; n++) { > + const float re = fft_data[n].re; > + const float im = fft_data[n].im; > + > + fft_data[n].re = re * fft_coef[n].re - im * fft_coef[n].im; > + fft_data[n].im = re * fft_coef[n].im + im * fft_coef[n].re; > + } > + > + av_fft_permute(s->ifft, fft_data); > + av_fft_calc(s->ifft, fft_data); > + > + start = s->start; > + end = s->end; > + k = end; > + > + for (n = 0, j = start; j < k && n < s->fft_length; n++, j++) { > + buf[j] = fft_data[n].re; > + } > + > + for (; n < s->fft_length; n++, j++) { > + buf[j] = fft_data[n].re; > + } > + > + start += s->hop_size; > + end = j; > + } > + > + s->start = start; > + s->end = end; > + > + if (start >= nb_samples) { > + float *dst, *buf; > + > + start -= nb_samples; > + end -= nb_samples; > + > + s->start = start; > + s->end = end; > + > + out = ff_get_audio_buffer(outlink, nb_samples); > + if (!out) > + return AVERROR(ENOMEM); > + > + out->pts = s->pts; > + s->pts += nb_samples; Is pts handled correctly here? Seem it is not derived from input pts. > + > + for (ch = 0; ch < s->nb_channels; ch++) { > + dst = (float *)out->extended_data[ch]; > + buf = (float *)s->buffer->extended_data[ch]; > + > + for (n = 0; n < nb_samples; n++) > + dst[n] = buf[n]; > + memmove(buf, buf + nb_samples, nb_samples * 4); > + } > + > + ret = ff_filter_frame(outlink, out); > + } > + > + av_audio_fifo_drain(s->fifo[0], FFMIN(nb_samples, s->hop_size)); > + av_frame_free(&s->in[0]); > + > + return ret; > +} > + > +static int convert_coeffs(AVFilterContext *ctx) > +{ > + FIRContext *s = ctx->priv; > + int ch, n; > + > + s->nb_taps = av_audio_fifo_size(s->fifo[1]); > + if (s->nb_taps > 32768) { > + av_log(ctx, AV_LOG_ERROR, "Too big number of taps: %d > 32768.\n", s->nb_taps); > + return AVERROR(EINVAL); > + } > + > + for (n = 1; (1 << n) < s->nb_taps; n++); > + s->n = n + 2; > + s->fft_length = 1 << s->n; > + > + for (ch = 0; ch < ctx->inputs[0]->channels; ch++) { > + s->fft_data[ch] = av_calloc(s->fft_length, sizeof(**s->fft_data)); > + if (!s->fft_data[ch]) > + return AVERROR(ENOMEM); > + } > + > + for (ch = 0; ch < ctx->inputs[1]->channels; ch++) { > + s->fft_coef[ch] = av_calloc(s->fft_length, sizeof(**s->fft_coef)); > + if (!s->fft_coef[ch]) > + return AVERROR(ENOMEM); > + } > + > + s->hop_size = s->fft_length - s->nb_taps + 1; > + if (s->hop_size <= 0) { > + av_log(ctx, AV_LOG_ERROR, "Too big number of taps.\n"); > + return AVERROR(EINVAL); > + } > + > + s->buffer = ff_get_audio_buffer(ctx->inputs[0], s->fft_length * 2); > + if (!s->buffer) > + return AVERROR(ENOMEM); > + > + s->fft = av_fft_init(s->n, 0); > + s->ifft = av_fft_init(s->n, 1); > + if (!s->fft || !s->ifft) > + return AVERROR(ENOMEM); > + > + s->in[1] = ff_get_audio_buffer(ctx->inputs[1], s->nb_taps); > + if (!s->in[1]) > + return AVERROR(ENOMEM); > + > + av_audio_fifo_read(s->fifo[1], (void **)s->in[1]->extended_data, s->nb_taps); > + for (ch = 0; ch < ctx->inputs[1]->channels; ch++) { > + FFTComplex *fft_coef = s->fft_coef[ch]; > + const float *re = (const float *)s->in[1]->extended_data[!s->one2many * ch]; > + const float scale = 1.f / s->fft_length; > + const int offset = (s->fft_length - s->nb_taps); > + > + memset(fft_coef, 0, sizeof(*fft_coef) * s->fft_length); > + for (n = 0; n < s->nb_taps; n++) { > + fft_coef[n + offset].re = re[n] * scale; > + } > + av_fft_permute(s->fft, fft_coef); > + av_fft_calc(s->fft, fft_coef); > + } > + > + av_frame_free(&s->in[1]); > + s->have_coeffs = 1; > + > + return 0; > +} > + > +static int read_coeffs(AVFilterLink *link, AVFrame *frame) > +{ > + AVFilterContext *ctx = link->dst; > + FIRContext *s = ctx->priv; > + > + av_audio_fifo_write(s->fifo[1], (void **)frame->extended_data, > + frame->nb_samples); > + av_frame_free(&frame); > + > + return 0; > +} > + > +static int filter_frame(AVFilterLink *link, AVFrame *frame) > +{ > + AVFilterContext *ctx = link->dst; > + FIRContext *s = ctx->priv; > + AVFilterLink *outlink = ctx->outputs[0]; > + int ret = 0; > + > + av_audio_fifo_write(s->fifo[0], (void **)frame->extended_data, > + frame->nb_samples); > + av_frame_free(&frame); > + > + if (!s->have_coeffs && s->eof_coeffs) { > + ret = convert_coeffs(ctx); > + if (ret < 0) > + return ret; > + } > + > + if (s->have_coeffs) { > + while (av_audio_fifo_size(s->fifo[0]) >= s->fft_length) { > + ret = fir_filter(s, outlink); > + if (ret < 0) > + break; > + } > + } > + return ret; > +} > + > +static int request_frame(AVFilterLink *outlink) > +{ > + AVFilterContext *ctx = outlink->src; > + FIRContext *s = ctx->priv; > + int ret; > + > + if (!s->eof_coeffs) { > + ret = ff_request_frame(ctx->inputs[1]); > + if (ret == AVERROR_EOF) { > + s->eof_coeffs = 1; > + ret = 0; > + } > + return ret; > + } > + ret = ff_request_frame(ctx->inputs[0]); > + if (ret == AVERROR_EOF && s->have_coeffs) { > + while (av_audio_fifo_size(s->fifo[0]) > 0) { > + ret = fir_filter(s, outlink); > + if (ret < 0) > + return ret; > + } > + ret = AVERROR_EOF; > + } > + return ret; > +} > + > +static int query_formats(AVFilterContext *ctx) > +{ > + AVFilterFormats *formats; > + AVFilterChannelLayouts *layouts = NULL; > + static const enum AVSampleFormat sample_fmts[] = { > + AV_SAMPLE_FMT_FLTP, > + AV_SAMPLE_FMT_NONE > + }; > + int ret, i; > + > + layouts = ff_all_channel_counts(); > + if ((ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts)) < 0) > + return ret; > + > + for (i = 0; i < 2; i++) { > + layouts = ff_all_channel_counts(); > + if ((ret = ff_channel_layouts_ref(layouts, &ctx->inputs[i]->out_channel_layouts)) < 0) > + return ret; > + } > + > + formats = ff_make_format_list(sample_fmts); > + if ((ret = ff_set_common_formats(ctx, formats)) < 0) > + return ret; > + > + formats = ff_all_samplerates(); > + return ff_set_common_samplerates(ctx, formats); > +} > + > +static int config_output(AVFilterLink *outlink) > +{ > + AVFilterContext *ctx = outlink->src; > + FIRContext *s = ctx->priv; > + > + if (ctx->inputs[0]->channels != ctx->inputs[1]->channels && > + ctx->inputs[1]->channels != 1) { > + av_log(ctx, AV_LOG_ERROR, > + "Second input must have same number of channels as first input or " > + "exactly 1 channel.\n"); > + return AVERROR(EINVAL); > + } > + > + s->one2many = ctx->inputs[1]->channels == 1; > + outlink->sample_rate = ctx->inputs[0]->sample_rate; > + outlink->time_base = ctx->inputs[0]->time_base; > + outlink->channel_layout = ctx->inputs[0]->channel_layout; > + outlink->channels = ctx->inputs[0]->channels; > + > + s->fifo[0] = av_audio_fifo_alloc(ctx->inputs[0]->format, ctx->inputs[0]->channels, 1024); > + s->fifo[1] = av_audio_fifo_alloc(ctx->inputs[1]->format, ctx->inputs[1]->channels, 1024); > + if (!s->fifo[0] || !s->fifo[1]) > + return AVERROR(ENOMEM); > + > + s->fft_data = av_calloc(outlink->channels, sizeof(*s->fft_data)); > + s->fft_coef = av_calloc(ctx->inputs[1]->channels, sizeof(*s->fft_coef)); > + if (!s->fft_data || !s->fft_coef) > + return AVERROR(ENOMEM); > + s->nb_channels = outlink->channels; > + > + return 0; > +} > + > +static av_cold void uninit(AVFilterContext *ctx) > +{ > + FIRContext *s = ctx->priv; > + int ch; > + > + for (ch = 0; ch < s->nb_channels; ch++) { > + if (s->fft_data) > + av_freep(&s->fft_data[ch]); > + } > + av_freep(&s->fft_data); > + > + for (ch = 0; ch < s->nb_channels; ch++) { > + if (s->fft_coef) > + av_freep(&s->fft_coef[ch]); > + } > + av_freep(&s->fft_coef); > + > + av_fft_end(s->fft); > + av_fft_end(s->ifft); > + > + av_frame_free(&s->in[0]); > + av_frame_free(&s->in[1]); > + > + av_audio_fifo_free(s->fifo[0]); > + av_audio_fifo_free(s->fifo[1]); > +} > + > +static const AVFilterPad afirfilter_inputs[] = { > + { > + .name = "main", > + .type = AVMEDIA_TYPE_AUDIO, > + .filter_frame = filter_frame, > + },{ > + .name = "coefficients", > + .type = AVMEDIA_TYPE_AUDIO, > + .filter_frame = read_coeffs, > + }, > + { NULL } > +}; > + > +static const AVFilterPad afirfilter_outputs[] = { > + { > + .name = "default", > + .type = AVMEDIA_TYPE_AUDIO, > + .config_props = config_output, > + .request_frame = request_frame, > + }, > + { NULL } > +}; > + > +AVFilter ff_af_afirfilter = { > + .name = "afirfilter", > + .description = NULL_IF_CONFIG_SMALL("Apply Finite Impulse Response filter with supplied coefficients in 2nd stream."), > + .priv_size = sizeof(FIRContext), > + .query_formats = query_formats, > + .uninit = uninit, > + .inputs = afirfilter_inputs, > + .outputs = afirfilter_outputs, > +}; > diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c > index 8fb87eb..8bfe1ae 100644 > --- a/libavfilter/allfilters.c > +++ b/libavfilter/allfilters.c > @@ -50,6 +50,7 @@ static void register_all(void) > REGISTER_FILTER(AEVAL, aeval, af); > REGISTER_FILTER(AFADE, afade, af); > REGISTER_FILTER(AFFTFILT, afftfilt, af); > + REGISTER_FILTER(AFIRFILTER, afirfilter, af); > REGISTER_FILTER(AFORMAT, aformat, af); > REGISTER_FILTER(AGATE, agate, af); > REGISTER_FILTER(AINTERLEAVE, ainterleave, af); > -- > 2.9.3 > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
On 5/2/17, Muhammad Faiz <mfcc64@gmail.com> wrote: > On Mon, May 1, 2017 at 3:30 PM, Paul B Mahol <onemda@gmail.com> wrote: >> Signed-off-by: Paul B Mahol <onemda@gmail.com> >> --- >> configure | 2 + >> doc/filters.texi | 10 ++ >> libavfilter/Makefile | 1 + >> libavfilter/af_afirfilter.c | 409 >> ++++++++++++++++++++++++++++++++++++++++++++ >> libavfilter/allfilters.c | 1 + >> 5 files changed, 423 insertions(+) >> create mode 100644 libavfilter/af_afirfilter.c >> >> diff --git a/configure b/configure >> index b3cb5b0..7fc7af4 100755 >> --- a/configure >> +++ b/configure >> @@ -3078,6 +3078,8 @@ unix_protocol_select="network" >> # filters >> afftfilt_filter_deps="avcodec" >> afftfilt_filter_select="fft" >> +afirfilter_filter_deps="avcodec" >> +afirfilter_filter_select="fft" >> amovie_filter_deps="avcodec avformat" >> aresample_filter_deps="swresample" >> ass_filter_deps="libass" >> diff --git a/doc/filters.texi b/doc/filters.texi >> index 119e747..ea343d1 100644 >> --- a/doc/filters.texi >> +++ b/doc/filters.texi >> @@ -878,6 +878,16 @@ afftfilt="1-clip((b/nb)*b,0,1)" >> @end example >> @end itemize >> >> +@section afirfilter >> + >> +Apply an Arbitary Frequency Impulse Response filter. >> + >> +This filter uses second stream as FIR coefficients. >> +If second stream holds single channel, it will be used >> +for all input channels in first stream, otherwise >> +number of channels in second stream must be same as >> +number of channels in first stream. >> + >> @anchor{aformat} >> @section aformat >> >> diff --git a/libavfilter/Makefile b/libavfilter/Makefile >> index 66c36e4..1a0f24b 100644 >> --- a/libavfilter/Makefile >> +++ b/libavfilter/Makefile >> @@ -38,6 +38,7 @@ OBJS-$(CONFIG_AEMPHASIS_FILTER) += >> af_aemphasis.o >> OBJS-$(CONFIG_AEVAL_FILTER) += aeval.o >> OBJS-$(CONFIG_AFADE_FILTER) += af_afade.o >> OBJS-$(CONFIG_AFFTFILT_FILTER) += af_afftfilt.o >> window_func.o >> +OBJS-$(CONFIG_AFIRFILTER_FILTER) += af_afirfilter.o >> OBJS-$(CONFIG_AFORMAT_FILTER) += af_aformat.o >> OBJS-$(CONFIG_AGATE_FILTER) += af_agate.o >> OBJS-$(CONFIG_AINTERLEAVE_FILTER) += f_interleave.o >> diff --git a/libavfilter/af_afirfilter.c b/libavfilter/af_afirfilter.c >> new file mode 100644 >> index 0000000..ef2488a >> --- /dev/null >> +++ b/libavfilter/af_afirfilter.c >> @@ -0,0 +1,409 @@ >> +/* >> + * Copyright (c) 2017 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 >> + */ >> + >> +/** >> + * @file >> + * An arbitrary audio FIR filter >> + */ >> + >> +#include "libavutil/audio_fifo.h" >> +#include "libavutil/avassert.h" >> +#include "libavutil/channel_layout.h" >> +#include "libavutil/common.h" >> +#include "libavutil/opt.h" >> +#include "libavcodec/avfft.h" >> + >> +#include "audio.h" >> +#include "avfilter.h" >> +#include "formats.h" >> +#include "internal.h" >> + >> +typedef struct FIRContext { >> + const AVClass *class; >> + >> + int n; >> + int eof_coeffs; >> + int have_coeffs; >> + int nb_taps; >> + int fft_length; >> + int nb_channels; >> + int one2many; >> + >> + FFTContext *fft, *ifft; >> + FFTComplex **fft_data; >> + FFTComplex **fft_coef; > > Probably you may use rdft for performance reason. I will concentrate on correctness of output first. > > > >> + >> + AVAudioFifo *fifo[2]; >> + AVFrame *in[2]; >> + AVFrame *buffer; >> + int64_t pts; >> + int hop_size; >> + int start, end; >> +} FIRContext; >> + >> +static int fir_filter(FIRContext *s, AVFilterLink *outlink) >> +{ >> + AVFilterContext *ctx = outlink->src; >> + int start = s->start, end = s->end; >> + int ret = 0, n, ch, j, k; >> + int nb_samples; >> + AVFrame *out; >> + >> + nb_samples = FFMIN(s->fft_length, av_audio_fifo_size(s->fifo[0])); >> + >> + s->in[0] = ff_get_audio_buffer(ctx->inputs[0], nb_samples); >> + if (!s->in[0]) >> + return AVERROR(ENOMEM); >> + >> + av_audio_fifo_peek(s->fifo[0], (void **)s->in[0]->extended_data, >> nb_samples); >> + >> + for (ch = 0; ch < outlink->channels; ch++) { >> + const float *src = (float *)s->in[0]->extended_data[ch]; >> + float *buf = (float *)s->buffer->extended_data[ch]; >> + FFTComplex *fft_data = s->fft_data[ch]; >> + FFTComplex *fft_coef = s->fft_coef[ch]; >> + >> + memset(fft_data, 0, sizeof(*fft_data) * s->fft_length); >> + for (n = 0; n < nb_samples; n++) { >> + fft_data[n].re = src[n]; >> + fft_data[n].im = 0; >> + } >> + >> + av_fft_permute(s->fft, fft_data); >> + av_fft_calc(s->fft, fft_data); >> + >> + fft_data[0].re *= fft_coef[0].re; >> + fft_data[0].im *= fft_coef[0].im; >> + for (n = 1; n < s->fft_length; n++) { >> + const float re = fft_data[n].re; >> + const float im = fft_data[n].im; >> + >> + fft_data[n].re = re * fft_coef[n].re - im * fft_coef[n].im; >> + fft_data[n].im = re * fft_coef[n].im + im * fft_coef[n].re; >> + } >> + >> + av_fft_permute(s->ifft, fft_data); >> + av_fft_calc(s->ifft, fft_data); >> + >> + start = s->start; >> + end = s->end; >> + k = end; >> + >> + for (n = 0, j = start; j < k && n < s->fft_length; n++, j++) { >> + buf[j] = fft_data[n].re; >> + } >> + >> + for (; n < s->fft_length; n++, j++) { >> + buf[j] = fft_data[n].re; >> + } >> + >> + start += s->hop_size; >> + end = j; >> + } >> + >> + s->start = start; >> + s->end = end; >> + >> + if (start >= nb_samples) { >> + float *dst, *buf; >> + >> + start -= nb_samples; >> + end -= nb_samples; >> + >> + s->start = start; >> + s->end = end; >> + >> + out = ff_get_audio_buffer(outlink, nb_samples); >> + if (!out) >> + return AVERROR(ENOMEM); >> + >> + out->pts = s->pts; >> + s->pts += nb_samples; > > Is pts handled correctly here? Seem it is not derived from input pts. > It can not be derived in any other way.
On Wed, May 3, 2017 at 1:47 AM, Paul B Mahol <onemda@gmail.com> wrote: > On 5/2/17, Muhammad Faiz <mfcc64@gmail.com> wrote: >> On Mon, May 1, 2017 at 3:30 PM, Paul B Mahol <onemda@gmail.com> wrote: >>> Signed-off-by: Paul B Mahol <onemda@gmail.com> >>> --- >>> configure | 2 + >>> doc/filters.texi | 10 ++ >>> libavfilter/Makefile | 1 + >>> libavfilter/af_afirfilter.c | 409 >>> ++++++++++++++++++++++++++++++++++++++++++++ >>> libavfilter/allfilters.c | 1 + >>> 5 files changed, 423 insertions(+) >>> create mode 100644 libavfilter/af_afirfilter.c >>> >>> diff --git a/configure b/configure >>> index b3cb5b0..7fc7af4 100755 >>> --- a/configure >>> +++ b/configure >>> @@ -3078,6 +3078,8 @@ unix_protocol_select="network" >>> # filters >>> afftfilt_filter_deps="avcodec" >>> afftfilt_filter_select="fft" >>> +afirfilter_filter_deps="avcodec" >>> +afirfilter_filter_select="fft" >>> amovie_filter_deps="avcodec avformat" >>> aresample_filter_deps="swresample" >>> ass_filter_deps="libass" >>> diff --git a/doc/filters.texi b/doc/filters.texi >>> index 119e747..ea343d1 100644 >>> --- a/doc/filters.texi >>> +++ b/doc/filters.texi >>> @@ -878,6 +878,16 @@ afftfilt="1-clip((b/nb)*b,0,1)" >>> @end example >>> @end itemize >>> >>> +@section afirfilter >>> + >>> +Apply an Arbitary Frequency Impulse Response filter. >>> + >>> +This filter uses second stream as FIR coefficients. >>> +If second stream holds single channel, it will be used >>> +for all input channels in first stream, otherwise >>> +number of channels in second stream must be same as >>> +number of channels in first stream. >>> + >>> @anchor{aformat} >>> @section aformat >>> >>> diff --git a/libavfilter/Makefile b/libavfilter/Makefile >>> index 66c36e4..1a0f24b 100644 >>> --- a/libavfilter/Makefile >>> +++ b/libavfilter/Makefile >>> @@ -38,6 +38,7 @@ OBJS-$(CONFIG_AEMPHASIS_FILTER) += >>> af_aemphasis.o >>> OBJS-$(CONFIG_AEVAL_FILTER) += aeval.o >>> OBJS-$(CONFIG_AFADE_FILTER) += af_afade.o >>> OBJS-$(CONFIG_AFFTFILT_FILTER) += af_afftfilt.o >>> window_func.o >>> +OBJS-$(CONFIG_AFIRFILTER_FILTER) += af_afirfilter.o >>> OBJS-$(CONFIG_AFORMAT_FILTER) += af_aformat.o >>> OBJS-$(CONFIG_AGATE_FILTER) += af_agate.o >>> OBJS-$(CONFIG_AINTERLEAVE_FILTER) += f_interleave.o >>> diff --git a/libavfilter/af_afirfilter.c b/libavfilter/af_afirfilter.c >>> new file mode 100644 >>> index 0000000..ef2488a >>> --- /dev/null >>> +++ b/libavfilter/af_afirfilter.c >>> @@ -0,0 +1,409 @@ >>> +/* >>> + * Copyright (c) 2017 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 >>> + */ >>> + >>> +/** >>> + * @file >>> + * An arbitrary audio FIR filter >>> + */ >>> + >>> +#include "libavutil/audio_fifo.h" >>> +#include "libavutil/avassert.h" >>> +#include "libavutil/channel_layout.h" >>> +#include "libavutil/common.h" >>> +#include "libavutil/opt.h" >>> +#include "libavcodec/avfft.h" >>> + >>> +#include "audio.h" >>> +#include "avfilter.h" >>> +#include "formats.h" >>> +#include "internal.h" >>> + >>> +typedef struct FIRContext { >>> + const AVClass *class; >>> + >>> + int n; >>> + int eof_coeffs; >>> + int have_coeffs; >>> + int nb_taps; >>> + int fft_length; >>> + int nb_channels; >>> + int one2many; >>> + >>> + FFTContext *fft, *ifft; >>> + FFTComplex **fft_data; >>> + FFTComplex **fft_coef; >> >> Probably you may use rdft for performance reason. > > I will concentrate on correctness of output first. OK. > >> >> >> >>> + >>> + AVAudioFifo *fifo[2]; >>> + AVFrame *in[2]; >>> + AVFrame *buffer; >>> + int64_t pts; >>> + int hop_size; >>> + int start, end; >>> +} FIRContext; >>> + >>> +static int fir_filter(FIRContext *s, AVFilterLink *outlink) >>> +{ >>> + AVFilterContext *ctx = outlink->src; >>> + int start = s->start, end = s->end; >>> + int ret = 0, n, ch, j, k; >>> + int nb_samples; >>> + AVFrame *out; >>> + >>> + nb_samples = FFMIN(s->fft_length, av_audio_fifo_size(s->fifo[0])); >>> + >>> + s->in[0] = ff_get_audio_buffer(ctx->inputs[0], nb_samples); >>> + if (!s->in[0]) >>> + return AVERROR(ENOMEM); >>> + >>> + av_audio_fifo_peek(s->fifo[0], (void **)s->in[0]->extended_data, >>> nb_samples); >>> + >>> + for (ch = 0; ch < outlink->channels; ch++) { >>> + const float *src = (float *)s->in[0]->extended_data[ch]; >>> + float *buf = (float *)s->buffer->extended_data[ch]; >>> + FFTComplex *fft_data = s->fft_data[ch]; >>> + FFTComplex *fft_coef = s->fft_coef[ch]; >>> + >>> + memset(fft_data, 0, sizeof(*fft_data) * s->fft_length); >>> + for (n = 0; n < nb_samples; n++) { >>> + fft_data[n].re = src[n]; >>> + fft_data[n].im = 0; >>> + } >>> + >>> + av_fft_permute(s->fft, fft_data); >>> + av_fft_calc(s->fft, fft_data); >>> + >>> + fft_data[0].re *= fft_coef[0].re; >>> + fft_data[0].im *= fft_coef[0].im; >>> + for (n = 1; n < s->fft_length; n++) { >>> + const float re = fft_data[n].re; >>> + const float im = fft_data[n].im; >>> + >>> + fft_data[n].re = re * fft_coef[n].re - im * fft_coef[n].im; >>> + fft_data[n].im = re * fft_coef[n].im + im * fft_coef[n].re; >>> + } >>> + >>> + av_fft_permute(s->ifft, fft_data); >>> + av_fft_calc(s->ifft, fft_data); >>> + >>> + start = s->start; >>> + end = s->end; >>> + k = end; >>> + >>> + for (n = 0, j = start; j < k && n < s->fft_length; n++, j++) { >>> + buf[j] = fft_data[n].re; >>> + } >>> + >>> + for (; n < s->fft_length; n++, j++) { >>> + buf[j] = fft_data[n].re; >>> + } >>> + >>> + start += s->hop_size; >>> + end = j; >>> + } >>> + >>> + s->start = start; >>> + s->end = end; >>> + >>> + if (start >= nb_samples) { >>> + float *dst, *buf; >>> + >>> + start -= nb_samples; >>> + end -= nb_samples; >>> + >>> + s->start = start; >>> + s->end = end; >>> + >>> + out = ff_get_audio_buffer(outlink, nb_samples); >>> + if (!out) >>> + return AVERROR(ENOMEM); >>> + >>> + out->pts = s->pts; >>> + s->pts += nb_samples; >> >> Is pts handled correctly here? Seem it is not derived from input pts. >> > > It can not be derived in any other way. Probably, at least, first pts should be derived from input pts. Also, is time_base always 1/sample_rate? Thank's.
On Mon, May 01, 2017 at 10:30:20 +0200, Paul B Mahol wrote:
> + .name = "afirfilter",
Does a filter have any reason to have "filter" in its name? It seems
unusual.
Moritz
On Wed, May 3, 2017 at 4:12 PM, Muhammad Faiz <mfcc64@gmail.com> wrote: > On Wed, May 3, 2017 at 1:47 AM, Paul B Mahol <onemda@gmail.com> wrote: >> On 5/2/17, Muhammad Faiz <mfcc64@gmail.com> wrote: >>> On Mon, May 1, 2017 at 3:30 PM, Paul B Mahol <onemda@gmail.com> wrote: >>>> Signed-off-by: Paul B Mahol <onemda@gmail.com> >>>> --- >>>> configure | 2 + >>>> doc/filters.texi | 10 ++ >>>> libavfilter/Makefile | 1 + >>>> libavfilter/af_afirfilter.c | 409 >>>> ++++++++++++++++++++++++++++++++++++++++++++ >>>> libavfilter/allfilters.c | 1 + >>>> 5 files changed, 423 insertions(+) >>>> create mode 100644 libavfilter/af_afirfilter.c >>>> >>>> diff --git a/configure b/configure >>>> index b3cb5b0..7fc7af4 100755 >>>> --- a/configure >>>> +++ b/configure >>>> @@ -3078,6 +3078,8 @@ unix_protocol_select="network" >>>> # filters >>>> afftfilt_filter_deps="avcodec" >>>> afftfilt_filter_select="fft" >>>> +afirfilter_filter_deps="avcodec" >>>> +afirfilter_filter_select="fft" >>>> amovie_filter_deps="avcodec avformat" >>>> aresample_filter_deps="swresample" >>>> ass_filter_deps="libass" >>>> diff --git a/doc/filters.texi b/doc/filters.texi >>>> index 119e747..ea343d1 100644 >>>> --- a/doc/filters.texi >>>> +++ b/doc/filters.texi >>>> @@ -878,6 +878,16 @@ afftfilt="1-clip((b/nb)*b,0,1)" >>>> @end example >>>> @end itemize >>>> >>>> +@section afirfilter >>>> + >>>> +Apply an Arbitary Frequency Impulse Response filter. >>>> + >>>> +This filter uses second stream as FIR coefficients. >>>> +If second stream holds single channel, it will be used >>>> +for all input channels in first stream, otherwise >>>> +number of channels in second stream must be same as >>>> +number of channels in first stream. >>>> + >>>> @anchor{aformat} >>>> @section aformat >>>> >>>> diff --git a/libavfilter/Makefile b/libavfilter/Makefile >>>> index 66c36e4..1a0f24b 100644 >>>> --- a/libavfilter/Makefile >>>> +++ b/libavfilter/Makefile >>>> @@ -38,6 +38,7 @@ OBJS-$(CONFIG_AEMPHASIS_FILTER) += >>>> af_aemphasis.o >>>> OBJS-$(CONFIG_AEVAL_FILTER) += aeval.o >>>> OBJS-$(CONFIG_AFADE_FILTER) += af_afade.o >>>> OBJS-$(CONFIG_AFFTFILT_FILTER) += af_afftfilt.o >>>> window_func.o >>>> +OBJS-$(CONFIG_AFIRFILTER_FILTER) += af_afirfilter.o >>>> OBJS-$(CONFIG_AFORMAT_FILTER) += af_aformat.o >>>> OBJS-$(CONFIG_AGATE_FILTER) += af_agate.o >>>> OBJS-$(CONFIG_AINTERLEAVE_FILTER) += f_interleave.o >>>> diff --git a/libavfilter/af_afirfilter.c b/libavfilter/af_afirfilter.c >>>> new file mode 100644 >>>> index 0000000..ef2488a >>>> --- /dev/null >>>> +++ b/libavfilter/af_afirfilter.c >>>> @@ -0,0 +1,409 @@ >>>> +/* >>>> + * Copyright (c) 2017 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 >>>> + */ >>>> + >>>> +/** >>>> + * @file >>>> + * An arbitrary audio FIR filter >>>> + */ >>>> + >>>> +#include "libavutil/audio_fifo.h" >>>> +#include "libavutil/avassert.h" >>>> +#include "libavutil/channel_layout.h" >>>> +#include "libavutil/common.h" >>>> +#include "libavutil/opt.h" >>>> +#include "libavcodec/avfft.h" >>>> + >>>> +#include "audio.h" >>>> +#include "avfilter.h" >>>> +#include "formats.h" >>>> +#include "internal.h" >>>> + >>>> +typedef struct FIRContext { >>>> + const AVClass *class; >>>> + >>>> + int n; >>>> + int eof_coeffs; >>>> + int have_coeffs; >>>> + int nb_taps; >>>> + int fft_length; >>>> + int nb_channels; >>>> + int one2many; >>>> + >>>> + FFTContext *fft, *ifft; >>>> + FFTComplex **fft_data; >>>> + FFTComplex **fft_coef; >>> >>> Probably you may use rdft for performance reason. >> >> I will concentrate on correctness of output first. > > OK. > >> >>> >>> >>> >>>> + >>>> + AVAudioFifo *fifo[2]; >>>> + AVFrame *in[2]; >>>> + AVFrame *buffer; >>>> + int64_t pts; >>>> + int hop_size; >>>> + int start, end; >>>> +} FIRContext; >>>> + >>>> +static int fir_filter(FIRContext *s, AVFilterLink *outlink) >>>> +{ >>>> + AVFilterContext *ctx = outlink->src; >>>> + int start = s->start, end = s->end; >>>> + int ret = 0, n, ch, j, k; >>>> + int nb_samples; >>>> + AVFrame *out; >>>> + >>>> + nb_samples = FFMIN(s->fft_length, av_audio_fifo_size(s->fifo[0])); >>>> + >>>> + s->in[0] = ff_get_audio_buffer(ctx->inputs[0], nb_samples); >>>> + if (!s->in[0]) >>>> + return AVERROR(ENOMEM); >>>> + >>>> + av_audio_fifo_peek(s->fifo[0], (void **)s->in[0]->extended_data, >>>> nb_samples); >>>> + >>>> + for (ch = 0; ch < outlink->channels; ch++) { >>>> + const float *src = (float *)s->in[0]->extended_data[ch]; >>>> + float *buf = (float *)s->buffer->extended_data[ch]; >>>> + FFTComplex *fft_data = s->fft_data[ch]; >>>> + FFTComplex *fft_coef = s->fft_coef[ch]; >>>> + >>>> + memset(fft_data, 0, sizeof(*fft_data) * s->fft_length); >>>> + for (n = 0; n < nb_samples; n++) { >>>> + fft_data[n].re = src[n]; >>>> + fft_data[n].im = 0; >>>> + } >>>> + >>>> + av_fft_permute(s->fft, fft_data); >>>> + av_fft_calc(s->fft, fft_data); >>>> + >>>> + fft_data[0].re *= fft_coef[0].re; >>>> + fft_data[0].im *= fft_coef[0].im; >>>> + for (n = 1; n < s->fft_length; n++) { >>>> + const float re = fft_data[n].re; >>>> + const float im = fft_data[n].im; >>>> + >>>> + fft_data[n].re = re * fft_coef[n].re - im * fft_coef[n].im; >>>> + fft_data[n].im = re * fft_coef[n].im + im * fft_coef[n].re; >>>> + } >>>> + >>>> + av_fft_permute(s->ifft, fft_data); >>>> + av_fft_calc(s->ifft, fft_data); >>>> + >>>> + start = s->start; >>>> + end = s->end; >>>> + k = end; >>>> + >>>> + for (n = 0, j = start; j < k && n < s->fft_length; n++, j++) { >>>> + buf[j] = fft_data[n].re; >>>> + } >>>> + >>>> + for (; n < s->fft_length; n++, j++) { >>>> + buf[j] = fft_data[n].re; >>>> + } >>>> + >>>> + start += s->hop_size; >>>> + end = j; >>>> + } >>>> + >>>> + s->start = start; >>>> + s->end = end; >>>> + >>>> + if (start >= nb_samples) { >>>> + float *dst, *buf; >>>> + >>>> + start -= nb_samples; >>>> + end -= nb_samples; >>>> + >>>> + s->start = start; >>>> + s->end = end; >>>> + >>>> + out = ff_get_audio_buffer(outlink, nb_samples); >>>> + if (!out) >>>> + return AVERROR(ENOMEM); >>>> + >>>> + out->pts = s->pts; >>>> + s->pts += nb_samples; >>> >>> Is pts handled correctly here? Seem it is not derived from input pts. >>> >> >> It can not be derived in any other way. > > Probably, at least, first pts should be derived from input pts. > Also, is time_base always 1/sample_rate? > > Thank's. Probably, like in asetnsamples filter. Thank's.
On 5/3/17, Moritz Barsnick <barsnick@gmx.net> wrote: > On Mon, May 01, 2017 at 10:30:20 +0200, Paul B Mahol wrote: >> + .name = "afirfilter", > > Does a filter have any reason to have "filter" in its name? It seems > unusual. > Renamed.
On 5/5/17, Muhammad Faiz <mfcc64@gmail.com> wrote: >>>> Is pts handled correctly here? Seem it is not derived from input pts. >>>> >>> >>> It can not be derived in any other way. >> >> Probably, at least, first pts should be derived from input pts. >> Also, is time_base always 1/sample_rate? >> >> Thank's. > > Probably, like in asetnsamples filter. Done. Have an idea where artifacst come out for some IRs?
On Sat, May 6, 2017 at 2:33 AM, Paul B Mahol <onemda@gmail.com> wrote: > On 5/5/17, Muhammad Faiz <mfcc64@gmail.com> wrote: >>>>> Is pts handled correctly here? Seem it is not derived from input pts. >>>>> >>>> >>>> It can not be derived in any other way. >>> >>> Probably, at least, first pts should be derived from input pts. >>> Also, is time_base always 1/sample_rate? >>> >>> Thank's. >> >> Probably, like in asetnsamples filter. > > Done. Have an idea where artifacst come out for some IRs? I have no idea what's wrong here? I currently don't understand partitioned convolution. My test case: ./ffplay -f lavfi "aevalsrc='if(lt(t,1), if(n,0,1), sin(1000*t*t))', aformat= channel_layouts=mono, asplit [afir_in0], firequalizer= zero_phase=on:gain_entry='entry(0, 0); entry(100, -4); entry(1000, -30); entry(5000, -50); entry(6000, -10); entry(10000, -5)', aformat= channel_layouts=mono,a split [merge1], atrim= end_sample=883 [afir_in1]; [afir_in0][afir_in1] afir, aformat=channel_layouts=mono, [merge1] amerge, asplit[out0],showspectrum=s=1024x512[out1]" Note that your old patch didn't generate artifact. Thank's.
On 5/6/17, Muhammad Faiz <mfcc64@gmail.com> wrote: > On Sat, May 6, 2017 at 2:33 AM, Paul B Mahol <onemda@gmail.com> wrote: >> On 5/5/17, Muhammad Faiz <mfcc64@gmail.com> wrote: >>>>>> Is pts handled correctly here? Seem it is not derived from input pts. >>>>>> >>>>> >>>>> It can not be derived in any other way. >>>> >>>> Probably, at least, first pts should be derived from input pts. >>>> Also, is time_base always 1/sample_rate? >>>> >>>> Thank's. >>> >>> Probably, like in asetnsamples filter. >> >> Done. Have an idea where artifacst come out for some IRs? > > I have no idea what's wrong here? I currently don't understand > partitioned convolution. > > My test case: > ./ffplay -f lavfi "aevalsrc='if(lt(t,1), if(n,0,1), sin(1000*t*t))', > aformat= channel_layouts=mono, asplit [afir_in0], firequalizer= > zero_phase=on:gain_entry='entry(0, 0); entry(100, -4); entry(1000, > -30); entry(5000, -50); entry(6000, -10); entry(10000, -5)', aformat= > channel_layouts=mono,a > split [merge1], atrim= end_sample=883 [afir_in1]; [afir_in0][afir_in1] > afir, aformat=channel_layouts=mono, [merge1] amerge, > asplit[out0],showspectrum=s=1024x512[out1]" > > Note that your old patch didn't generate artifact. Artifacts are gone if you use bigger IR size, >2048. UPOLS should be rather trivial to understand and to implement.
On Sun, May 7, 2017 at 2:49 AM, Paul B Mahol <onemda@gmail.com> wrote: > On 5/6/17, Muhammad Faiz <mfcc64@gmail.com> wrote: >> On Sat, May 6, 2017 at 2:33 AM, Paul B Mahol <onemda@gmail.com> wrote: >>> On 5/5/17, Muhammad Faiz <mfcc64@gmail.com> wrote: >>>>>>> Is pts handled correctly here? Seem it is not derived from input pts. >>>>>>> >>>>>> >>>>>> It can not be derived in any other way. >>>>> >>>>> Probably, at least, first pts should be derived from input pts. >>>>> Also, is time_base always 1/sample_rate? >>>>> >>>>> Thank's. >>>> >>>> Probably, like in asetnsamples filter. >>> >>> Done. Have an idea where artifacst come out for some IRs? >> >> I have no idea what's wrong here? I currently don't understand >> partitioned convolution. >> >> My test case: >> ./ffplay -f lavfi "aevalsrc='if(lt(t,1), if(n,0,1), sin(1000*t*t))', >> aformat= channel_layouts=mono, asplit [afir_in0], firequalizer= >> zero_phase=on:gain_entry='entry(0, 0); entry(100, -4); entry(1000, >> -30); entry(5000, -50); entry(6000, -10); entry(10000, -5)', aformat= >> channel_layouts=mono,a >> split [merge1], atrim= end_sample=883 [afir_in1]; [afir_in0][afir_in1] >> afir, aformat=channel_layouts=mono, [merge1] amerge, >> asplit[out0],showspectrum=s=1024x512[out1]" >> >> Note that your old patch didn't generate artifact. > > Artifacts are gone if you use bigger IR size, >2048. > > UPOLS should be rather trivial to understand and to implement. Still happens at 8821 (44100 Hz with 0.1s delay): ./ffplay -f lavfi "aevalsrc='if(lt(t,1), if(n,0,1), sin(1000*t*t))', aformat=channel_layouts=mono, asplit [afir_in0], fir equalizer=zero_phase=on:gain_entry='entry(0, 0); entry(100, -4); entry(1000, -30); entry(5000, -50); entry(6000, -10); entry(10000, -5)':delay=0.1, aformat=channel_layo uts=mono,asplit [merge1], atrim=end_sample=8821 [afir_in1]; [afir_in0][afir_in1] afir, aformat=channel_layouts=mono, [merge1] amerge, asplit[out0],showspectrum=s=1024x5 12[out1]"
diff --git a/configure b/configure index b3cb5b0..7fc7af4 100755 --- a/configure +++ b/configure @@ -3078,6 +3078,8 @@ unix_protocol_select="network" # filters afftfilt_filter_deps="avcodec" afftfilt_filter_select="fft" +afirfilter_filter_deps="avcodec" +afirfilter_filter_select="fft" amovie_filter_deps="avcodec avformat" aresample_filter_deps="swresample" ass_filter_deps="libass" diff --git a/doc/filters.texi b/doc/filters.texi index 119e747..ea343d1 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -878,6 +878,16 @@ afftfilt="1-clip((b/nb)*b,0,1)" @end example @end itemize +@section afirfilter + +Apply an Arbitary Frequency Impulse Response filter. + +This filter uses second stream as FIR coefficients. +If second stream holds single channel, it will be used +for all input channels in first stream, otherwise +number of channels in second stream must be same as +number of channels in first stream. + @anchor{aformat} @section aformat diff --git a/libavfilter/Makefile b/libavfilter/Makefile index 66c36e4..1a0f24b 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -38,6 +38,7 @@ OBJS-$(CONFIG_AEMPHASIS_FILTER) += af_aemphasis.o OBJS-$(CONFIG_AEVAL_FILTER) += aeval.o OBJS-$(CONFIG_AFADE_FILTER) += af_afade.o OBJS-$(CONFIG_AFFTFILT_FILTER) += af_afftfilt.o window_func.o +OBJS-$(CONFIG_AFIRFILTER_FILTER) += af_afirfilter.o OBJS-$(CONFIG_AFORMAT_FILTER) += af_aformat.o OBJS-$(CONFIG_AGATE_FILTER) += af_agate.o OBJS-$(CONFIG_AINTERLEAVE_FILTER) += f_interleave.o diff --git a/libavfilter/af_afirfilter.c b/libavfilter/af_afirfilter.c new file mode 100644 index 0000000..ef2488a --- /dev/null +++ b/libavfilter/af_afirfilter.c @@ -0,0 +1,409 @@ +/* + * Copyright (c) 2017 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 + */ + +/** + * @file + * An arbitrary audio FIR filter + */ + +#include "libavutil/audio_fifo.h" +#include "libavutil/avassert.h" +#include "libavutil/channel_layout.h" +#include "libavutil/common.h" +#include "libavutil/opt.h" +#include "libavcodec/avfft.h" + +#include "audio.h" +#include "avfilter.h" +#include "formats.h" +#include "internal.h" + +typedef struct FIRContext { + const AVClass *class; + + int n; + int eof_coeffs; + int have_coeffs; + int nb_taps; + int fft_length; + int nb_channels; + int one2many; + + FFTContext *fft, *ifft; + FFTComplex **fft_data; + FFTComplex **fft_coef; + + AVAudioFifo *fifo[2]; + AVFrame *in[2]; + AVFrame *buffer; + int64_t pts; + int hop_size; + int start, end; +} FIRContext; + +static int fir_filter(FIRContext *s, AVFilterLink *outlink) +{ + AVFilterContext *ctx = outlink->src; + int start = s->start, end = s->end; + int ret = 0, n, ch, j, k; + int nb_samples; + AVFrame *out; + + nb_samples = FFMIN(s->fft_length, av_audio_fifo_size(s->fifo[0])); + + s->in[0] = ff_get_audio_buffer(ctx->inputs[0], nb_samples); + if (!s->in[0]) + return AVERROR(ENOMEM); + + av_audio_fifo_peek(s->fifo[0], (void **)s->in[0]->extended_data, nb_samples); + + for (ch = 0; ch < outlink->channels; ch++) { + const float *src = (float *)s->in[0]->extended_data[ch]; + float *buf = (float *)s->buffer->extended_data[ch]; + FFTComplex *fft_data = s->fft_data[ch]; + FFTComplex *fft_coef = s->fft_coef[ch]; + + memset(fft_data, 0, sizeof(*fft_data) * s->fft_length); + for (n = 0; n < nb_samples; n++) { + fft_data[n].re = src[n]; + fft_data[n].im = 0; + } + + av_fft_permute(s->fft, fft_data); + av_fft_calc(s->fft, fft_data); + + fft_data[0].re *= fft_coef[0].re; + fft_data[0].im *= fft_coef[0].im; + for (n = 1; n < s->fft_length; n++) { + const float re = fft_data[n].re; + const float im = fft_data[n].im; + + fft_data[n].re = re * fft_coef[n].re - im * fft_coef[n].im; + fft_data[n].im = re * fft_coef[n].im + im * fft_coef[n].re; + } + + av_fft_permute(s->ifft, fft_data); + av_fft_calc(s->ifft, fft_data); + + start = s->start; + end = s->end; + k = end; + + for (n = 0, j = start; j < k && n < s->fft_length; n++, j++) { + buf[j] = fft_data[n].re; + } + + for (; n < s->fft_length; n++, j++) { + buf[j] = fft_data[n].re; + } + + start += s->hop_size; + end = j; + } + + s->start = start; + s->end = end; + + if (start >= nb_samples) { + float *dst, *buf; + + start -= nb_samples; + end -= nb_samples; + + s->start = start; + s->end = end; + + out = ff_get_audio_buffer(outlink, nb_samples); + if (!out) + return AVERROR(ENOMEM); + + out->pts = s->pts; + s->pts += nb_samples; + + for (ch = 0; ch < s->nb_channels; ch++) { + dst = (float *)out->extended_data[ch]; + buf = (float *)s->buffer->extended_data[ch]; + + for (n = 0; n < nb_samples; n++) + dst[n] = buf[n]; + memmove(buf, buf + nb_samples, nb_samples * 4); + } + + ret = ff_filter_frame(outlink, out); + } + + av_audio_fifo_drain(s->fifo[0], FFMIN(nb_samples, s->hop_size)); + av_frame_free(&s->in[0]); + + return ret; +} + +static int convert_coeffs(AVFilterContext *ctx) +{ + FIRContext *s = ctx->priv; + int ch, n; + + s->nb_taps = av_audio_fifo_size(s->fifo[1]); + if (s->nb_taps > 32768) { + av_log(ctx, AV_LOG_ERROR, "Too big number of taps: %d > 32768.\n", s->nb_taps); + return AVERROR(EINVAL); + } + + for (n = 1; (1 << n) < s->nb_taps; n++); + s->n = n + 2; + s->fft_length = 1 << s->n; + + for (ch = 0; ch < ctx->inputs[0]->channels; ch++) { + s->fft_data[ch] = av_calloc(s->fft_length, sizeof(**s->fft_data)); + if (!s->fft_data[ch]) + return AVERROR(ENOMEM); + } + + for (ch = 0; ch < ctx->inputs[1]->channels; ch++) { + s->fft_coef[ch] = av_calloc(s->fft_length, sizeof(**s->fft_coef)); + if (!s->fft_coef[ch]) + return AVERROR(ENOMEM); + } + + s->hop_size = s->fft_length - s->nb_taps + 1; + if (s->hop_size <= 0) { + av_log(ctx, AV_LOG_ERROR, "Too big number of taps.\n"); + return AVERROR(EINVAL); + } + + s->buffer = ff_get_audio_buffer(ctx->inputs[0], s->fft_length * 2); + if (!s->buffer) + return AVERROR(ENOMEM); + + s->fft = av_fft_init(s->n, 0); + s->ifft = av_fft_init(s->n, 1); + if (!s->fft || !s->ifft) + return AVERROR(ENOMEM); + + s->in[1] = ff_get_audio_buffer(ctx->inputs[1], s->nb_taps); + if (!s->in[1]) + return AVERROR(ENOMEM); + + av_audio_fifo_read(s->fifo[1], (void **)s->in[1]->extended_data, s->nb_taps); + for (ch = 0; ch < ctx->inputs[1]->channels; ch++) { + FFTComplex *fft_coef = s->fft_coef[ch]; + const float *re = (const float *)s->in[1]->extended_data[!s->one2many * ch]; + const float scale = 1.f / s->fft_length; + const int offset = (s->fft_length - s->nb_taps); + + memset(fft_coef, 0, sizeof(*fft_coef) * s->fft_length); + for (n = 0; n < s->nb_taps; n++) { + fft_coef[n + offset].re = re[n] * scale; + } + av_fft_permute(s->fft, fft_coef); + av_fft_calc(s->fft, fft_coef); + } + + av_frame_free(&s->in[1]); + s->have_coeffs = 1; + + return 0; +} + +static int read_coeffs(AVFilterLink *link, AVFrame *frame) +{ + AVFilterContext *ctx = link->dst; + FIRContext *s = ctx->priv; + + av_audio_fifo_write(s->fifo[1], (void **)frame->extended_data, + frame->nb_samples); + av_frame_free(&frame); + + return 0; +} + +static int filter_frame(AVFilterLink *link, AVFrame *frame) +{ + AVFilterContext *ctx = link->dst; + FIRContext *s = ctx->priv; + AVFilterLink *outlink = ctx->outputs[0]; + int ret = 0; + + av_audio_fifo_write(s->fifo[0], (void **)frame->extended_data, + frame->nb_samples); + av_frame_free(&frame); + + if (!s->have_coeffs && s->eof_coeffs) { + ret = convert_coeffs(ctx); + if (ret < 0) + return ret; + } + + if (s->have_coeffs) { + while (av_audio_fifo_size(s->fifo[0]) >= s->fft_length) { + ret = fir_filter(s, outlink); + if (ret < 0) + break; + } + } + return ret; +} + +static int request_frame(AVFilterLink *outlink) +{ + AVFilterContext *ctx = outlink->src; + FIRContext *s = ctx->priv; + int ret; + + if (!s->eof_coeffs) { + ret = ff_request_frame(ctx->inputs[1]); + if (ret == AVERROR_EOF) { + s->eof_coeffs = 1; + ret = 0; + } + return ret; + } + ret = ff_request_frame(ctx->inputs[0]); + if (ret == AVERROR_EOF && s->have_coeffs) { + while (av_audio_fifo_size(s->fifo[0]) > 0) { + ret = fir_filter(s, outlink); + if (ret < 0) + return ret; + } + ret = AVERROR_EOF; + } + return ret; +} + +static int query_formats(AVFilterContext *ctx) +{ + AVFilterFormats *formats; + AVFilterChannelLayouts *layouts = NULL; + static const enum AVSampleFormat sample_fmts[] = { + AV_SAMPLE_FMT_FLTP, + AV_SAMPLE_FMT_NONE + }; + int ret, i; + + layouts = ff_all_channel_counts(); + if ((ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts)) < 0) + return ret; + + for (i = 0; i < 2; i++) { + layouts = ff_all_channel_counts(); + if ((ret = ff_channel_layouts_ref(layouts, &ctx->inputs[i]->out_channel_layouts)) < 0) + return ret; + } + + formats = ff_make_format_list(sample_fmts); + if ((ret = ff_set_common_formats(ctx, formats)) < 0) + return ret; + + formats = ff_all_samplerates(); + return ff_set_common_samplerates(ctx, formats); +} + +static int config_output(AVFilterLink *outlink) +{ + AVFilterContext *ctx = outlink->src; + FIRContext *s = ctx->priv; + + if (ctx->inputs[0]->channels != ctx->inputs[1]->channels && + ctx->inputs[1]->channels != 1) { + av_log(ctx, AV_LOG_ERROR, + "Second input must have same number of channels as first input or " + "exactly 1 channel.\n"); + return AVERROR(EINVAL); + } + + s->one2many = ctx->inputs[1]->channels == 1; + outlink->sample_rate = ctx->inputs[0]->sample_rate; + outlink->time_base = ctx->inputs[0]->time_base; + outlink->channel_layout = ctx->inputs[0]->channel_layout; + outlink->channels = ctx->inputs[0]->channels; + + s->fifo[0] = av_audio_fifo_alloc(ctx->inputs[0]->format, ctx->inputs[0]->channels, 1024); + s->fifo[1] = av_audio_fifo_alloc(ctx->inputs[1]->format, ctx->inputs[1]->channels, 1024); + if (!s->fifo[0] || !s->fifo[1]) + return AVERROR(ENOMEM); + + s->fft_data = av_calloc(outlink->channels, sizeof(*s->fft_data)); + s->fft_coef = av_calloc(ctx->inputs[1]->channels, sizeof(*s->fft_coef)); + if (!s->fft_data || !s->fft_coef) + return AVERROR(ENOMEM); + s->nb_channels = outlink->channels; + + return 0; +} + +static av_cold void uninit(AVFilterContext *ctx) +{ + FIRContext *s = ctx->priv; + int ch; + + for (ch = 0; ch < s->nb_channels; ch++) { + if (s->fft_data) + av_freep(&s->fft_data[ch]); + } + av_freep(&s->fft_data); + + for (ch = 0; ch < s->nb_channels; ch++) { + if (s->fft_coef) + av_freep(&s->fft_coef[ch]); + } + av_freep(&s->fft_coef); + + av_fft_end(s->fft); + av_fft_end(s->ifft); + + av_frame_free(&s->in[0]); + av_frame_free(&s->in[1]); + + av_audio_fifo_free(s->fifo[0]); + av_audio_fifo_free(s->fifo[1]); +} + +static const AVFilterPad afirfilter_inputs[] = { + { + .name = "main", + .type = AVMEDIA_TYPE_AUDIO, + .filter_frame = filter_frame, + },{ + .name = "coefficients", + .type = AVMEDIA_TYPE_AUDIO, + .filter_frame = read_coeffs, + }, + { NULL } +}; + +static const AVFilterPad afirfilter_outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_AUDIO, + .config_props = config_output, + .request_frame = request_frame, + }, + { NULL } +}; + +AVFilter ff_af_afirfilter = { + .name = "afirfilter", + .description = NULL_IF_CONFIG_SMALL("Apply Finite Impulse Response filter with supplied coefficients in 2nd stream."), + .priv_size = sizeof(FIRContext), + .query_formats = query_formats, + .uninit = uninit, + .inputs = afirfilter_inputs, + .outputs = afirfilter_outputs, +}; diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index 8fb87eb..8bfe1ae 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -50,6 +50,7 @@ static void register_all(void) REGISTER_FILTER(AEVAL, aeval, af); REGISTER_FILTER(AFADE, afade, af); REGISTER_FILTER(AFFTFILT, afftfilt, af); + REGISTER_FILTER(AFIRFILTER, afirfilter, af); REGISTER_FILTER(AFORMAT, aformat, af); REGISTER_FILTER(AGATE, agate, af); REGISTER_FILTER(AINTERLEAVE, ainterleave, af);
Signed-off-by: Paul B Mahol <onemda@gmail.com> --- configure | 2 + doc/filters.texi | 10 ++ libavfilter/Makefile | 1 + libavfilter/af_afirfilter.c | 409 ++++++++++++++++++++++++++++++++++++++++++++ libavfilter/allfilters.c | 1 + 5 files changed, 423 insertions(+) create mode 100644 libavfilter/af_afirfilter.c