From patchwork Thu Jun 8 06:13:48 2017 Content-Type: text/plain; charset="utf-8" MIME-Version: 1.0 Content-Transfer-Encoding: 7bit X-Patchwork-Submitter: Paul B Mahol X-Patchwork-Id: 3869 Delivered-To: ffmpegpatchwork@gmail.com Received: by 10.103.10.2 with SMTP id 2csp2708820vsk; Wed, 7 Jun 2017 23:22:24 -0700 (PDT) X-Received: by 10.28.210.13 with SMTP id j13mr2005087wmg.110.1496902944057; Wed, 07 Jun 2017 23:22:24 -0700 (PDT) ARC-Seal: i=1; a=rsa-sha256; t=1496902944; cv=none; d=google.com; s=arc-20160816; b=m8hL67FjXEvfOHdUD9rqqTPuRFjNobU2nnBP7MU4Nk2Gwk1+4Tjkp1UAKTBDLyMeM7 m5wMCPfa4iPWWXRFt4f/+WcM0lSoquawI0tsnBDA5Hq0/X2pQuPi7UkZ5ZHTidi1DL3h V2x8No0TJZBtnsf8V/neDyqi34V1yNi8B96dEoOOqFRs92dz9BwJpL/91o/np4C1Fz/d ALdfybWacjZEMt/ddjDPjzdiCu2fYyOjFrCJM7w9UZf23NEsevkWSWB5Ns+A35Nl4BKt BMC2sfCPeKUbuSiZabQIB0MTpRbOdcfQ8OggWlrSdKPL/KPIAnNWgPMDHRX0iG5Uz73Y WRDQ== ARC-Message-Signature: i=1; a=rsa-sha256; c=relaxed/relaxed; d=google.com; s=arc-20160816; h=sender:errors-to:content-transfer-encoding:mime-version:reply-to :list-subscribe:list-help:list-post:list-archive:list-unsubscribe :list-id:precedence:subject:message-id:date:to:from:dkim-signature :delivered-to:arc-authentication-results; bh=Guo5mSc/yRUIGkVuIblK6YGxYU+InjxAlclYJJsF5SU=; b=Ec0ddxxo/6xvI+WscB/z8bs9jOyRIwjXtx0cdFZvI8eUWXgJvPE52H0nhMnul/gpXk DD6v0sl0dJBs2QFqL7JLf+NWs2ciDaPV00+jBS4e8NEqPhmkUfMO5D9qNfpPJejDai5Q k2KpUzTvvbfP1awg+gdUXXSntzOH4tL5+ro7WMhL/F/1ICCfFRwqv0Wt+slNLQfxfUgO adPs4fiBeLZ2sTIDpSuv/qWUQNzxD4e5TNw2axTLu7bTgKLBd8qcNqyT+KgVK+THUmnn k9We+poNWFNZh/f7yUKv7soduYdXfWxCXzKTZRw9c+NsvPmG3+VI2sLDR1lWmNVPD/L9 Dlng== ARC-Authentication-Results: i=1; mx.google.com; dkim=neutral (body hash did not verify) header.i=@gmail.com; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=gmail.com Return-Path: Received: from ffbox0-bg.mplayerhq.hu (ffbox0-bg.ffmpeg.org. [79.124.17.100]) by mx.google.com with ESMTP id 193si4791723wmm.48.2017.06.07.23.22.23; Wed, 07 Jun 2017 23:22:24 -0700 (PDT) Received-SPF: pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) client-ip=79.124.17.100; Authentication-Results: mx.google.com; dkim=neutral (body hash did not verify) header.i=@gmail.com; spf=pass (google.com: domain of ffmpeg-devel-bounces@ffmpeg.org designates 79.124.17.100 as permitted sender) smtp.mailfrom=ffmpeg-devel-bounces@ffmpeg.org; dmarc=fail (p=NONE sp=NONE dis=NONE) header.from=gmail.com Received: from [127.0.1.1] (localhost [127.0.0.1]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTP id EDF5C689C9B; Thu, 8 Jun 2017 09:22:19 +0300 (EEST) X-Original-To: ffmpeg-devel@ffmpeg.org Delivered-To: ffmpeg-devel@ffmpeg.org Received: from mail-wr0-f195.google.com (mail-wr0-f195.google.com [209.85.128.195]) by ffbox0-bg.mplayerhq.hu (Postfix) with ESMTPS id E3645689B83 for ; Thu, 8 Jun 2017 09:22:13 +0300 (EEST) Received: by mail-wr0-f195.google.com with SMTP id u101so2861925wrc.1 for ; Wed, 07 Jun 2017 23:22:14 -0700 (PDT) DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=gmail.com; s=20161025; h=from:to:subject:date:message-id; bh=0AAaB/09WRZEAKMmkxnP7w+Y0FUYEnZNCpnrLUhKJow=; b=O6Lv6eRpeLAPCn6aNMVmQwdm61bxovhIyi+AIkX7P9G+uAUeiuAdnS9O43H0Eps7Ak LXHq/MmqByl5IQz7Zsqe07cclxBMlh54KO6bXVLpR0eDUKd0ZFS+i74YrA1kw5ZT4Jw2 dnsv26vR28jbJcpm+jKNm19MFj1+g4IcOLOUmk/fURq1S4B1g85jm8IkviVgO3stLQyx +ETCanlcU9O2rVoB4WYnH8xR4VOZpMstixOzl5eRVJ8oO0AsMuqtuDKexaFC2HzEXKXn FiD/oT//MWTcmaWZaBfdmAYV1NRcOpPixnt6hwX3GtuQpTAUcYBVBOtmdnYaBwT/cFbl afbw== X-Google-DKIM-Signature: v=1; a=rsa-sha256; c=relaxed/relaxed; d=1e100.net; s=20161025; h=x-gm-message-state:from:to:subject:date:message-id; bh=0AAaB/09WRZEAKMmkxnP7w+Y0FUYEnZNCpnrLUhKJow=; b=t9znT05rHzazoBG0DwKtpnKA+Cj+aGawRDMycYe6LSi+Ic5GEIXP2JFUQ2fQlW3Toz t2irYPM/rQ4Pe4IO/q9KDpz03vZwgc2CnUqG6oh67WjCrKv52JYlCmoe7VJEv/YBF5cf sYU3dlC0SBlNuWwGKCGIWNYQECmtCIXO8dOc1CqdEyDi8zikCvnAP+ZbaMIK5c5S59j7 vIp0Iiw7jf1vVm+LBd/OjuZPXYVcH8yqJy06Po+N29TF4ksWlYHerI7oBhty1TaA6oeV rTsMZeoFTaHBgLYYZyNkVOtzPqMOWRxkhSfATPYOLBU8zwUU8eItsjIjZtaMZthl6aMI vkcg== X-Gm-Message-State: AODbwcD14Vrfh0LJoOjffD3eWbuzR3A4TTxiXSUZHl5weeb0Tq8tjT+I iRO2hdECaDA3AmuW X-Received: by 10.223.147.33 with SMTP id 30mr23421027wro.103.1496902441427; Wed, 07 Jun 2017 23:14:01 -0700 (PDT) Received: from localhost.localdomain ([94.250.174.60]) by smtp.gmail.com with ESMTPSA id v13sm6012467wmd.5.2017.06.07.23.13.58 for (version=TLS1_2 cipher=ECDHE-RSA-AES128-GCM-SHA256 bits=128/128); Wed, 07 Jun 2017 23:14:00 -0700 (PDT) From: Paul B Mahol To: ffmpeg-devel@ffmpeg.org Date: Thu, 8 Jun 2017 08:13:48 +0200 Message-Id: <20170608061348.1043-1-onemda@gmail.com> X-Mailer: git-send-email 2.9.3 Subject: [FFmpeg-devel] [PATCH] avfilter: add native headphone spatialization filter X-BeenThere: ffmpeg-devel@ffmpeg.org X-Mailman-Version: 2.1.20 Precedence: list List-Id: FFmpeg development discussions and patches List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , Reply-To: FFmpeg development discussions and patches MIME-Version: 1.0 Errors-To: ffmpeg-devel-bounces@ffmpeg.org Sender: "ffmpeg-devel" Signed-off-by: Paul B Mahol --- doc/filters.texi | 41 +++ libavfilter/Makefile | 1 + libavfilter/af_headphone.c | 803 +++++++++++++++++++++++++++++++++++++++++++++ libavfilter/allfilters.c | 1 + 4 files changed, 846 insertions(+) create mode 100644 libavfilter/af_headphone.c diff --git a/doc/filters.texi b/doc/filters.texi index 65eef89..41fcb11 100644 --- a/doc/filters.texi +++ b/doc/filters.texi @@ -2789,6 +2789,47 @@ Samples where the target gain does not match between channels @end table @end table +@section headphone + +Apply head-related transfer functions (HRTFs) to create virtual +loudspeakers around the user for binaural listening via headphones. +The HRTFs are provided via additional streams. + +The filter accepts the following options: + +@table @option +@item map +Set mapping of input streams for convolution. +The argument is a '|'-separated list of channel names in order as they +are given as additional stream inputs for filter. +This also specify number of input streams. + +@item gain +Set gain applied to audio. Value is in dB. Default is 0. + +@item type +Set processing type. Can be @var{time} or @var{freq}. @var{time} is +processing audio in time domain which is slow. +@var{freq} is processing audio in frequency domain which is fast. +Default is @var{freq}. + +@item lfegain +Set custom gain for LFE channels. Value is in dB. Default is 0. +@end table + +@subsection Examples + +@itemize +@item +Full example using wav files as coefficients with amovie filters for 7.1 downmix, +each amovie filter use stereo file with IR coefficients as input. +The files give coefficients for each position of virtual loudspeaker: +@example +ffmpeg -i input.wav -lavfi-complex "amovie=azi_270_ele_0_DFC.wav[sr],amovie=azi_90_ele_0_DFC.wav[sl],amovie=azi_225_ele_0_DFC.wav[bl],amovie=azi_135_ele_0_DFC.wav[br],amovie=azi_0_ele_0_DFC.wav,asplit[fc][lfe],amovie=azi_35_ele_0_DFC.wav[fl],amovie=azi_325_ele_0_DFC.wav[fr],[a:0][fl][fr][fc][lfe][br][bl][sl][sr]headphone=FL|FR|FC|LFE|BL|BR|SL|SR" +output.wav +@end example +@end itemize + @section highpass Apply a high-pass filter with 3dB point frequency. diff --git a/libavfilter/Makefile b/libavfilter/Makefile index c88dfb3..04ec9b8 100644 --- a/libavfilter/Makefile +++ b/libavfilter/Makefile @@ -92,6 +92,7 @@ OBJS-$(CONFIG_EXTRASTEREO_FILTER) += af_extrastereo.o OBJS-$(CONFIG_FIREQUALIZER_FILTER) += af_firequalizer.o OBJS-$(CONFIG_FLANGER_FILTER) += af_flanger.o generate_wave_table.o OBJS-$(CONFIG_HDCD_FILTER) += af_hdcd.o +OBJS-$(CONFIG_HEADPHONE_FILTER) += af_headphone.o OBJS-$(CONFIG_HIGHPASS_FILTER) += af_biquads.o OBJS-$(CONFIG_JOIN_FILTER) += af_join.o OBJS-$(CONFIG_LADSPA_FILTER) += af_ladspa.o diff --git a/libavfilter/af_headphone.c b/libavfilter/af_headphone.c new file mode 100644 index 0000000..1fe6988 --- /dev/null +++ b/libavfilter/af_headphone.c @@ -0,0 +1,803 @@ +/* + * Copyright (C) 2017 Paul B Mahol + * Copyright (C) 2013-2015 Andreas Fuchs, Wolfgang Hrauda + * 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 + +#include "libavutil/audio_fifo.h" +#include "libavutil/avstring.h" +#include "libavutil/channel_layout.h" +#include "libavutil/float_dsp.h" +#include "libavutil/intmath.h" +#include "libavutil/opt.h" +#include "libavcodec/avfft.h" + +#include "avfilter.h" +#include "internal.h" +#include "audio.h" + +#define TIME_DOMAIN 0 +#define FREQUENCY_DOMAIN 1 + +typedef struct HeadphoneContext { + const AVClass *class; + + char *map; + int type; + + int lfe_channel; + + int have_coeffs; + int eof_coeffs; + int64_t pts; + + int ir_len; + + int mapping[64]; + + int nb_inputs; + + int nb_irs; + + float gain; + float lfe_gain, gain_lfe; + + float *ringbuffer[2]; + int write[2]; + + int buffer_length; + int n_fft; + int size; + + int *delay[2]; + float *data_ir[2]; + float *temp_src[2]; + FFTComplex *temp_fft[2]; + + FFTContext *fft[2], *ifft[2]; + FFTComplex *data_hrtf[2]; + + AVFloatDSPContext *fdsp; + struct headphone_inputs { + AVAudioFifo *fifo; + AVFrame *frame; + int ir_len; + int delay_l; + int delay_r; + int eof; + } *in; +} HeadphoneContext; + +static int parse_channel_name(char **arg, int *rchannel, char *buf) +{ + int len, i, channel_id = 0; + int64_t layout, layout0; + + if (sscanf(*arg, "%7[A-Z]%n", buf, &len)) { + layout0 = layout = av_get_channel_layout(buf); + for (i = 32; i > 0; i >>= 1) { + if (layout >= 1LL << i) { + channel_id += i; + layout >>= i; + } + } + if (channel_id >= 64 || layout0 != 1LL << channel_id) + return AVERROR(EINVAL); + *rchannel = channel_id; + *arg += len; + return 0; + } + return AVERROR(EINVAL); +} + +static void parse_map(AVFilterContext *ctx) +{ + HeadphoneContext *s = ctx->priv; + char *arg, *tokenizer, *p, *args = av_strdup(s->map); + int i; + + if (!args) + return; + p = args; + + s->nb_inputs = 1; + + for (i = 0; i < 64; i++) { + s->mapping[i] = -1; + } + + while ((arg = av_strtok(p, "|", &tokenizer))) { + int out_ch_id; + char buf[8]; + + p = NULL; + if (parse_channel_name(&arg, &out_ch_id, buf)) { + av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", buf); + continue; + } + s->mapping[s->nb_inputs - 1] = out_ch_id; + s->nb_inputs++; + } + s->nb_irs = s->nb_inputs - 1; + + av_free(args); +} + +typedef struct ThreadData { + AVFrame *in, *out; + int *write; + int **delay; + float **ir; + int *n_clippings; + float **ringbuffer; + float **temp_src; + FFTComplex **temp_fft; +} ThreadData; + +static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) +{ + HeadphoneContext *s = ctx->priv; + ThreadData *td = arg; + AVFrame *in = td->in, *out = td->out; + int offset = jobnr; + int *write = &td->write[jobnr]; + const int *const delay = td->delay[jobnr]; + const float *const ir = td->ir[jobnr]; + int *n_clippings = &td->n_clippings[jobnr]; + float *ringbuffer = td->ringbuffer[jobnr]; + float *temp_src = td->temp_src[jobnr]; + const int ir_len = s->ir_len; + const float *src = (const float *)in->data[0]; + float *dst = (float *)out->data[0]; + const int in_channels = in->channels; + const int buffer_length = s->buffer_length; + const uint32_t modulo = (uint32_t)buffer_length - 1; + float *buffer[16]; + int wr = *write; + int read; + int i, l; + + dst += offset; + for (l = 0; l < in_channels; l++) { + buffer[l] = ringbuffer + l * buffer_length; + } + + for (i = 0; i < in->nb_samples; i++) { + const float *temp_ir = ir; + + *dst = 0; + for (l = 0; l < in_channels; l++) { + *(buffer[l] + wr) = src[l]; + } + + for (l = 0; l < in_channels; l++) { + const float *const bptr = buffer[l]; + + if (l == s->lfe_channel) { + *dst += *(buffer[s->lfe_channel] + wr) * s->gain_lfe; + temp_ir += FFALIGN(ir_len, 16); + continue; + } + + read = (wr - *(delay + l) - (ir_len - 1) + buffer_length) & modulo; + + if (read + ir_len < buffer_length) { + memcpy(temp_src, bptr + read, ir_len * sizeof(*temp_src)); + } else { + int len = FFMIN(ir_len - (read % ir_len), buffer_length - read); + + memcpy(temp_src, bptr + read, len * sizeof(*temp_src)); + memcpy(temp_src + len, bptr, (ir_len - len) * sizeof(*temp_src)); + } + + dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, ir_len); + temp_ir += FFALIGN(ir_len, 16); + } + + if (fabs(*dst) > 1) + *n_clippings += 1; + + dst += 2; + src += in_channels; + wr = (wr + 1) & modulo; + } + + *write = wr; + + return 0; +} + +static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs) +{ + HeadphoneContext *s = ctx->priv; + ThreadData *td = arg; + AVFrame *in = td->in, *out = td->out; + int offset = jobnr; + int *write = &td->write[jobnr]; + FFTComplex *hrtf = s->data_hrtf[jobnr]; + int *n_clippings = &td->n_clippings[jobnr]; + float *ringbuffer = td->ringbuffer[jobnr]; + const int ir_len = s->ir_len; + const float *src = (const float *)in->data[0]; + float *dst = (float *)out->data[0]; + const int in_channels = in->channels; + const int buffer_length = s->buffer_length; + const uint32_t modulo = (uint32_t)buffer_length - 1; + FFTComplex *fft_in = s->temp_fft[jobnr]; + FFTContext *ifft = s->ifft[jobnr]; + FFTContext *fft = s->fft[jobnr]; + const int n_fft = s->n_fft; + const float fft_scale = 1.0f / s->n_fft; + FFTComplex *hrtf_offset; + int wr = *write; + int n_read; + int i, j; + + dst += offset; + + n_read = FFMIN(s->ir_len, in->nb_samples); + for (j = 0; j < n_read; j++) { + dst[2 * j] = ringbuffer[wr]; + ringbuffer[wr] = 0.0; + wr = (wr + 1) & modulo; + } + + for (j = n_read; j < in->nb_samples; j++) { + dst[2 * j] = 0; + } + + for (i = 0; i < in_channels; i++) { + if (i == s->lfe_channel) { + for (j = 0; j < in->nb_samples; j++) { + dst[2 * j] += src[i + j * in_channels] * s->gain_lfe; + } + continue; + } + + offset = i * n_fft; + hrtf_offset = hrtf + offset; + + memset(fft_in, 0, sizeof(FFTComplex) * n_fft); + + for (j = 0; j < in->nb_samples; j++) { + fft_in[j].re = src[j * in_channels + i]; + } + + av_fft_permute(fft, fft_in); + av_fft_calc(fft, fft_in); + for (j = 0; j < n_fft; j++) { + const FFTComplex *hcomplex = hrtf_offset + j; + const float re = fft_in[j].re; + const float im = fft_in[j].im; + + fft_in[j].re = re * hcomplex->re - im * hcomplex->im; + fft_in[j].im = re * hcomplex->im + im * hcomplex->re; + } + + av_fft_permute(ifft, fft_in); + av_fft_calc(ifft, fft_in); + + for (j = 0; j < in->nb_samples; j++) { + dst[2 * j] += fft_in[j].re * fft_scale; + } + + for (j = 0; j < ir_len - 1; j++) { + int write_pos = (wr + j) & modulo; + + *(ringbuffer + write_pos) += fft_in[in->nb_samples + j].re * fft_scale; + } + } + + for (i = 0; i < out->nb_samples; i++) { + if (fabs(*dst) > 1) { + n_clippings[0]++; + } + + dst += 2; + } + + *write = wr; + + return 0; +} + +static int read_ir(AVFilterLink *inlink, AVFrame *frame) +{ + AVFilterContext *ctx = inlink->dst; + HeadphoneContext *s = ctx->priv; + int ir_len, max_ir_len, input_number; + + for (input_number = 0; input_number < s->nb_inputs; input_number++) + if (inlink == ctx->inputs[input_number]) + break; + + av_audio_fifo_write(s->in[input_number].fifo, (void **)frame->extended_data, + frame->nb_samples); + av_frame_free(&frame); + + ir_len = av_audio_fifo_size(s->in[input_number].fifo); + max_ir_len = 4096; + if (ir_len > max_ir_len) { + av_log(ctx, AV_LOG_ERROR, "Too big length of IRs: %d > %d.\n", ir_len, max_ir_len); + return AVERROR(EINVAL); + } + s->in[input_number].ir_len = ir_len; + s->ir_len = FFMAX(ir_len, s->ir_len); + + return 0; +} + +static int headphone_frame(HeadphoneContext *s, AVFilterLink *outlink) +{ + AVFilterContext *ctx = outlink->src; + AVFrame *in = s->in[0].frame; + int n_clippings[2] = { 0 }; + ThreadData td; + AVFrame *out; + + av_audio_fifo_read(s->in[0].fifo, (void **)in->extended_data, s->size); + + out = ff_get_audio_buffer(outlink, in->nb_samples); + if (!out) { + av_frame_free(&in); + return AVERROR(ENOMEM); + } + out->pts = s->pts; + if (s->pts != AV_NOPTS_VALUE) + s->pts += av_rescale_q(out->nb_samples, (AVRational){1, outlink->sample_rate}, outlink->time_base); + + td.in = in; td.out = out; td.write = s->write; + td.delay = s->delay; td.ir = s->data_ir; td.n_clippings = n_clippings; + td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src; + td.temp_fft = s->temp_fft; + + if (s->type == TIME_DOMAIN) { + ctx->internal->execute(ctx, headphone_convolute, &td, NULL, 2); + } else { + ctx->internal->execute(ctx, headphone_fast_convolute, &td, NULL, 2); + } + emms_c(); + + if (n_clippings[0] + n_clippings[1] > 0) { + av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n", + n_clippings[0] + n_clippings[1], out->nb_samples * 2); + } + + return ff_filter_frame(outlink, out); +} + +static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink) +{ + struct HeadphoneContext *s = ctx->priv; + const int ir_len = s->ir_len; + int nb_irs = s->nb_irs; + int nb_input_channels = ctx->inputs[0]->channels; + float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10); + FFTComplex *data_hrtf_l = NULL; + FFTComplex *data_hrtf_r = NULL; + FFTComplex *fft_in_l = NULL; + FFTComplex *fft_in_r = NULL; + float *data_ir_l = NULL; + float *data_ir_r = NULL; + int offset = 0; + int n_fft; + int i, j; + + s->buffer_length = 1 << (32 - ff_clz(s->ir_len)); + s->n_fft = n_fft = 1 << (32 - ff_clz(s->ir_len + inlink->sample_rate)); + + if (s->type == FREQUENCY_DOMAIN) { + fft_in_l = av_calloc(n_fft, sizeof(*fft_in_l)); + fft_in_r = av_calloc(n_fft, sizeof(*fft_in_r)); + if (!fft_in_l || !fft_in_r) { + return AVERROR(ENOMEM); + } + + av_fft_end(s->fft[0]); + av_fft_end(s->fft[1]); + s->fft[0] = av_fft_init(log2(s->n_fft), 0); + s->fft[1] = av_fft_init(log2(s->n_fft), 0); + av_fft_end(s->ifft[0]); + av_fft_end(s->ifft[1]); + s->ifft[0] = av_fft_init(log2(s->n_fft), 1); + s->ifft[1] = av_fft_init(log2(s->n_fft), 1); + + if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) { + av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft); + return AVERROR(ENOMEM); + } + } + + s->data_ir[0] = av_calloc(FFALIGN(s->ir_len, 16), sizeof(float) * s->nb_irs); + s->data_ir[1] = av_calloc(FFALIGN(s->ir_len, 16), sizeof(float) * s->nb_irs); + s->delay[0] = av_malloc_array(s->nb_irs, sizeof(float)); + s->delay[1] = av_malloc_array(s->nb_irs, sizeof(float)); + + if (s->type == TIME_DOMAIN) { + s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels); + s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels); + } else { + s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float)); + s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float)); + s->temp_fft[0] = av_malloc_array(s->n_fft, sizeof(FFTComplex)); + s->temp_fft[1] = av_malloc_array(s->n_fft, sizeof(FFTComplex)); + if (!s->temp_fft[0] || !s->temp_fft[1]) + return AVERROR(ENOMEM); + } + + if (!s->data_ir[0] || !s->data_ir[1] || + !s->ringbuffer[0] || !s->ringbuffer[1]) + return AVERROR(ENOMEM); + + s->in[0].frame = ff_get_audio_buffer(ctx->inputs[0], s->size); + if (!s->in[0].frame) + return AVERROR(ENOMEM); + for (i = 0; i < s->nb_irs; i++) { + s->in[i + 1].frame = ff_get_audio_buffer(ctx->inputs[i + 1], s->ir_len); + if (!s->in[i + 1].frame) + return AVERROR(ENOMEM); + } + + if (s->type == TIME_DOMAIN) { + s->temp_src[0] = av_calloc(FFALIGN(ir_len, 16), sizeof(float)); + s->temp_src[1] = av_calloc(FFALIGN(ir_len, 16), sizeof(float)); + + data_ir_l = av_calloc(nb_irs * FFALIGN(ir_len, 16), sizeof(*data_ir_l)); + data_ir_r = av_calloc(nb_irs * FFALIGN(ir_len, 16), sizeof(*data_ir_r)); + if (!data_ir_r || !data_ir_l || !s->temp_src[0] || !s->temp_src[1]) { + av_free(data_ir_l); + av_free(data_ir_r); + return AVERROR(ENOMEM); + } + } else { + data_hrtf_l = av_malloc_array(n_fft, sizeof(*data_hrtf_l) * nb_irs); + data_hrtf_r = av_malloc_array(n_fft, sizeof(*data_hrtf_r) * nb_irs); + if (!data_hrtf_r || !data_hrtf_l) { + av_free(data_hrtf_l); + av_free(data_hrtf_r); + return AVERROR(ENOMEM); + } + } + + for (i = 0; i < s->nb_irs; i++) { + int len = s->in[i + 1].ir_len; + int delay_l = s->in[i + 1].delay_l; + int delay_r = s->in[i + 1].delay_r; + int idx = -1; + float *ptr; + + for (j = 0; j < inlink->channels; j++) { + if (s->mapping[i] < 0) { + continue; + } + + if ((av_channel_layout_extract_channel(inlink->channel_layout, j)) == (1LL << s->mapping[i])) { + idx = j; + break; + } + } + if (idx == -1) + continue; + + av_audio_fifo_read(s->in[i + 1].fifo, (void **)s->in[i + 1].frame->extended_data, len); + ptr = (float *)s->in[i + 1].frame->extended_data[0]; + + if (s->type == TIME_DOMAIN) { + offset = idx * FFALIGN(len, 16); + for (j = 0; j < len; j++) { + *(data_ir_l + offset + j) = ptr[len * 2 - j * 2 - 2] * gain_lin; + *(data_ir_r + offset + j) = ptr[len * 2 - j * 2 - 1] * gain_lin; + } + } else { + memset(fft_in_l, 0, n_fft * sizeof(*fft_in_l)); + memset(fft_in_r, 0, n_fft * sizeof(*fft_in_r)); + + offset = idx * n_fft; + for (j = 0; j < len; j++) { + fft_in_l[delay_l + j].re = ptr[j * 2 ] * gain_lin; + fft_in_r[delay_r + j].re = ptr[j * 2 + 1] * gain_lin; + } + + av_fft_permute(s->fft[0], fft_in_l); + av_fft_calc(s->fft[0], fft_in_l); + memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l)); + av_fft_permute(s->fft[0], fft_in_r); + av_fft_calc(s->fft[0], fft_in_r); + memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r)); + } + } + + if (s->type == TIME_DOMAIN) { + memcpy(s->data_ir[0], data_ir_l, sizeof(float) * nb_irs * FFALIGN(ir_len, 16)); + memcpy(s->data_ir[1], data_ir_r, sizeof(float) * nb_irs * FFALIGN(ir_len, 16)); + + av_freep(&data_ir_l); + av_freep(&data_ir_r); + } else { + s->data_hrtf[0] = av_malloc_array(n_fft * s->nb_irs, sizeof(FFTComplex)); + s->data_hrtf[1] = av_malloc_array(n_fft * s->nb_irs, sizeof(FFTComplex)); + if (!s->data_hrtf[0] || !s->data_hrtf[1]) { + av_freep(&data_hrtf_l); + av_freep(&data_hrtf_r); + av_freep(&fft_in_l); + av_freep(&fft_in_r); + return AVERROR(ENOMEM); + } + + memcpy(s->data_hrtf[0], data_hrtf_l, + sizeof(FFTComplex) * nb_irs * n_fft); + memcpy(s->data_hrtf[1], data_hrtf_r, + sizeof(FFTComplex) * nb_irs * n_fft); + + av_freep(&data_hrtf_l); + av_freep(&data_hrtf_r); + + av_freep(&fft_in_l); + av_freep(&fft_in_r); + } + + s->have_coeffs = 1; + + return 0; +} + +static int filter_frame(AVFilterLink *inlink, AVFrame *in) +{ + AVFilterContext *ctx = inlink->dst; + HeadphoneContext *s = ctx->priv; + AVFilterLink *outlink = ctx->outputs[0]; + int ret = 0; + + av_audio_fifo_write(s->in[0].fifo, (void **)in->extended_data, + in->nb_samples); + if (s->pts == AV_NOPTS_VALUE) + s->pts = in->pts; + + av_frame_free(&in); + + if (!s->have_coeffs && s->eof_coeffs) { + ret = convert_coeffs(ctx, inlink); + if (ret < 0) + return ret; + } + + if (s->have_coeffs) { + while (av_audio_fifo_size(s->in[0].fifo) >= s->size) { + ret = headphone_frame(s, outlink); + if (ret < 0) + break; + } + } + return ret; +} + +static int query_formats(AVFilterContext *ctx) +{ + struct HeadphoneContext *s = ctx->priv; + AVFilterFormats *formats = NULL; + AVFilterChannelLayouts *layouts = NULL; + int ret, i; + + ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT); + if (ret) + return ret; + ret = ff_set_common_formats(ctx, formats); + if (ret) + return ret; + + layouts = ff_all_channel_layouts(); + if (!layouts) + return AVERROR(ENOMEM); + + ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts); + if (ret) + return ret; + + layouts = NULL; + ret = ff_add_channel_layout(&layouts, AV_CH_LAYOUT_STEREO); + if (ret) + return ret; + + for (i = 1; i < s->nb_inputs; i++) { + ret = ff_channel_layouts_ref(layouts, &ctx->inputs[i]->out_channel_layouts); + if (ret) + return ret; + } + + ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts); + if (ret) + return ret; + + formats = ff_all_samplerates(); + if (!formats) + return AVERROR(ENOMEM); + return ff_set_common_samplerates(ctx, formats); +} + +static int config_input(AVFilterLink *inlink) +{ + AVFilterContext *ctx = inlink->dst; + HeadphoneContext *s = ctx->priv; + + if (s->type == FREQUENCY_DOMAIN) { + inlink->partial_buf_size = + inlink->min_samples = + inlink->max_samples = inlink->sample_rate; + } + + return 0; +} + +static av_cold int init(AVFilterContext *ctx) +{ + HeadphoneContext *s = ctx->priv; + int i; + + AVFilterPad pad = { + .name = "in0", + .type = AVMEDIA_TYPE_AUDIO, + .config_props = config_input, + .filter_frame = filter_frame, + }; + ff_insert_inpad(ctx, 0, &pad); + + if (!s->map) { + av_log(ctx, AV_LOG_ERROR, "Valid mapping must be set.\n"); + return AVERROR(EINVAL); + } + + parse_map(ctx); + + s->in = av_calloc(s->nb_inputs, sizeof(*s->in)); + if (!s->in) + return AVERROR(ENOMEM); + + for (i = 1; i < s->nb_inputs; i++) { + char *name = av_asprintf("ir%d", i - 1); + AVFilterPad pad = { + .name = name, + .type = AVMEDIA_TYPE_AUDIO, + .filter_frame = read_ir, + }; + if (!name) + return AVERROR(ENOMEM); + ff_insert_inpad(ctx, i, &pad); + } + + s->fdsp = avpriv_float_dsp_alloc(0); + if (!s->fdsp) + return AVERROR(ENOMEM); + s->pts = AV_NOPTS_VALUE; + + return 0; +} + +static int config_output(AVFilterLink *outlink) +{ + AVFilterContext *ctx = outlink->src; + HeadphoneContext *s = ctx->priv; + AVFilterLink *inlink = ctx->inputs[0]; + int i; + + if (s->type == TIME_DOMAIN) + s->size = 1024; + else + s->size = inlink->sample_rate; + + for (i = 0; i < s->nb_inputs; i++) { + s->in[i].fifo = av_audio_fifo_alloc(ctx->inputs[i]->format, ctx->inputs[i]->channels, 1024); + if (!s->in[i].fifo) + return AVERROR(ENOMEM); + } + s->gain_lfe = expf((s->gain - 3 * inlink->channels - 6 + s->lfe_gain) / 20 * M_LN10); + + return 0; +} + +static int request_frame(AVFilterLink *outlink) +{ + AVFilterContext *ctx = outlink->src; + HeadphoneContext *s = ctx->priv; + int i, ret; + + for (i = 1; !s->eof_coeffs && i < s->nb_inputs; i++) { + if (!s->in[i].eof) { + ret = ff_request_frame(ctx->inputs[i]); + if (ret == AVERROR_EOF) { + s->in[i].eof = 1; + ret = 0; + } + return ret; + } else { + if (i == s->nb_inputs - 1) + s->eof_coeffs = 1; + } + } + return ff_request_frame(ctx->inputs[0]); +} + +static av_cold void uninit(AVFilterContext *ctx) +{ + HeadphoneContext *s = ctx->priv; + int i; + + av_fft_end(s->ifft[0]); + av_fft_end(s->ifft[1]); + av_fft_end(s->fft[0]); + av_fft_end(s->fft[1]); + av_freep(&s->delay[0]); + av_freep(&s->delay[1]); + av_freep(&s->data_ir[0]); + av_freep(&s->data_ir[1]); + av_freep(&s->ringbuffer[0]); + av_freep(&s->ringbuffer[1]); + av_freep(&s->temp_src[0]); + av_freep(&s->temp_src[1]); + av_freep(&s->temp_fft[0]); + av_freep(&s->temp_fft[1]); + av_freep(&s->data_hrtf[0]); + av_freep(&s->data_hrtf[1]); + av_freep(&s->fdsp); + + for (i = 0; i < s->nb_inputs; i++) { + av_frame_free(&s->in[i].frame); + av_audio_fifo_free(s->in[i].fifo); + if (ctx->input_pads && i) + av_freep(&ctx->input_pads[i].name); + } + av_freep(&s->in); +} + +#define OFFSET(x) offsetof(HeadphoneContext, x) +#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM + +static const AVOption headphone_options[] = { + { "map", "set channels convolution mappings", OFFSET(map), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS }, + { "gain", "set gain in dB", OFFSET(gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS }, + { "lfe", "set lfe gain in dB", OFFSET(lfe_gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS }, + { "type", "set processing", OFFSET(type), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, .flags = FLAGS, "type" }, + { "time", "time domain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, .flags = FLAGS, "type" }, + { "freq", "frequency domain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, .flags = FLAGS, "type" }, + { NULL } +}; + +AVFILTER_DEFINE_CLASS(headphone); + +static const AVFilterPad outputs[] = { + { + .name = "default", + .type = AVMEDIA_TYPE_AUDIO, + .config_props = config_output, + .request_frame = request_frame, + }, + { NULL } +}; + +AVFilter ff_af_headphone = { + .name = "headphone", + .description = NULL_IF_CONFIG_SMALL("Apply headphone binaural spatialization with HRTFs in additional streams."), + .priv_size = sizeof(HeadphoneContext), + .priv_class = &headphone_class, + .init = init, + .uninit = uninit, + .query_formats = query_formats, + .inputs = NULL, + .outputs = outputs, + .flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_DYNAMIC_INPUTS, +}; diff --git a/libavfilter/allfilters.c b/libavfilter/allfilters.c index 534c340..94f7cf3 100644 --- a/libavfilter/allfilters.c +++ b/libavfilter/allfilters.c @@ -105,6 +105,7 @@ static void register_all(void) REGISTER_FILTER(FIREQUALIZER, firequalizer, af); REGISTER_FILTER(FLANGER, flanger, af); REGISTER_FILTER(HDCD, hdcd, af); + REGISTER_FILTER(HEADPHONE, headphone, af); REGISTER_FILTER(HIGHPASS, highpass, af); REGISTER_FILTER(JOIN, join, af); REGISTER_FILTER(LADSPA, ladspa, af);