2 * Copyright (C) 2017 Paul B Mahol
3 * Copyright (C) 2013-2015 Andreas Fuchs, Wolfgang Hrauda
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU Lesser General Public
8 * License as published by the Free Software Foundation; either
9 * version 2.1 of the License, or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include "libavutil/avstring.h"
24 #include "libavutil/channel_layout.h"
25 #include "libavutil/float_dsp.h"
26 #include "libavutil/intmath.h"
27 #include "libavutil/opt.h"
28 #include "libavcodec/avfft.h"
36 #define FREQUENCY_DOMAIN 1
41 typedef struct HeadphoneContext {
62 float lfe_gain, gain_lfe;
75 FFTComplex *temp_fft[2];
77 FFTContext *fft[2], *ifft[2];
78 FFTComplex *data_hrtf[2];
80 AVFloatDSPContext *fdsp;
81 struct headphone_inputs {
90 static int parse_channel_name(HeadphoneContext *s, int x, char **arg, int *rchannel, char *buf)
92 int len, i, channel_id = 0;
93 int64_t layout, layout0;
95 if (sscanf(*arg, "%7[A-Z]%n", buf, &len)) {
96 layout0 = layout = av_get_channel_layout(buf);
97 if (layout == AV_CH_LOW_FREQUENCY)
99 for (i = 32; i > 0; i >>= 1) {
100 if (layout >= 1LL << i) {
105 if (channel_id >= 64 || layout0 != 1LL << channel_id)
106 return AVERROR(EINVAL);
107 *rchannel = channel_id;
111 return AVERROR(EINVAL);
114 static void parse_map(AVFilterContext *ctx)
116 HeadphoneContext *s = ctx->priv;
117 char *arg, *tokenizer, *p, *args = av_strdup(s->map);
127 for (i = 0; i < 64; i++) {
131 while ((arg = av_strtok(p, "|", &tokenizer))) {
136 if (parse_channel_name(s, s->nb_irs, &arg, &out_ch_id, buf)) {
137 av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", buf);
140 s->mapping[s->nb_irs] = out_ch_id;
144 if (s->hrir_fmt == HRIR_MULTI)
147 s->nb_inputs = s->nb_irs + 1;
152 typedef struct ThreadData {
160 FFTComplex **temp_fft;
163 static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
165 HeadphoneContext *s = ctx->priv;
166 ThreadData *td = arg;
167 AVFrame *in = td->in, *out = td->out;
169 int *write = &td->write[jobnr];
170 const int *const delay = td->delay[jobnr];
171 const float *const ir = td->ir[jobnr];
172 int *n_clippings = &td->n_clippings[jobnr];
173 float *ringbuffer = td->ringbuffer[jobnr];
174 float *temp_src = td->temp_src[jobnr];
175 const int ir_len = s->ir_len;
176 const int air_len = s->air_len;
177 const float *src = (const float *)in->data[0];
178 float *dst = (float *)out->data[0];
179 const int in_channels = in->channels;
180 const int buffer_length = s->buffer_length;
181 const uint32_t modulo = (uint32_t)buffer_length - 1;
188 for (l = 0; l < in_channels; l++) {
189 buffer[l] = ringbuffer + l * buffer_length;
192 for (i = 0; i < in->nb_samples; i++) {
193 const float *temp_ir = ir;
196 for (l = 0; l < in_channels; l++) {
197 *(buffer[l] + wr) = src[l];
200 for (l = 0; l < in_channels; l++) {
201 const float *const bptr = buffer[l];
203 if (l == s->lfe_channel) {
204 *dst += *(buffer[s->lfe_channel] + wr) * s->gain_lfe;
209 read = (wr - *(delay + l) - (air_len - 1) + buffer_length) & modulo;
211 if (read + air_len < buffer_length) {
212 memcpy(temp_src, bptr + read, air_len * sizeof(*temp_src));
214 int len = FFMIN(air_len - (read % air_len), buffer_length - read);
216 memcpy(temp_src, bptr + read, len * sizeof(*temp_src));
217 memcpy(temp_src + len, bptr, (air_len - len) * sizeof(*temp_src));
220 dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, FFALIGN(ir_len, 32));
224 if (fabsf(dst[0]) > 1)
229 wr = (wr + 1) & modulo;
237 static int headphone_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
239 HeadphoneContext *s = ctx->priv;
240 ThreadData *td = arg;
241 AVFrame *in = td->in, *out = td->out;
243 int *write = &td->write[jobnr];
244 FFTComplex *hrtf = s->data_hrtf[jobnr];
245 int *n_clippings = &td->n_clippings[jobnr];
246 float *ringbuffer = td->ringbuffer[jobnr];
247 const int ir_len = s->ir_len;
248 const float *src = (const float *)in->data[0];
249 float *dst = (float *)out->data[0];
250 const int in_channels = in->channels;
251 const int buffer_length = s->buffer_length;
252 const uint32_t modulo = (uint32_t)buffer_length - 1;
253 FFTComplex *fft_in = s->temp_fft[jobnr];
254 FFTContext *ifft = s->ifft[jobnr];
255 FFTContext *fft = s->fft[jobnr];
256 const int n_fft = s->n_fft;
257 const float fft_scale = 1.0f / s->n_fft;
258 FFTComplex *hrtf_offset;
265 n_read = FFMIN(ir_len, in->nb_samples);
266 for (j = 0; j < n_read; j++) {
267 dst[2 * j] = ringbuffer[wr];
268 ringbuffer[wr] = 0.0;
269 wr = (wr + 1) & modulo;
272 for (j = n_read; j < in->nb_samples; j++) {
276 for (i = 0; i < in_channels; i++) {
277 if (i == s->lfe_channel) {
278 for (j = 0; j < in->nb_samples; j++) {
279 dst[2 * j] += src[i + j * in_channels] * s->gain_lfe;
285 hrtf_offset = hrtf + offset;
287 memset(fft_in, 0, sizeof(FFTComplex) * n_fft);
289 for (j = 0; j < in->nb_samples; j++) {
290 fft_in[j].re = src[j * in_channels + i];
293 av_fft_permute(fft, fft_in);
294 av_fft_calc(fft, fft_in);
295 for (j = 0; j < n_fft; j++) {
296 const FFTComplex *hcomplex = hrtf_offset + j;
297 const float re = fft_in[j].re;
298 const float im = fft_in[j].im;
300 fft_in[j].re = re * hcomplex->re - im * hcomplex->im;
301 fft_in[j].im = re * hcomplex->im + im * hcomplex->re;
304 av_fft_permute(ifft, fft_in);
305 av_fft_calc(ifft, fft_in);
307 for (j = 0; j < in->nb_samples; j++) {
308 dst[2 * j] += fft_in[j].re * fft_scale;
311 for (j = 0; j < ir_len - 1; j++) {
312 int write_pos = (wr + j) & modulo;
314 *(ringbuffer + write_pos) += fft_in[in->nb_samples + j].re * fft_scale;
318 for (i = 0; i < out->nb_samples; i++) {
319 if (fabsf(dst[0]) > 1) {
331 static int check_ir(AVFilterLink *inlink, int input_number)
333 AVFilterContext *ctx = inlink->dst;
334 HeadphoneContext *s = ctx->priv;
335 int ir_len, max_ir_len;
337 ir_len = ff_inlink_queued_samples(inlink);
339 if (ir_len > max_ir_len) {
340 av_log(ctx, AV_LOG_ERROR, "Too big length of IRs: %d > %d.\n", ir_len, max_ir_len);
341 return AVERROR(EINVAL);
343 s->in[input_number].ir_len = ir_len;
344 s->ir_len = FFMAX(ir_len, s->ir_len);
349 static int headphone_frame(HeadphoneContext *s, AVFrame *in, AVFilterLink *outlink)
351 AVFilterContext *ctx = outlink->src;
352 int n_clippings[2] = { 0 };
356 out = ff_get_audio_buffer(outlink, in->nb_samples);
359 return AVERROR(ENOMEM);
363 td.in = in; td.out = out; td.write = s->write;
364 td.delay = s->delay; td.ir = s->data_ir; td.n_clippings = n_clippings;
365 td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
366 td.temp_fft = s->temp_fft;
368 if (s->type == TIME_DOMAIN) {
369 ctx->internal->execute(ctx, headphone_convolute, &td, NULL, 2);
371 ctx->internal->execute(ctx, headphone_fast_convolute, &td, NULL, 2);
375 if (n_clippings[0] + n_clippings[1] > 0) {
376 av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n",
377 n_clippings[0] + n_clippings[1], out->nb_samples * 2);
381 return ff_filter_frame(outlink, out);
384 static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
386 struct HeadphoneContext *s = ctx->priv;
387 const int ir_len = s->ir_len;
388 int nb_irs = s->nb_irs;
389 int nb_input_channels = ctx->inputs[0]->channels;
390 float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10);
391 FFTComplex *data_hrtf_l = NULL;
392 FFTComplex *data_hrtf_r = NULL;
393 FFTComplex *fft_in_l = NULL;
394 FFTComplex *fft_in_r = NULL;
395 float *data_ir_l = NULL;
396 float *data_ir_r = NULL;
397 int offset = 0, ret = 0;
401 s->air_len = 1 << (32 - ff_clz(ir_len));
402 s->buffer_length = 1 << (32 - ff_clz(s->air_len));
403 s->n_fft = n_fft = 1 << (32 - ff_clz(ir_len + s->size));
405 if (s->type == FREQUENCY_DOMAIN) {
406 fft_in_l = av_calloc(n_fft, sizeof(*fft_in_l));
407 fft_in_r = av_calloc(n_fft, sizeof(*fft_in_r));
408 if (!fft_in_l || !fft_in_r) {
409 ret = AVERROR(ENOMEM);
413 av_fft_end(s->fft[0]);
414 av_fft_end(s->fft[1]);
415 s->fft[0] = av_fft_init(log2(s->n_fft), 0);
416 s->fft[1] = av_fft_init(log2(s->n_fft), 0);
417 av_fft_end(s->ifft[0]);
418 av_fft_end(s->ifft[1]);
419 s->ifft[0] = av_fft_init(log2(s->n_fft), 1);
420 s->ifft[1] = av_fft_init(log2(s->n_fft), 1);
422 if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {
423 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);
424 ret = AVERROR(ENOMEM);
429 s->data_ir[0] = av_calloc(s->air_len, sizeof(float) * s->nb_irs);
430 s->data_ir[1] = av_calloc(s->air_len, sizeof(float) * s->nb_irs);
431 s->delay[0] = av_calloc(s->nb_irs, sizeof(float));
432 s->delay[1] = av_calloc(s->nb_irs, sizeof(float));
434 if (s->type == TIME_DOMAIN) {
435 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
436 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
438 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float));
439 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));
440 s->temp_fft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
441 s->temp_fft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
442 if (!s->temp_fft[0] || !s->temp_fft[1]) {
443 ret = AVERROR(ENOMEM);
448 if (!s->data_ir[0] || !s->data_ir[1] ||
449 !s->ringbuffer[0] || !s->ringbuffer[1]) {
450 ret = AVERROR(ENOMEM);
454 if (s->type == TIME_DOMAIN) {
455 s->temp_src[0] = av_calloc(s->air_len, sizeof(float));
456 s->temp_src[1] = av_calloc(s->air_len, sizeof(float));
458 data_ir_l = av_calloc(nb_irs * s->air_len, sizeof(*data_ir_l));
459 data_ir_r = av_calloc(nb_irs * s->air_len, sizeof(*data_ir_r));
460 if (!data_ir_r || !data_ir_l || !s->temp_src[0] || !s->temp_src[1]) {
461 ret = AVERROR(ENOMEM);
465 data_hrtf_l = av_calloc(n_fft, sizeof(*data_hrtf_l) * nb_irs);
466 data_hrtf_r = av_calloc(n_fft, sizeof(*data_hrtf_r) * nb_irs);
467 if (!data_hrtf_r || !data_hrtf_l) {
468 ret = AVERROR(ENOMEM);
473 for (i = 0; i < s->nb_inputs - 1; i++) {
474 int len = s->in[i + 1].ir_len;
475 int delay_l = s->in[i + 1].delay_l;
476 int delay_r = s->in[i + 1].delay_r;
479 ret = ff_inlink_consume_samples(ctx->inputs[i + 1], len, len, &s->in[i + 1].frame);
482 ptr = (float *)s->in[i + 1].frame->extended_data[0];
484 if (s->hrir_fmt == HRIR_STEREO) {
487 for (j = 0; j < inlink->channels; j++) {
488 if (s->mapping[i] < 0) {
492 if ((av_channel_layout_extract_channel(inlink->channel_layout, j)) == (1LL << s->mapping[i])) {
500 if (s->type == TIME_DOMAIN) {
501 offset = idx * s->air_len;
502 for (j = 0; j < len; j++) {
503 data_ir_l[offset + j] = ptr[len * 2 - j * 2 - 2] * gain_lin;
504 data_ir_r[offset + j] = ptr[len * 2 - j * 2 - 1] * gain_lin;
507 memset(fft_in_l, 0, n_fft * sizeof(*fft_in_l));
508 memset(fft_in_r, 0, n_fft * sizeof(*fft_in_r));
510 offset = idx * n_fft;
511 for (j = 0; j < len; j++) {
512 fft_in_l[delay_l + j].re = ptr[j * 2 ] * gain_lin;
513 fft_in_r[delay_r + j].re = ptr[j * 2 + 1] * gain_lin;
516 av_fft_permute(s->fft[0], fft_in_l);
517 av_fft_calc(s->fft[0], fft_in_l);
518 memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l));
519 av_fft_permute(s->fft[0], fft_in_r);
520 av_fft_calc(s->fft[0], fft_in_r);
521 memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r));
524 int I, N = ctx->inputs[1]->channels;
526 for (k = 0; k < N / 2; k++) {
529 for (j = 0; j < inlink->channels; j++) {
530 if (s->mapping[k] < 0) {
534 if ((av_channel_layout_extract_channel(inlink->channel_layout, j)) == (1LL << s->mapping[k])) {
543 if (s->type == TIME_DOMAIN) {
544 offset = idx * s->air_len;
545 for (j = 0; j < len; j++) {
546 data_ir_l[offset + j] = ptr[len * N - j * N - N + I ] * gain_lin;
547 data_ir_r[offset + j] = ptr[len * N - j * N - N + I + 1] * gain_lin;
550 memset(fft_in_l, 0, n_fft * sizeof(*fft_in_l));
551 memset(fft_in_r, 0, n_fft * sizeof(*fft_in_r));
553 offset = idx * n_fft;
554 for (j = 0; j < len; j++) {
555 fft_in_l[delay_l + j].re = ptr[j * N + I ] * gain_lin;
556 fft_in_r[delay_r + j].re = ptr[j * N + I + 1] * gain_lin;
559 av_fft_permute(s->fft[0], fft_in_l);
560 av_fft_calc(s->fft[0], fft_in_l);
561 memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l));
562 av_fft_permute(s->fft[0], fft_in_r);
563 av_fft_calc(s->fft[0], fft_in_r);
564 memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r));
569 av_frame_free(&s->in[i + 1].frame);
572 if (s->type == TIME_DOMAIN) {
573 memcpy(s->data_ir[0], data_ir_l, sizeof(float) * nb_irs * s->air_len);
574 memcpy(s->data_ir[1], data_ir_r, sizeof(float) * nb_irs * s->air_len);
576 s->data_hrtf[0] = av_calloc(n_fft * s->nb_irs, sizeof(FFTComplex));
577 s->data_hrtf[1] = av_calloc(n_fft * s->nb_irs, sizeof(FFTComplex));
578 if (!s->data_hrtf[0] || !s->data_hrtf[1]) {
579 ret = AVERROR(ENOMEM);
583 memcpy(s->data_hrtf[0], data_hrtf_l,
584 sizeof(FFTComplex) * nb_irs * n_fft);
585 memcpy(s->data_hrtf[1], data_hrtf_r,
586 sizeof(FFTComplex) * nb_irs * n_fft);
593 for (i = 0; i < s->nb_inputs - 1; i++)
594 av_frame_free(&s->in[i + 1].frame);
596 av_freep(&data_ir_l);
597 av_freep(&data_ir_r);
599 av_freep(&data_hrtf_l);
600 av_freep(&data_hrtf_r);
608 static int activate(AVFilterContext *ctx)
610 HeadphoneContext *s = ctx->priv;
611 AVFilterLink *inlink = ctx->inputs[0];
612 AVFilterLink *outlink = ctx->outputs[0];
616 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
618 for (i = 1; i < s->nb_inputs; i++) {
622 if ((ret = check_ir(ctx->inputs[i], i)) < 0)
626 if (ff_outlink_get_status(ctx->inputs[i]) == AVERROR_EOF)
631 for (i = 1; i < s->nb_inputs; i++) {
636 if (i != s->nb_inputs) {
637 if (ff_outlink_frame_wanted(ctx->outputs[0])) {
638 for (i = 1; i < s->nb_inputs; i++) {
640 ff_inlink_request_frame(ctx->inputs[i]);
650 if (!s->have_hrirs && s->eof_hrirs) {
651 ret = convert_coeffs(ctx, inlink);
656 if ((ret = ff_inlink_consume_samples(ctx->inputs[0], s->size, s->size, &in)) > 0) {
657 ret = headphone_frame(s, in, outlink);
665 FF_FILTER_FORWARD_STATUS(ctx->inputs[0], ctx->outputs[0]);
666 if (ff_outlink_frame_wanted(ctx->outputs[0]))
667 ff_inlink_request_frame(ctx->inputs[0]);
672 static int query_formats(AVFilterContext *ctx)
674 struct HeadphoneContext *s = ctx->priv;
675 AVFilterFormats *formats = NULL;
676 AVFilterChannelLayouts *layouts = NULL;
677 AVFilterChannelLayouts *stereo_layout = NULL;
678 AVFilterChannelLayouts *hrir_layouts = NULL;
681 ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
684 ret = ff_set_common_formats(ctx, formats);
688 layouts = ff_all_channel_layouts();
690 return AVERROR(ENOMEM);
692 ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts);
696 ret = ff_add_channel_layout(&stereo_layout, AV_CH_LAYOUT_STEREO);
700 if (s->hrir_fmt == HRIR_MULTI) {
701 hrir_layouts = ff_all_channel_counts();
703 ret = AVERROR(ENOMEM);
704 ret = ff_channel_layouts_ref(hrir_layouts, &ctx->inputs[1]->out_channel_layouts);
708 for (i = 1; i < s->nb_inputs; i++) {
709 ret = ff_channel_layouts_ref(stereo_layout, &ctx->inputs[i]->out_channel_layouts);
715 ret = ff_channel_layouts_ref(stereo_layout, &ctx->outputs[0]->in_channel_layouts);
719 formats = ff_all_samplerates();
721 return AVERROR(ENOMEM);
722 return ff_set_common_samplerates(ctx, formats);
725 static int config_input(AVFilterLink *inlink)
727 AVFilterContext *ctx = inlink->dst;
728 HeadphoneContext *s = ctx->priv;
730 if (s->nb_irs < inlink->channels) {
731 av_log(ctx, AV_LOG_ERROR, "Number of HRIRs must be >= %d.\n", inlink->channels);
732 return AVERROR(EINVAL);
738 static av_cold int init(AVFilterContext *ctx)
740 HeadphoneContext *s = ctx->priv;
745 .type = AVMEDIA_TYPE_AUDIO,
746 .config_props = config_input,
748 if ((ret = ff_insert_inpad(ctx, 0, &pad)) < 0)
752 av_log(ctx, AV_LOG_ERROR, "Valid mapping must be set.\n");
753 return AVERROR(EINVAL);
758 s->in = av_calloc(s->nb_inputs, sizeof(*s->in));
760 return AVERROR(ENOMEM);
762 for (i = 1; i < s->nb_inputs; i++) {
763 char *name = av_asprintf("hrir%d", i - 1);
766 .type = AVMEDIA_TYPE_AUDIO,
769 return AVERROR(ENOMEM);
770 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
776 s->fdsp = avpriv_float_dsp_alloc(0);
778 return AVERROR(ENOMEM);
783 static int config_output(AVFilterLink *outlink)
785 AVFilterContext *ctx = outlink->src;
786 HeadphoneContext *s = ctx->priv;
787 AVFilterLink *inlink = ctx->inputs[0];
789 if (s->hrir_fmt == HRIR_MULTI) {
790 AVFilterLink *hrir_link = ctx->inputs[1];
792 if (hrir_link->channels < inlink->channels * 2) {
793 av_log(ctx, AV_LOG_ERROR, "Number of channels in HRIR stream must be >= %d.\n", inlink->channels * 2);
794 return AVERROR(EINVAL);
798 s->gain_lfe = expf((s->gain - 3 * inlink->channels + s->lfe_gain) / 20 * M_LN10);
803 static av_cold void uninit(AVFilterContext *ctx)
805 HeadphoneContext *s = ctx->priv;
808 av_fft_end(s->ifft[0]);
809 av_fft_end(s->ifft[1]);
810 av_fft_end(s->fft[0]);
811 av_fft_end(s->fft[1]);
812 av_freep(&s->delay[0]);
813 av_freep(&s->delay[1]);
814 av_freep(&s->data_ir[0]);
815 av_freep(&s->data_ir[1]);
816 av_freep(&s->ringbuffer[0]);
817 av_freep(&s->ringbuffer[1]);
818 av_freep(&s->temp_src[0]);
819 av_freep(&s->temp_src[1]);
820 av_freep(&s->temp_fft[0]);
821 av_freep(&s->temp_fft[1]);
822 av_freep(&s->data_hrtf[0]);
823 av_freep(&s->data_hrtf[1]);
826 for (i = 0; i < s->nb_inputs; i++) {
827 if (ctx->input_pads && i)
828 av_freep(&ctx->input_pads[i].name);
833 #define OFFSET(x) offsetof(HeadphoneContext, x)
834 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
836 static const AVOption headphone_options[] = {
837 { "map", "set channels convolution mappings", OFFSET(map), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
838 { "gain", "set gain in dB", OFFSET(gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
839 { "lfe", "set lfe gain in dB", OFFSET(lfe_gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
840 { "type", "set processing", OFFSET(type), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, .flags = FLAGS, "type" },
841 { "time", "time domain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, .flags = FLAGS, "type" },
842 { "freq", "frequency domain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, .flags = FLAGS, "type" },
843 { "size", "set frame size", OFFSET(size), AV_OPT_TYPE_INT, {.i64=1024},1024,96000, .flags = FLAGS },
844 { "hrir", "set hrir format", OFFSET(hrir_fmt), AV_OPT_TYPE_INT, {.i64=HRIR_STEREO}, 0, 1, .flags = FLAGS, "hrir" },
845 { "stereo", "hrir files have exactly 2 channels", 0, AV_OPT_TYPE_CONST, {.i64=HRIR_STEREO}, 0, 0, .flags = FLAGS, "hrir" },
846 { "multich", "single multichannel hrir file", 0, AV_OPT_TYPE_CONST, {.i64=HRIR_MULTI}, 0, 0, .flags = FLAGS, "hrir" },
850 AVFILTER_DEFINE_CLASS(headphone);
852 static const AVFilterPad outputs[] = {
855 .type = AVMEDIA_TYPE_AUDIO,
856 .config_props = config_output,
861 AVFilter ff_af_headphone = {
863 .description = NULL_IF_CONFIG_SMALL("Apply headphone binaural spatialization with HRTFs in additional streams."),
864 .priv_size = sizeof(HeadphoneContext),
865 .priv_class = &headphone_class,
868 .query_formats = query_formats,
869 .activate = activate,
872 .flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_DYNAMIC_INPUTS,