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/audio_fifo.h"
24 #include "libavutil/avstring.h"
25 #include "libavutil/channel_layout.h"
26 #include "libavutil/float_dsp.h"
27 #include "libavutil/intmath.h"
28 #include "libavutil/opt.h"
29 #include "libavcodec/avfft.h"
37 #define FREQUENCY_DOMAIN 1
42 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 {
91 static int parse_channel_name(HeadphoneContext *s, int x, char **arg, int *rchannel, char *buf)
93 int len, i, channel_id = 0;
94 int64_t layout, layout0;
96 if (sscanf(*arg, "%7[A-Z]%n", buf, &len)) {
97 layout0 = layout = av_get_channel_layout(buf);
98 if (layout == AV_CH_LOW_FREQUENCY)
100 for (i = 32; i > 0; i >>= 1) {
101 if (layout >= 1LL << i) {
106 if (channel_id >= 64 || layout0 != 1LL << channel_id)
107 return AVERROR(EINVAL);
108 *rchannel = channel_id;
112 return AVERROR(EINVAL);
115 static void parse_map(AVFilterContext *ctx)
117 HeadphoneContext *s = ctx->priv;
118 char *arg, *tokenizer, *p, *args = av_strdup(s->map);
128 for (i = 0; i < 64; i++) {
132 while ((arg = av_strtok(p, "|", &tokenizer))) {
137 if (parse_channel_name(s, s->nb_irs, &arg, &out_ch_id, buf)) {
138 av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", buf);
141 s->mapping[s->nb_irs] = out_ch_id;
145 if (s->hrir_fmt == HRIR_MULTI)
148 s->nb_inputs = s->nb_irs + 1;
153 typedef struct ThreadData {
161 FFTComplex **temp_fft;
164 static int headphone_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
166 HeadphoneContext *s = ctx->priv;
167 ThreadData *td = arg;
168 AVFrame *in = td->in, *out = td->out;
170 int *write = &td->write[jobnr];
171 const int *const delay = td->delay[jobnr];
172 const float *const ir = td->ir[jobnr];
173 int *n_clippings = &td->n_clippings[jobnr];
174 float *ringbuffer = td->ringbuffer[jobnr];
175 float *temp_src = td->temp_src[jobnr];
176 const int ir_len = s->ir_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;
205 temp_ir += FFALIGN(ir_len, 16);
209 read = (wr - *(delay + l) - (ir_len - 1) + buffer_length) & modulo;
211 if (read + ir_len < buffer_length) {
212 memcpy(temp_src, bptr + read, ir_len * sizeof(*temp_src));
214 int len = FFMIN(ir_len - (read % ir_len), buffer_length - read);
216 memcpy(temp_src, bptr + read, len * sizeof(*temp_src));
217 memcpy(temp_src + len, bptr, (ir_len - len) * sizeof(*temp_src));
220 dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, ir_len);
221 temp_ir += FFALIGN(ir_len, 16);
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(s->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 (fabs(*dst) > 1) {
331 static int read_ir(AVFilterLink *inlink, int input_number, AVFrame *frame)
333 AVFilterContext *ctx = inlink->dst;
334 HeadphoneContext *s = ctx->priv;
335 int ir_len, max_ir_len, ret;
337 ret = av_audio_fifo_write(s->in[input_number].fifo, (void **)frame->extended_data,
339 av_frame_free(&frame);
344 ir_len = av_audio_fifo_size(s->in[input_number].fifo);
346 if (ir_len > max_ir_len) {
347 av_log(ctx, AV_LOG_ERROR, "Too big length of IRs: %d > %d.\n", ir_len, max_ir_len);
348 return AVERROR(EINVAL);
350 s->in[input_number].ir_len = ir_len;
351 s->ir_len = FFMAX(ir_len, s->ir_len);
356 static int headphone_frame(HeadphoneContext *s, AVFrame *in, AVFilterLink *outlink)
358 AVFilterContext *ctx = outlink->src;
359 int n_clippings[2] = { 0 };
363 out = ff_get_audio_buffer(outlink, in->nb_samples);
366 return AVERROR(ENOMEM);
370 td.in = in; td.out = out; td.write = s->write;
371 td.delay = s->delay; td.ir = s->data_ir; td.n_clippings = n_clippings;
372 td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
373 td.temp_fft = s->temp_fft;
375 if (s->type == TIME_DOMAIN) {
376 ctx->internal->execute(ctx, headphone_convolute, &td, NULL, 2);
378 ctx->internal->execute(ctx, headphone_fast_convolute, &td, NULL, 2);
382 if (n_clippings[0] + n_clippings[1] > 0) {
383 av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n",
384 n_clippings[0] + n_clippings[1], out->nb_samples * 2);
388 return ff_filter_frame(outlink, out);
391 static int convert_coeffs(AVFilterContext *ctx, AVFilterLink *inlink)
393 struct HeadphoneContext *s = ctx->priv;
394 const int ir_len = s->ir_len;
395 int nb_irs = s->nb_irs;
396 int nb_input_channels = ctx->inputs[0]->channels;
397 float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10);
398 FFTComplex *data_hrtf_l = NULL;
399 FFTComplex *data_hrtf_r = NULL;
400 FFTComplex *fft_in_l = NULL;
401 FFTComplex *fft_in_r = NULL;
402 float *data_ir_l = NULL;
403 float *data_ir_r = NULL;
404 int offset = 0, ret = 0;
408 s->buffer_length = 1 << (32 - ff_clz(s->ir_len));
409 s->n_fft = n_fft = 1 << (32 - ff_clz(s->ir_len + s->size));
411 if (s->type == FREQUENCY_DOMAIN) {
412 fft_in_l = av_calloc(n_fft, sizeof(*fft_in_l));
413 fft_in_r = av_calloc(n_fft, sizeof(*fft_in_r));
414 if (!fft_in_l || !fft_in_r) {
415 ret = AVERROR(ENOMEM);
419 av_fft_end(s->fft[0]);
420 av_fft_end(s->fft[1]);
421 s->fft[0] = av_fft_init(log2(s->n_fft), 0);
422 s->fft[1] = av_fft_init(log2(s->n_fft), 0);
423 av_fft_end(s->ifft[0]);
424 av_fft_end(s->ifft[1]);
425 s->ifft[0] = av_fft_init(log2(s->n_fft), 1);
426 s->ifft[1] = av_fft_init(log2(s->n_fft), 1);
428 if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {
429 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);
430 ret = AVERROR(ENOMEM);
435 s->data_ir[0] = av_calloc(FFALIGN(s->ir_len, 16), sizeof(float) * s->nb_irs);
436 s->data_ir[1] = av_calloc(FFALIGN(s->ir_len, 16), sizeof(float) * s->nb_irs);
437 s->delay[0] = av_calloc(s->nb_irs, sizeof(float));
438 s->delay[1] = av_calloc(s->nb_irs, sizeof(float));
440 if (s->type == TIME_DOMAIN) {
441 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
442 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
444 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float));
445 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));
446 s->temp_fft[0] = av_calloc(s->n_fft, sizeof(FFTComplex));
447 s->temp_fft[1] = av_calloc(s->n_fft, sizeof(FFTComplex));
448 if (!s->temp_fft[0] || !s->temp_fft[1]) {
449 ret = AVERROR(ENOMEM);
454 if (!s->data_ir[0] || !s->data_ir[1] ||
455 !s->ringbuffer[0] || !s->ringbuffer[1]) {
456 ret = AVERROR(ENOMEM);
460 for (i = 0; i < s->nb_inputs - 1; i++) {
461 s->in[i + 1].frame = ff_get_audio_buffer(ctx->inputs[i + 1], s->ir_len);
462 if (!s->in[i + 1].frame) {
463 ret = AVERROR(ENOMEM);
468 if (s->type == TIME_DOMAIN) {
469 s->temp_src[0] = av_calloc(FFALIGN(ir_len, 16), sizeof(float));
470 s->temp_src[1] = av_calloc(FFALIGN(ir_len, 16), sizeof(float));
472 data_ir_l = av_calloc(nb_irs * FFALIGN(ir_len, 16), sizeof(*data_ir_l));
473 data_ir_r = av_calloc(nb_irs * FFALIGN(ir_len, 16), sizeof(*data_ir_r));
474 if (!data_ir_r || !data_ir_l || !s->temp_src[0] || !s->temp_src[1]) {
475 ret = AVERROR(ENOMEM);
479 data_hrtf_l = av_calloc(n_fft, sizeof(*data_hrtf_l) * nb_irs);
480 data_hrtf_r = av_calloc(n_fft, sizeof(*data_hrtf_r) * nb_irs);
481 if (!data_hrtf_r || !data_hrtf_l) {
482 ret = AVERROR(ENOMEM);
487 for (i = 0; i < s->nb_inputs - 1; i++) {
488 int len = s->in[i + 1].ir_len;
489 int delay_l = s->in[i + 1].delay_l;
490 int delay_r = s->in[i + 1].delay_r;
493 av_audio_fifo_read(s->in[i + 1].fifo, (void **)s->in[i + 1].frame->extended_data, len);
494 ptr = (float *)s->in[i + 1].frame->extended_data[0];
496 if (s->hrir_fmt == HRIR_STEREO) {
499 for (j = 0; j < inlink->channels; j++) {
500 if (s->mapping[i] < 0) {
504 if ((av_channel_layout_extract_channel(inlink->channel_layout, j)) == (1LL << s->mapping[i])) {
512 if (s->type == TIME_DOMAIN) {
513 offset = idx * FFALIGN(len, 16);
514 for (j = 0; j < len; j++) {
515 data_ir_l[offset + j] = ptr[len * 2 - j * 2 - 2] * gain_lin;
516 data_ir_r[offset + j] = ptr[len * 2 - j * 2 - 1] * gain_lin;
519 memset(fft_in_l, 0, n_fft * sizeof(*fft_in_l));
520 memset(fft_in_r, 0, n_fft * sizeof(*fft_in_r));
522 offset = idx * n_fft;
523 for (j = 0; j < len; j++) {
524 fft_in_l[delay_l + j].re = ptr[j * 2 ] * gain_lin;
525 fft_in_r[delay_r + j].re = ptr[j * 2 + 1] * gain_lin;
528 av_fft_permute(s->fft[0], fft_in_l);
529 av_fft_calc(s->fft[0], fft_in_l);
530 memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l));
531 av_fft_permute(s->fft[0], fft_in_r);
532 av_fft_calc(s->fft[0], fft_in_r);
533 memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r));
536 int I, N = ctx->inputs[1]->channels;
538 for (k = 0; k < N / 2; k++) {
541 for (j = 0; j < inlink->channels; j++) {
542 if (s->mapping[k] < 0) {
546 if ((av_channel_layout_extract_channel(inlink->channel_layout, j)) == (1LL << s->mapping[k])) {
555 if (s->type == TIME_DOMAIN) {
556 offset = idx * FFALIGN(len, 16);
557 for (j = 0; j < len; j++) {
558 data_ir_l[offset + j] = ptr[len * N - j * N - N + I ] * gain_lin;
559 data_ir_r[offset + j] = ptr[len * N - j * N - N + I + 1] * gain_lin;
562 memset(fft_in_l, 0, n_fft * sizeof(*fft_in_l));
563 memset(fft_in_r, 0, n_fft * sizeof(*fft_in_r));
565 offset = idx * n_fft;
566 for (j = 0; j < len; j++) {
567 fft_in_l[delay_l + j].re = ptr[j * N + I ] * gain_lin;
568 fft_in_r[delay_r + j].re = ptr[j * N + I + 1] * gain_lin;
571 av_fft_permute(s->fft[0], fft_in_l);
572 av_fft_calc(s->fft[0], fft_in_l);
573 memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l));
574 av_fft_permute(s->fft[0], fft_in_r);
575 av_fft_calc(s->fft[0], fft_in_r);
576 memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r));
582 if (s->type == TIME_DOMAIN) {
583 memcpy(s->data_ir[0], data_ir_l, sizeof(float) * nb_irs * FFALIGN(ir_len, 16));
584 memcpy(s->data_ir[1], data_ir_r, sizeof(float) * nb_irs * FFALIGN(ir_len, 16));
586 s->data_hrtf[0] = av_calloc(n_fft * s->nb_irs, sizeof(FFTComplex));
587 s->data_hrtf[1] = av_calloc(n_fft * s->nb_irs, sizeof(FFTComplex));
588 if (!s->data_hrtf[0] || !s->data_hrtf[1]) {
589 ret = AVERROR(ENOMEM);
593 memcpy(s->data_hrtf[0], data_hrtf_l,
594 sizeof(FFTComplex) * nb_irs * n_fft);
595 memcpy(s->data_hrtf[1], data_hrtf_r,
596 sizeof(FFTComplex) * nb_irs * n_fft);
603 av_freep(&data_ir_l);
604 av_freep(&data_ir_r);
606 av_freep(&data_hrtf_l);
607 av_freep(&data_hrtf_r);
615 static int activate(AVFilterContext *ctx)
617 HeadphoneContext *s = ctx->priv;
618 AVFilterLink *inlink = ctx->inputs[0];
619 AVFilterLink *outlink = ctx->outputs[0];
623 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
625 for (i = 1; i < s->nb_inputs; i++) {
633 if ((ret = ff_inlink_consume_frame(ctx->inputs[i], &ir)) > 0) {
634 ret = read_ir(ctx->inputs[i], i, ir);
642 if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
643 if (status == AVERROR_EOF) {
650 for (i = 1; i < s->nb_inputs; i++) {
655 if (i != s->nb_inputs) {
656 if (ff_outlink_frame_wanted(ctx->outputs[0])) {
657 for (i = 1; i < s->nb_inputs; i++) {
659 ff_inlink_request_frame(ctx->inputs[i]);
668 if (!s->have_hrirs && s->eof_hrirs) {
669 ret = convert_coeffs(ctx, inlink);
674 if ((ret = ff_inlink_consume_samples(ctx->inputs[0], s->size, s->size, &in)) > 0) {
675 ret = headphone_frame(s, in, outlink);
683 FF_FILTER_FORWARD_STATUS(ctx->inputs[0], ctx->outputs[0]);
684 if (ff_outlink_frame_wanted(ctx->outputs[0]))
685 ff_inlink_request_frame(ctx->inputs[0]);
690 static int query_formats(AVFilterContext *ctx)
692 struct HeadphoneContext *s = ctx->priv;
693 AVFilterFormats *formats = NULL;
694 AVFilterChannelLayouts *layouts = NULL;
695 AVFilterChannelLayouts *stereo_layout = NULL;
696 AVFilterChannelLayouts *hrir_layouts = NULL;
699 ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
702 ret = ff_set_common_formats(ctx, formats);
706 layouts = ff_all_channel_layouts();
708 return AVERROR(ENOMEM);
710 ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts);
714 ret = ff_add_channel_layout(&stereo_layout, AV_CH_LAYOUT_STEREO);
718 if (s->hrir_fmt == HRIR_MULTI) {
719 hrir_layouts = ff_all_channel_counts();
721 ret = AVERROR(ENOMEM);
722 ret = ff_channel_layouts_ref(hrir_layouts, &ctx->inputs[1]->out_channel_layouts);
726 for (i = 1; i < s->nb_inputs; i++) {
727 ret = ff_channel_layouts_ref(stereo_layout, &ctx->inputs[i]->out_channel_layouts);
733 ret = ff_channel_layouts_ref(stereo_layout, &ctx->outputs[0]->in_channel_layouts);
737 formats = ff_all_samplerates();
739 return AVERROR(ENOMEM);
740 return ff_set_common_samplerates(ctx, formats);
743 static int config_input(AVFilterLink *inlink)
745 AVFilterContext *ctx = inlink->dst;
746 HeadphoneContext *s = ctx->priv;
748 if (s->nb_irs < inlink->channels) {
749 av_log(ctx, AV_LOG_ERROR, "Number of HRIRs must be >= %d.\n", inlink->channels);
750 return AVERROR(EINVAL);
756 static av_cold int init(AVFilterContext *ctx)
758 HeadphoneContext *s = ctx->priv;
763 .type = AVMEDIA_TYPE_AUDIO,
764 .config_props = config_input,
766 if ((ret = ff_insert_inpad(ctx, 0, &pad)) < 0)
770 av_log(ctx, AV_LOG_ERROR, "Valid mapping must be set.\n");
771 return AVERROR(EINVAL);
776 s->in = av_calloc(s->nb_inputs, sizeof(*s->in));
778 return AVERROR(ENOMEM);
780 for (i = 1; i < s->nb_inputs; i++) {
781 char *name = av_asprintf("hrir%d", i - 1);
784 .type = AVMEDIA_TYPE_AUDIO,
787 return AVERROR(ENOMEM);
788 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
794 s->fdsp = avpriv_float_dsp_alloc(0);
796 return AVERROR(ENOMEM);
801 static int config_output(AVFilterLink *outlink)
803 AVFilterContext *ctx = outlink->src;
804 HeadphoneContext *s = ctx->priv;
805 AVFilterLink *inlink = ctx->inputs[0];
808 if (s->hrir_fmt == HRIR_MULTI) {
809 AVFilterLink *hrir_link = ctx->inputs[1];
811 if (hrir_link->channels < inlink->channels * 2) {
812 av_log(ctx, AV_LOG_ERROR, "Number of channels in HRIR stream must be >= %d.\n", inlink->channels * 2);
813 return AVERROR(EINVAL);
817 for (i = 0; i < s->nb_inputs; i++) {
818 s->in[i].fifo = av_audio_fifo_alloc(ctx->inputs[i]->format, ctx->inputs[i]->channels, 1024);
820 return AVERROR(ENOMEM);
822 s->gain_lfe = expf((s->gain - 3 * inlink->channels - 6 + s->lfe_gain) / 20 * M_LN10);
827 static av_cold void uninit(AVFilterContext *ctx)
829 HeadphoneContext *s = ctx->priv;
832 av_fft_end(s->ifft[0]);
833 av_fft_end(s->ifft[1]);
834 av_fft_end(s->fft[0]);
835 av_fft_end(s->fft[1]);
836 av_freep(&s->delay[0]);
837 av_freep(&s->delay[1]);
838 av_freep(&s->data_ir[0]);
839 av_freep(&s->data_ir[1]);
840 av_freep(&s->ringbuffer[0]);
841 av_freep(&s->ringbuffer[1]);
842 av_freep(&s->temp_src[0]);
843 av_freep(&s->temp_src[1]);
844 av_freep(&s->temp_fft[0]);
845 av_freep(&s->temp_fft[1]);
846 av_freep(&s->data_hrtf[0]);
847 av_freep(&s->data_hrtf[1]);
850 for (i = 0; i < s->nb_inputs; i++) {
851 av_frame_free(&s->in[i].frame);
852 av_audio_fifo_free(s->in[i].fifo);
853 if (ctx->input_pads && i)
854 av_freep(&ctx->input_pads[i].name);
859 #define OFFSET(x) offsetof(HeadphoneContext, x)
860 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
862 static const AVOption headphone_options[] = {
863 { "map", "set channels convolution mappings", OFFSET(map), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
864 { "gain", "set gain in dB", OFFSET(gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
865 { "lfe", "set lfe gain in dB", OFFSET(lfe_gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
866 { "type", "set processing", OFFSET(type), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, .flags = FLAGS, "type" },
867 { "time", "time domain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, .flags = FLAGS, "type" },
868 { "freq", "frequency domain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, .flags = FLAGS, "type" },
869 { "size", "set frame size", OFFSET(size), AV_OPT_TYPE_INT, {.i64=1024},1024,96000, .flags = FLAGS },
870 { "hrir", "set hrir format", OFFSET(hrir_fmt), AV_OPT_TYPE_INT, {.i64=HRIR_STEREO}, 0, 1, .flags = FLAGS, "hrir" },
871 { "stereo", "hrir files have exactly 2 channels", 0, AV_OPT_TYPE_CONST, {.i64=HRIR_STEREO}, 0, 0, .flags = FLAGS, "hrir" },
872 { "multich", "single multichannel hrir file", 0, AV_OPT_TYPE_CONST, {.i64=HRIR_MULTI}, 0, 0, .flags = FLAGS, "hrir" },
876 AVFILTER_DEFINE_CLASS(headphone);
878 static const AVFilterPad outputs[] = {
881 .type = AVMEDIA_TYPE_AUDIO,
882 .config_props = config_output,
887 AVFilter ff_af_headphone = {
889 .description = NULL_IF_CONFIG_SMALL("Apply headphone binaural spatialization with HRTFs in additional streams."),
890 .priv_size = sizeof(HeadphoneContext),
891 .priv_class = &headphone_class,
894 .query_formats = query_formats,
895 .activate = activate,
898 .flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_DYNAMIC_INPUTS,