summaryrefslogtreecommitdiff
path: root/libavfilter/af_sofalizer.c (plain)
blob: 252f524c231d2abccd02701a5ae08d332b51aa07
1/*****************************************************************************
2 * sofalizer.c : SOFAlizer filter for virtual binaural acoustics
3 *****************************************************************************
4 * Copyright (C) 2013-2015 Andreas Fuchs, Wolfgang Hrauda,
5 * Acoustics Research Institute (ARI), Vienna, Austria
6 *
7 * Authors: Andreas Fuchs <andi.fuchs.mail@gmail.com>
8 * Wolfgang Hrauda <wolfgang.hrauda@gmx.at>
9 *
10 * SOFAlizer project coordinator at ARI, main developer of SOFA:
11 * Piotr Majdak <piotr@majdak.at>
12 *
13 * This program is free software; you can redistribute it and/or modify it
14 * under the terms of the GNU Lesser General Public License as published by
15 * the Free Software Foundation; either version 2.1 of the License, or
16 * (at your option) any later version.
17 *
18 * This program is distributed in the hope that it will be useful,
19 * but WITHOUT ANY WARRANTY; without even the implied warranty of
20 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
21 * GNU Lesser General Public License for more details.
22 *
23 * You should have received a copy of the GNU Lesser General Public License
24 * along with this program; if not, write to the Free Software Foundation,
25 * Inc., 51 Franklin Street, Fifth Floor, Boston MA 02110-1301, USA.
26 *****************************************************************************/
27
28#include <math.h>
29#include <netcdf.h>
30
31#include "libavcodec/avfft.h"
32#include "libavutil/avstring.h"
33#include "libavutil/channel_layout.h"
34#include "libavutil/float_dsp.h"
35#include "libavutil/intmath.h"
36#include "libavutil/opt.h"
37#include "avfilter.h"
38#include "internal.h"
39#include "audio.h"
40
41#define TIME_DOMAIN 0
42#define FREQUENCY_DOMAIN 1
43
44typedef struct NCSofa { /* contains data of one SOFA file */
45 int ncid; /* netCDF ID of the opened SOFA file */
46 int n_samples; /* length of one impulse response (IR) */
47 int m_dim; /* number of measurement positions */
48 int *data_delay; /* broadband delay of each IR */
49 /* all measurement positions for each receiver (i.e. ear): */
50 float *sp_a; /* azimuth angles */
51 float *sp_e; /* elevation angles */
52 float *sp_r; /* radii */
53 /* data at each measurement position for each receiver: */
54 float *data_ir; /* IRs (time-domain) */
55} NCSofa;
56
57typedef struct VirtualSpeaker {
58 uint8_t set;
59 float azim;
60 float elev;
61} VirtualSpeaker;
62
63typedef struct SOFAlizerContext {
64 const AVClass *class;
65
66 char *filename; /* name of SOFA file */
67 NCSofa sofa; /* contains data of the SOFA file */
68
69 int sample_rate; /* sample rate from SOFA file */
70 float *speaker_azim; /* azimuth of the virtual loudspeakers */
71 float *speaker_elev; /* elevation of the virtual loudspeakers */
72 char *speakers_pos; /* custom positions of the virtual loudspeakers */
73 float gain_lfe; /* gain applied to LFE channel */
74 int lfe_channel; /* LFE channel position in channel layout */
75
76 int n_conv; /* number of channels to convolute */
77
78 /* buffer variables (for convolution) */
79 float *ringbuffer[2]; /* buffers input samples, length of one buffer: */
80 /* no. input ch. (incl. LFE) x buffer_length */
81 int write[2]; /* current write position to ringbuffer */
82 int buffer_length; /* is: longest IR plus max. delay in all SOFA files */
83 /* then choose next power of 2 */
84 int n_fft; /* number of samples in one FFT block */
85
86 /* netCDF variables */
87 int *delay[2]; /* broadband delay for each channel/IR to be convolved */
88
89 float *data_ir[2]; /* IRs for all channels to be convolved */
90 /* (this excludes the LFE) */
91 float *temp_src[2];
92 FFTComplex *temp_fft[2];
93
94 /* control variables */
95 float gain; /* filter gain (in dB) */
96 float rotation; /* rotation of virtual loudspeakers (in degrees) */
97 float elevation; /* elevation of virtual loudspeakers (in deg.) */
98 float radius; /* distance virtual loudspeakers to listener (in metres) */
99 int type; /* processing type */
100
101 VirtualSpeaker vspkrpos[64];
102
103 FFTContext *fft[2], *ifft[2];
104 FFTComplex *data_hrtf[2];
105
106 AVFloatDSPContext *fdsp;
107} SOFAlizerContext;
108
109static int close_sofa(struct NCSofa *sofa)
110{
111 av_freep(&sofa->data_delay);
112 av_freep(&sofa->sp_a);
113 av_freep(&sofa->sp_e);
114 av_freep(&sofa->sp_r);
115 av_freep(&sofa->data_ir);
116 nc_close(sofa->ncid);
117 sofa->ncid = 0;
118
119 return 0;
120}
121
122static int load_sofa(AVFilterContext *ctx, char *filename, int *samplingrate)
123{
124 struct SOFAlizerContext *s = ctx->priv;
125 /* variables associated with content of SOFA file: */
126 int ncid, n_dims, n_vars, n_gatts, n_unlim_dim_id, status;
127 char data_delay_dim_name[NC_MAX_NAME];
128 float *sp_a, *sp_e, *sp_r, *data_ir;
129 char *sofa_conventions;
130 char dim_name[NC_MAX_NAME]; /* names of netCDF dimensions */
131 size_t *dim_length; /* lengths of netCDF dimensions */
132 char *text;
133 unsigned int sample_rate;
134 int data_delay_dim_id[2];
135 int samplingrate_id;
136 int data_delay_id;
137 int n_samples;
138 int m_dim_id = -1;
139 int n_dim_id = -1;
140 int data_ir_id;
141 size_t att_len;
142 int m_dim;
143 int *data_delay;
144 int sp_id;
145 int i, ret;
146
147 s->sofa.ncid = 0;
148 status = nc_open(filename, NC_NOWRITE, &ncid); /* open SOFA file read-only */
149 if (status != NC_NOERR) {
150 av_log(ctx, AV_LOG_ERROR, "Can't find SOFA-file '%s'\n", filename);
151 return AVERROR(EINVAL);
152 }
153
154 /* get number of dimensions, vars, global attributes and Id of unlimited dimensions: */
155 nc_inq(ncid, &n_dims, &n_vars, &n_gatts, &n_unlim_dim_id);
156
157 /* -- get number of measurements ("M") and length of one IR ("N") -- */
158 dim_length = av_malloc_array(n_dims, sizeof(*dim_length));
159 if (!dim_length) {
160 nc_close(ncid);
161 return AVERROR(ENOMEM);
162 }
163
164 for (i = 0; i < n_dims; i++) { /* go through all dimensions of file */
165 nc_inq_dim(ncid, i, (char *)&dim_name, &dim_length[i]); /* get dimensions */
166 if (!strncmp("M", (const char *)&dim_name, 1)) /* get ID of dimension "M" */
167 m_dim_id = i;
168 if (!strncmp("N", (const char *)&dim_name, 1)) /* get ID of dimension "N" */
169 n_dim_id = i;
170 }
171
172 if ((m_dim_id == -1) || (n_dim_id == -1)) { /* dimension "M" or "N" couldn't be found */
173 av_log(ctx, AV_LOG_ERROR, "Can't find required dimensions in SOFA file.\n");
174 av_freep(&dim_length);
175 nc_close(ncid);
176 return AVERROR(EINVAL);
177 }
178
179 n_samples = dim_length[n_dim_id]; /* get length of one IR */
180 m_dim = dim_length[m_dim_id]; /* get number of measurements */
181
182 av_freep(&dim_length);
183
184 /* -- check file type -- */
185 /* get length of attritube "Conventions" */
186 status = nc_inq_attlen(ncid, NC_GLOBAL, "Conventions", &att_len);
187 if (status != NC_NOERR) {
188 av_log(ctx, AV_LOG_ERROR, "Can't get length of attribute \"Conventions\".\n");
189 nc_close(ncid);
190 return AVERROR_INVALIDDATA;
191 }
192
193 /* check whether file is SOFA file */
194 text = av_malloc(att_len + 1);
195 if (!text) {
196 nc_close(ncid);
197 return AVERROR(ENOMEM);
198 }
199
200 nc_get_att_text(ncid, NC_GLOBAL, "Conventions", text);
201 *(text + att_len) = 0;
202 if (strncmp("SOFA", text, 4)) {
203 av_log(ctx, AV_LOG_ERROR, "Not a SOFA file!\n");
204 av_freep(&text);
205 nc_close(ncid);
206 return AVERROR(EINVAL);
207 }
208 av_freep(&text);
209
210 status = nc_inq_attlen(ncid, NC_GLOBAL, "License", &att_len);
211 if (status == NC_NOERR) {
212 text = av_malloc(att_len + 1);
213 if (text) {
214 nc_get_att_text(ncid, NC_GLOBAL, "License", text);
215 *(text + att_len) = 0;
216 av_log(ctx, AV_LOG_INFO, "SOFA file License: %s\n", text);
217 av_freep(&text);
218 }
219 }
220
221 status = nc_inq_attlen(ncid, NC_GLOBAL, "SourceDescription", &att_len);
222 if (status == NC_NOERR) {
223 text = av_malloc(att_len + 1);
224 if (text) {
225 nc_get_att_text(ncid, NC_GLOBAL, "SourceDescription", text);
226 *(text + att_len) = 0;
227 av_log(ctx, AV_LOG_INFO, "SOFA file SourceDescription: %s\n", text);
228 av_freep(&text);
229 }
230 }
231
232 status = nc_inq_attlen(ncid, NC_GLOBAL, "Comment", &att_len);
233 if (status == NC_NOERR) {
234 text = av_malloc(att_len + 1);
235 if (text) {
236 nc_get_att_text(ncid, NC_GLOBAL, "Comment", text);
237 *(text + att_len) = 0;
238 av_log(ctx, AV_LOG_INFO, "SOFA file Comment: %s\n", text);
239 av_freep(&text);
240 }
241 }
242
243 status = nc_inq_attlen(ncid, NC_GLOBAL, "SOFAConventions", &att_len);
244 if (status != NC_NOERR) {
245 av_log(ctx, AV_LOG_ERROR, "Can't get length of attribute \"SOFAConventions\".\n");
246 nc_close(ncid);
247 return AVERROR_INVALIDDATA;
248 }
249
250 sofa_conventions = av_malloc(att_len + 1);
251 if (!sofa_conventions) {
252 nc_close(ncid);
253 return AVERROR(ENOMEM);
254 }
255
256 nc_get_att_text(ncid, NC_GLOBAL, "SOFAConventions", sofa_conventions);
257 *(sofa_conventions + att_len) = 0;
258 if (strncmp("SimpleFreeFieldHRIR", sofa_conventions, att_len)) {
259 av_log(ctx, AV_LOG_ERROR, "Not a SimpleFreeFieldHRIR file!\n");
260 av_freep(&sofa_conventions);
261 nc_close(ncid);
262 return AVERROR(EINVAL);
263 }
264 av_freep(&sofa_conventions);
265
266 /* -- get sampling rate of HRTFs -- */
267 /* read ID, then value */
268 status = nc_inq_varid(ncid, "Data.SamplingRate", &samplingrate_id);
269 status += nc_get_var_uint(ncid, samplingrate_id, &sample_rate);
270 if (status != NC_NOERR) {
271 av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.SamplingRate.\n");
272 nc_close(ncid);
273 return AVERROR(EINVAL);
274 }
275 *samplingrate = sample_rate; /* remember sampling rate */
276
277 /* -- allocate memory for one value for each measurement position: -- */
278 sp_a = s->sofa.sp_a = av_malloc_array(m_dim, sizeof(float));
279 sp_e = s->sofa.sp_e = av_malloc_array(m_dim, sizeof(float));
280 sp_r = s->sofa.sp_r = av_malloc_array(m_dim, sizeof(float));
281 /* delay and IR values required for each ear and measurement position: */
282 data_delay = s->sofa.data_delay = av_calloc(m_dim, 2 * sizeof(int));
283 data_ir = s->sofa.data_ir = av_calloc(m_dim * FFALIGN(n_samples, 16), sizeof(float) * 2);
284
285 if (!data_delay || !sp_a || !sp_e || !sp_r || !data_ir) {
286 /* if memory could not be allocated */
287 close_sofa(&s->sofa);
288 return AVERROR(ENOMEM);
289 }
290
291 /* get impulse responses (HRTFs): */
292 /* get corresponding ID */
293 status = nc_inq_varid(ncid, "Data.IR", &data_ir_id);
294 status += nc_get_var_float(ncid, data_ir_id, data_ir); /* read and store IRs */
295 if (status != NC_NOERR) {
296 av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.IR!\n");
297 ret = AVERROR(EINVAL);
298 goto error;
299 }
300
301 /* get source positions of the HRTFs in the SOFA file: */
302 status = nc_inq_varid(ncid, "SourcePosition", &sp_id); /* get corresponding ID */
303 status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 0 } ,
304 (size_t[2]){ m_dim, 1}, sp_a); /* read & store azimuth angles */
305 status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 1 } ,
306 (size_t[2]){ m_dim, 1}, sp_e); /* read & store elevation angles */
307 status += nc_get_vara_float(ncid, sp_id, (size_t[2]){ 0, 2 } ,
308 (size_t[2]){ m_dim, 1}, sp_r); /* read & store radii */
309 if (status != NC_NOERR) { /* if any source position variable coudn't be read */
310 av_log(ctx, AV_LOG_ERROR, "Couldn't read SourcePosition.\n");
311 ret = AVERROR(EINVAL);
312 goto error;
313 }
314
315 /* read Data.Delay, check for errors and fit it to data_delay */
316 status = nc_inq_varid(ncid, "Data.Delay", &data_delay_id);
317 status += nc_inq_vardimid(ncid, data_delay_id, &data_delay_dim_id[0]);
318 status += nc_inq_dimname(ncid, data_delay_dim_id[0], data_delay_dim_name);
319 if (status != NC_NOERR) {
320 av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay.\n");
321 ret = AVERROR(EINVAL);
322 goto error;
323 }
324
325 /* Data.Delay dimension check */
326 /* dimension of Data.Delay is [I R]: */
327 if (!strncmp(data_delay_dim_name, "I", 2)) {
328 /* check 2 characters to assure string is 0-terminated after "I" */
329 int delay[2]; /* delays get from SOFA file: */
330 int *data_delay_r;
331
332 av_log(ctx, AV_LOG_DEBUG, "Data.Delay has dimension [I R]\n");
333 status = nc_get_var_int(ncid, data_delay_id, &delay[0]);
334 if (status != NC_NOERR) {
335 av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay\n");
336 ret = AVERROR(EINVAL);
337 goto error;
338 }
339 data_delay_r = data_delay + m_dim;
340 for (i = 0; i < m_dim; i++) { /* extend given dimension [I R] to [M R] */
341 /* assign constant delay value for all measurements to data_delay fields */
342 data_delay[i] = delay[0];
343 data_delay_r[i] = delay[1];
344 }
345 /* dimension of Data.Delay is [M R] */
346 } else if (!strncmp(data_delay_dim_name, "M", 2)) {
347 av_log(ctx, AV_LOG_ERROR, "Data.Delay in dimension [M R]\n");
348 /* get delays from SOFA file: */
349 status = nc_get_var_int(ncid, data_delay_id, data_delay);
350 if (status != NC_NOERR) {
351 av_log(ctx, AV_LOG_ERROR, "Couldn't read Data.Delay\n");
352 ret = AVERROR(EINVAL);
353 goto error;
354 }
355 } else { /* dimension of Data.Delay is neither [I R] nor [M R] */
356 av_log(ctx, AV_LOG_ERROR, "Data.Delay does not have the required dimensions [I R] or [M R].\n");
357 ret = AVERROR(EINVAL);
358 goto error;
359 }
360
361 /* save information in SOFA struct: */
362 s->sofa.m_dim = m_dim; /* no. measurement positions */
363 s->sofa.n_samples = n_samples; /* length on one IR */
364 s->sofa.ncid = ncid; /* netCDF ID of SOFA file */
365 nc_close(ncid); /* close SOFA file */
366
367 av_log(ctx, AV_LOG_DEBUG, "m_dim: %d n_samples %d\n", m_dim, n_samples);
368
369 return 0;
370
371error:
372 close_sofa(&s->sofa);
373 return ret;
374}
375
376static int parse_channel_name(char **arg, int *rchannel, char *buf)
377{
378 int len, i, channel_id = 0;
379 int64_t layout, layout0;
380
381 /* try to parse a channel name, e.g. "FL" */
382 if (sscanf(*arg, "%7[A-Z]%n", buf, &len)) {
383 layout0 = layout = av_get_channel_layout(buf);
384 /* channel_id <- first set bit in layout */
385 for (i = 32; i > 0; i >>= 1) {
386 if (layout >= (int64_t)1 << i) {
387 channel_id += i;
388 layout >>= i;
389 }
390 }
391 /* reject layouts that are not a single channel */
392 if (channel_id >= 64 || layout0 != (int64_t)1 << channel_id)
393 return AVERROR(EINVAL);
394 *rchannel = channel_id;
395 *arg += len;
396 return 0;
397 }
398 return AVERROR(EINVAL);
399}
400
401static void parse_speaker_pos(AVFilterContext *ctx, int64_t in_channel_layout)
402{
403 SOFAlizerContext *s = ctx->priv;
404 char *arg, *tokenizer, *p, *args = av_strdup(s->speakers_pos);
405
406 if (!args)
407 return;
408 p = args;
409
410 while ((arg = av_strtok(p, "|", &tokenizer))) {
411 char buf[8];
412 float azim, elev;
413 int out_ch_id;
414
415 p = NULL;
416 if (parse_channel_name(&arg, &out_ch_id, buf)) {
417 av_log(ctx, AV_LOG_WARNING, "Failed to parse \'%s\' as channel name.\n", buf);
418 continue;
419 }
420 if (sscanf(arg, "%f %f", &azim, &elev) == 2) {
421 s->vspkrpos[out_ch_id].set = 1;
422 s->vspkrpos[out_ch_id].azim = azim;
423 s->vspkrpos[out_ch_id].elev = elev;
424 } else if (sscanf(arg, "%f", &azim) == 1) {
425 s->vspkrpos[out_ch_id].set = 1;
426 s->vspkrpos[out_ch_id].azim = azim;
427 s->vspkrpos[out_ch_id].elev = 0;
428 }
429 }
430
431 av_free(args);
432}
433
434static int get_speaker_pos(AVFilterContext *ctx,
435 float *speaker_azim, float *speaker_elev)
436{
437 struct SOFAlizerContext *s = ctx->priv;
438 uint64_t channels_layout = ctx->inputs[0]->channel_layout;
439 float azim[16] = { 0 };
440 float elev[16] = { 0 };
441 int m, ch, n_conv = ctx->inputs[0]->channels; /* get no. input channels */
442
443 if (n_conv > 16)
444 return AVERROR(EINVAL);
445
446 s->lfe_channel = -1;
447
448 if (s->speakers_pos)
449 parse_speaker_pos(ctx, channels_layout);
450
451 /* set speaker positions according to input channel configuration: */
452 for (m = 0, ch = 0; ch < n_conv && m < 64; m++) {
453 uint64_t mask = channels_layout & (1ULL << m);
454
455 switch (mask) {
456 case AV_CH_FRONT_LEFT: azim[ch] = 30; break;
457 case AV_CH_FRONT_RIGHT: azim[ch] = 330; break;
458 case AV_CH_FRONT_CENTER: azim[ch] = 0; break;
459 case AV_CH_LOW_FREQUENCY:
460 case AV_CH_LOW_FREQUENCY_2: s->lfe_channel = ch; break;
461 case AV_CH_BACK_LEFT: azim[ch] = 150; break;
462 case AV_CH_BACK_RIGHT: azim[ch] = 210; break;
463 case AV_CH_BACK_CENTER: azim[ch] = 180; break;
464 case AV_CH_SIDE_LEFT: azim[ch] = 90; break;
465 case AV_CH_SIDE_RIGHT: azim[ch] = 270; break;
466 case AV_CH_FRONT_LEFT_OF_CENTER: azim[ch] = 15; break;
467 case AV_CH_FRONT_RIGHT_OF_CENTER: azim[ch] = 345; break;
468 case AV_CH_TOP_CENTER: azim[ch] = 0;
469 elev[ch] = 90; break;
470 case AV_CH_TOP_FRONT_LEFT: azim[ch] = 30;
471 elev[ch] = 45; break;
472 case AV_CH_TOP_FRONT_CENTER: azim[ch] = 0;
473 elev[ch] = 45; break;
474 case AV_CH_TOP_FRONT_RIGHT: azim[ch] = 330;
475 elev[ch] = 45; break;
476 case AV_CH_TOP_BACK_LEFT: azim[ch] = 150;
477 elev[ch] = 45; break;
478 case AV_CH_TOP_BACK_RIGHT: azim[ch] = 210;
479 elev[ch] = 45; break;
480 case AV_CH_TOP_BACK_CENTER: azim[ch] = 180;
481 elev[ch] = 45; break;
482 case AV_CH_WIDE_LEFT: azim[ch] = 90; break;
483 case AV_CH_WIDE_RIGHT: azim[ch] = 270; break;
484 case AV_CH_SURROUND_DIRECT_LEFT: azim[ch] = 90; break;
485 case AV_CH_SURROUND_DIRECT_RIGHT: azim[ch] = 270; break;
486 case AV_CH_STEREO_LEFT: azim[ch] = 90; break;
487 case AV_CH_STEREO_RIGHT: azim[ch] = 270; break;
488 case 0: break;
489 default:
490 return AVERROR(EINVAL);
491 }
492
493 if (s->vspkrpos[m].set) {
494 azim[ch] = s->vspkrpos[m].azim;
495 elev[ch] = s->vspkrpos[m].elev;
496 }
497
498 if (mask)
499 ch++;
500 }
501
502 memcpy(speaker_azim, azim, n_conv * sizeof(float));
503 memcpy(speaker_elev, elev, n_conv * sizeof(float));
504
505 return 0;
506
507}
508
509static int max_delay(struct NCSofa *sofa)
510{
511 int i, max = 0;
512
513 for (i = 0; i < sofa->m_dim * 2; i++) {
514 /* search maximum delay in given SOFA file */
515 max = FFMAX(max, sofa->data_delay[i]);
516 }
517
518 return max;
519}
520
521static int find_m(SOFAlizerContext *s, int azim, int elev, float radius)
522{
523 /* get source positions and M of currently selected SOFA file */
524 float *sp_a = s->sofa.sp_a; /* azimuth angle */
525 float *sp_e = s->sofa.sp_e; /* elevation angle */
526 float *sp_r = s->sofa.sp_r; /* radius */
527 int m_dim = s->sofa.m_dim; /* no. measurements */
528 int best_id = 0; /* index m currently closest to desired source pos. */
529 float delta = 1000; /* offset between desired and currently best pos. */
530 float current;
531 int i;
532
533 for (i = 0; i < m_dim; i++) {
534 /* search through all measurements in currently selected SOFA file */
535 /* distance of current to desired source position: */
536 current = fabs(sp_a[i] - azim) +
537 fabs(sp_e[i] - elev) +
538 fabs(sp_r[i] - radius);
539 if (current <= delta) {
540 /* if current distance is smaller than smallest distance so far */
541 delta = current;
542 best_id = i; /* remember index */
543 }
544 }
545
546 return best_id;
547}
548
549static int compensate_volume(AVFilterContext *ctx)
550{
551 struct SOFAlizerContext *s = ctx->priv;
552 float compensate;
553 float energy = 0;
554 float *ir;
555 int m;
556
557 if (s->sofa.ncid) {
558 /* find IR at front center position in the SOFA file (IR closest to 0°,0°,1m) */
559 struct NCSofa *sofa = &s->sofa;
560 m = find_m(s, 0, 0, 1);
561 /* get energy of that IR and compensate volume */
562 ir = sofa->data_ir + 2 * m * sofa->n_samples;
563 if (sofa->n_samples & 31) {
564 energy = avpriv_scalarproduct_float_c(ir, ir, sofa->n_samples);
565 } else {
566 energy = s->fdsp->scalarproduct_float(ir, ir, sofa->n_samples);
567 }
568 compensate = 256 / (sofa->n_samples * sqrt(energy));
569 av_log(ctx, AV_LOG_DEBUG, "Compensate-factor: %f\n", compensate);
570 ir = sofa->data_ir;
571 /* apply volume compensation to IRs */
572 if (sofa->n_samples & 31) {
573 int i;
574 for (i = 0; i < sofa->n_samples * sofa->m_dim * 2; i++) {
575 ir[i] = ir[i] * compensate;
576 }
577 } else {
578 s->fdsp->vector_fmul_scalar(ir, ir, compensate, sofa->n_samples * sofa->m_dim * 2);
579 emms_c();
580 }
581 }
582
583 return 0;
584}
585
586typedef struct ThreadData {
587 AVFrame *in, *out;
588 int *write;
589 int **delay;
590 float **ir;
591 int *n_clippings;
592 float **ringbuffer;
593 float **temp_src;
594 FFTComplex **temp_fft;
595} ThreadData;
596
597static int sofalizer_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
598{
599 SOFAlizerContext *s = ctx->priv;
600 ThreadData *td = arg;
601 AVFrame *in = td->in, *out = td->out;
602 int offset = jobnr;
603 int *write = &td->write[jobnr];
604 const int *const delay = td->delay[jobnr];
605 const float *const ir = td->ir[jobnr];
606 int *n_clippings = &td->n_clippings[jobnr];
607 float *ringbuffer = td->ringbuffer[jobnr];
608 float *temp_src = td->temp_src[jobnr];
609 const int n_samples = s->sofa.n_samples; /* length of one IR */
610 const float *src = (const float *)in->data[0]; /* get pointer to audio input buffer */
611 float *dst = (float *)out->data[0]; /* get pointer to audio output buffer */
612 const int in_channels = s->n_conv; /* number of input channels */
613 /* ring buffer length is: longest IR plus max. delay -> next power of 2 */
614 const int buffer_length = s->buffer_length;
615 /* -1 for AND instead of MODULO (applied to powers of 2): */
616 const uint32_t modulo = (uint32_t)buffer_length - 1;
617 float *buffer[16]; /* holds ringbuffer for each input channel */
618 int wr = *write;
619 int read;
620 int i, l;
621
622 dst += offset;
623 for (l = 0; l < in_channels; l++) {
624 /* get starting address of ringbuffer for each input channel */
625 buffer[l] = ringbuffer + l * buffer_length;
626 }
627
628 for (i = 0; i < in->nb_samples; i++) {
629 const float *temp_ir = ir; /* using same set of IRs for each sample */
630
631 *dst = 0;
632 for (l = 0; l < in_channels; l++) {
633 /* write current input sample to ringbuffer (for each channel) */
634 *(buffer[l] + wr) = src[l];
635 }
636
637 /* loop goes through all channels to be convolved */
638 for (l = 0; l < in_channels; l++) {
639 const float *const bptr = buffer[l];
640
641 if (l == s->lfe_channel) {
642 /* LFE is an input channel but requires no convolution */
643 /* apply gain to LFE signal and add to output buffer */
644 *dst += *(buffer[s->lfe_channel] + wr) * s->gain_lfe;
645 temp_ir += FFALIGN(n_samples, 16);
646 continue;
647 }
648
649 /* current read position in ringbuffer: input sample write position
650 * - delay for l-th ch. + diff. betw. IR length and buffer length
651 * (mod buffer length) */
652 read = (wr - *(delay + l) - (n_samples - 1) + buffer_length) & modulo;
653
654 if (read + n_samples < buffer_length) {
655 memcpy(temp_src, bptr + read, n_samples * sizeof(*temp_src));
656 } else {
657 int len = FFMIN(n_samples - (read % n_samples), buffer_length - read);
658
659 memcpy(temp_src, bptr + read, len * sizeof(*temp_src));
660 memcpy(temp_src + len, bptr, (n_samples - len) * sizeof(*temp_src));
661 }
662
663 /* multiply signal and IR, and add up the results */
664 dst[0] += s->fdsp->scalarproduct_float(temp_ir, temp_src, n_samples);
665 temp_ir += FFALIGN(n_samples, 16);
666 }
667
668 /* clippings counter */
669 if (fabs(*dst) > 1)
670 *n_clippings += 1;
671
672 /* move output buffer pointer by +2 to get to next sample of processed channel: */
673 dst += 2;
674 src += in_channels;
675 wr = (wr + 1) & modulo; /* update ringbuffer write position */
676 }
677
678 *write = wr; /* remember write position in ringbuffer for next call */
679
680 return 0;
681}
682
683static int sofalizer_fast_convolute(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
684{
685 SOFAlizerContext *s = ctx->priv;
686 ThreadData *td = arg;
687 AVFrame *in = td->in, *out = td->out;
688 int offset = jobnr;
689 int *write = &td->write[jobnr];
690 FFTComplex *hrtf = s->data_hrtf[jobnr]; /* get pointers to current HRTF data */
691 int *n_clippings = &td->n_clippings[jobnr];
692 float *ringbuffer = td->ringbuffer[jobnr];
693 const int n_samples = s->sofa.n_samples; /* length of one IR */
694 const float *src = (const float *)in->data[0]; /* get pointer to audio input buffer */
695 float *dst = (float *)out->data[0]; /* get pointer to audio output buffer */
696 const int in_channels = s->n_conv; /* number of input channels */
697 /* ring buffer length is: longest IR plus max. delay -> next power of 2 */
698 const int buffer_length = s->buffer_length;
699 /* -1 for AND instead of MODULO (applied to powers of 2): */
700 const uint32_t modulo = (uint32_t)buffer_length - 1;
701 FFTComplex *fft_in = s->temp_fft[jobnr]; /* temporary array for FFT input/output data */
702 FFTContext *ifft = s->ifft[jobnr];
703 FFTContext *fft = s->fft[jobnr];
704 const int n_conv = s->n_conv;
705 const int n_fft = s->n_fft;
706 const float fft_scale = 1.0f / s->n_fft;
707 FFTComplex *hrtf_offset;
708 int wr = *write;
709 int n_read;
710 int i, j;
711
712 dst += offset;
713
714 /* find minimum between number of samples and output buffer length:
715 * (important, if one IR is longer than the output buffer) */
716 n_read = FFMIN(s->sofa.n_samples, in->nb_samples);
717 for (j = 0; j < n_read; j++) {
718 /* initialize output buf with saved signal from overflow buf */
719 dst[2 * j] = ringbuffer[wr];
720 ringbuffer[wr] = 0.0; /* re-set read samples to zero */
721 /* update ringbuffer read/write position */
722 wr = (wr + 1) & modulo;
723 }
724
725 /* initialize rest of output buffer with 0 */
726 for (j = n_read; j < in->nb_samples; j++) {
727 dst[2 * j] = 0;
728 }
729
730 for (i = 0; i < n_conv; i++) {
731 if (i == s->lfe_channel) { /* LFE */
732 for (j = 0; j < in->nb_samples; j++) {
733 /* apply gain to LFE signal and add to output buffer */
734 dst[2 * j] += src[i + j * in_channels] * s->gain_lfe;
735 }
736 continue;
737 }
738
739 /* outer loop: go through all input channels to be convolved */
740 offset = i * n_fft; /* no. samples already processed */
741 hrtf_offset = hrtf + offset;
742
743 /* fill FFT input with 0 (we want to zero-pad) */
744 memset(fft_in, 0, sizeof(FFTComplex) * n_fft);
745
746 for (j = 0; j < in->nb_samples; j++) {
747 /* prepare input for FFT */
748 /* write all samples of current input channel to FFT input array */
749 fft_in[j].re = src[j * in_channels + i];
750 }
751
752 /* transform input signal of current channel to frequency domain */
753 av_fft_permute(fft, fft_in);
754 av_fft_calc(fft, fft_in);
755 for (j = 0; j < n_fft; j++) {
756 const FFTComplex *hcomplex = hrtf_offset + j;
757 const float re = fft_in[j].re;
758 const float im = fft_in[j].im;
759
760 /* complex multiplication of input signal and HRTFs */
761 /* output channel (real): */
762 fft_in[j].re = re * hcomplex->re - im * hcomplex->im;
763 /* output channel (imag): */
764 fft_in[j].im = re * hcomplex->im + im * hcomplex->re;
765 }
766
767 /* transform output signal of current channel back to time domain */
768 av_fft_permute(ifft, fft_in);
769 av_fft_calc(ifft, fft_in);
770
771 for (j = 0; j < in->nb_samples; j++) {
772 /* write output signal of current channel to output buffer */
773 dst[2 * j] += fft_in[j].re * fft_scale;
774 }
775
776 for (j = 0; j < n_samples - 1; j++) { /* overflow length is IR length - 1 */
777 /* write the rest of output signal to overflow buffer */
778 int write_pos = (wr + j) & modulo;
779
780 *(ringbuffer + write_pos) += fft_in[in->nb_samples + j].re * fft_scale;
781 }
782 }
783
784 /* go through all samples of current output buffer: count clippings */
785 for (i = 0; i < out->nb_samples; i++) {
786 /* clippings counter */
787 if (fabs(*dst) > 1) { /* if current output sample > 1 */
788 n_clippings[0]++;
789 }
790
791 /* move output buffer pointer by +2 to get to next sample of processed channel: */
792 dst += 2;
793 }
794
795 /* remember read/write position in ringbuffer for next call */
796 *write = wr;
797
798 return 0;
799}
800
801static int filter_frame(AVFilterLink *inlink, AVFrame *in)
802{
803 AVFilterContext *ctx = inlink->dst;
804 SOFAlizerContext *s = ctx->priv;
805 AVFilterLink *outlink = ctx->outputs[0];
806 int n_clippings[2] = { 0 };
807 ThreadData td;
808 AVFrame *out;
809
810 out = ff_get_audio_buffer(outlink, in->nb_samples);
811 if (!out) {
812 av_frame_free(&in);
813 return AVERROR(ENOMEM);
814 }
815 av_frame_copy_props(out, in);
816
817 td.in = in; td.out = out; td.write = s->write;
818 td.delay = s->delay; td.ir = s->data_ir; td.n_clippings = n_clippings;
819 td.ringbuffer = s->ringbuffer; td.temp_src = s->temp_src;
820 td.temp_fft = s->temp_fft;
821
822 if (s->type == TIME_DOMAIN) {
823 ctx->internal->execute(ctx, sofalizer_convolute, &td, NULL, 2);
824 } else {
825 ctx->internal->execute(ctx, sofalizer_fast_convolute, &td, NULL, 2);
826 }
827 emms_c();
828
829 /* display error message if clipping occurred */
830 if (n_clippings[0] + n_clippings[1] > 0) {
831 av_log(ctx, AV_LOG_WARNING, "%d of %d samples clipped. Please reduce gain.\n",
832 n_clippings[0] + n_clippings[1], out->nb_samples * 2);
833 }
834
835 av_frame_free(&in);
836 return ff_filter_frame(outlink, out);
837}
838
839static int query_formats(AVFilterContext *ctx)
840{
841 struct SOFAlizerContext *s = ctx->priv;
842 AVFilterFormats *formats = NULL;
843 AVFilterChannelLayouts *layouts = NULL;
844 int ret, sample_rates[] = { 48000, -1 };
845
846 ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
847 if (ret)
848 return ret;
849 ret = ff_set_common_formats(ctx, formats);
850 if (ret)
851 return ret;
852
853 layouts = ff_all_channel_layouts();
854 if (!layouts)
855 return AVERROR(ENOMEM);
856
857 ret = ff_channel_layouts_ref(layouts, &ctx->inputs[0]->out_channel_layouts);
858 if (ret)
859 return ret;
860
861 layouts = NULL;
862 ret = ff_add_channel_layout(&layouts, AV_CH_LAYOUT_STEREO);
863 if (ret)
864 return ret;
865
866 ret = ff_channel_layouts_ref(layouts, &ctx->outputs[0]->in_channel_layouts);
867 if (ret)
868 return ret;
869
870 sample_rates[0] = s->sample_rate;
871 formats = ff_make_format_list(sample_rates);
872 if (!formats)
873 return AVERROR(ENOMEM);
874 return ff_set_common_samplerates(ctx, formats);
875}
876
877static int load_data(AVFilterContext *ctx, int azim, int elev, float radius)
878{
879 struct SOFAlizerContext *s = ctx->priv;
880 const int n_samples = s->sofa.n_samples;
881 int n_conv = s->n_conv; /* no. channels to convolve */
882 int n_fft = s->n_fft;
883 int delay_l[16]; /* broadband delay for each IR */
884 int delay_r[16];
885 int nb_input_channels = ctx->inputs[0]->channels; /* no. input channels */
886 float gain_lin = expf((s->gain - 3 * nb_input_channels) / 20 * M_LN10); /* gain - 3dB/channel */
887 FFTComplex *data_hrtf_l = NULL;
888 FFTComplex *data_hrtf_r = NULL;
889 FFTComplex *fft_in_l = NULL;
890 FFTComplex *fft_in_r = NULL;
891 float *data_ir_l = NULL;
892 float *data_ir_r = NULL;
893 int offset = 0; /* used for faster pointer arithmetics in for-loop */
894 int m[16]; /* measurement index m of IR closest to required source positions */
895 int i, j, azim_orig = azim, elev_orig = elev;
896
897 if (!s->sofa.ncid) { /* if an invalid SOFA file has been selected */
898 av_log(ctx, AV_LOG_ERROR, "Selected SOFA file is invalid. Please select valid SOFA file.\n");
899 return AVERROR_INVALIDDATA;
900 }
901
902 if (s->type == TIME_DOMAIN) {
903 s->temp_src[0] = av_calloc(FFALIGN(n_samples, 16), sizeof(float));
904 s->temp_src[1] = av_calloc(FFALIGN(n_samples, 16), sizeof(float));
905
906 /* get temporary IR for L and R channel */
907 data_ir_l = av_calloc(n_conv * FFALIGN(n_samples, 16), sizeof(*data_ir_l));
908 data_ir_r = av_calloc(n_conv * FFALIGN(n_samples, 16), sizeof(*data_ir_r));
909 if (!data_ir_r || !data_ir_l || !s->temp_src[0] || !s->temp_src[1]) {
910 av_free(data_ir_l);
911 av_free(data_ir_r);
912 return AVERROR(ENOMEM);
913 }
914 } else {
915 /* get temporary HRTF memory for L and R channel */
916 data_hrtf_l = av_malloc_array(n_fft, sizeof(*data_hrtf_l) * n_conv);
917 data_hrtf_r = av_malloc_array(n_fft, sizeof(*data_hrtf_r) * n_conv);
918 if (!data_hrtf_r || !data_hrtf_l) {
919 av_free(data_hrtf_l);
920 av_free(data_hrtf_r);
921 return AVERROR(ENOMEM);
922 }
923 }
924
925 for (i = 0; i < s->n_conv; i++) {
926 /* load and store IRs and corresponding delays */
927 azim = (int)(s->speaker_azim[i] + azim_orig) % 360;
928 elev = (int)(s->speaker_elev[i] + elev_orig) % 90;
929 /* get id of IR closest to desired position */
930 m[i] = find_m(s, azim, elev, radius);
931
932 /* load the delays associated with the current IRs */
933 delay_l[i] = *(s->sofa.data_delay + 2 * m[i]);
934 delay_r[i] = *(s->sofa.data_delay + 2 * m[i] + 1);
935
936 if (s->type == TIME_DOMAIN) {
937 offset = i * FFALIGN(n_samples, 16); /* no. samples already written */
938 for (j = 0; j < n_samples; j++) {
939 /* load reversed IRs of the specified source position
940 * sample-by-sample for left and right ear; and apply gain */
941 *(data_ir_l + offset + j) = /* left channel */
942 *(s->sofa.data_ir + 2 * m[i] * n_samples + n_samples - 1 - j) * gain_lin;
943 *(data_ir_r + offset + j) = /* right channel */
944 *(s->sofa.data_ir + 2 * m[i] * n_samples + n_samples - 1 - j + n_samples) * gain_lin;
945 }
946 } else {
947 fft_in_l = av_calloc(n_fft, sizeof(*fft_in_l));
948 fft_in_r = av_calloc(n_fft, sizeof(*fft_in_r));
949 if (!fft_in_l || !fft_in_r) {
950 av_free(data_hrtf_l);
951 av_free(data_hrtf_r);
952 av_free(fft_in_l);
953 av_free(fft_in_r);
954 return AVERROR(ENOMEM);
955 }
956
957 offset = i * n_fft; /* no. samples already written */
958 for (j = 0; j < n_samples; j++) {
959 /* load non-reversed IRs of the specified source position
960 * sample-by-sample and apply gain,
961 * L channel is loaded to real part, R channel to imag part,
962 * IRs ared shifted by L and R delay */
963 fft_in_l[delay_l[i] + j].re = /* left channel */
964 *(s->sofa.data_ir + 2 * m[i] * n_samples + j) * gain_lin;
965 fft_in_r[delay_r[i] + j].re = /* right channel */
966 *(s->sofa.data_ir + (2 * m[i] + 1) * n_samples + j) * gain_lin;
967 }
968
969 /* actually transform to frequency domain (IRs -> HRTFs) */
970 av_fft_permute(s->fft[0], fft_in_l);
971 av_fft_calc(s->fft[0], fft_in_l);
972 memcpy(data_hrtf_l + offset, fft_in_l, n_fft * sizeof(*fft_in_l));
973 av_fft_permute(s->fft[0], fft_in_r);
974 av_fft_calc(s->fft[0], fft_in_r);
975 memcpy(data_hrtf_r + offset, fft_in_r, n_fft * sizeof(*fft_in_r));
976 }
977
978 av_log(ctx, AV_LOG_DEBUG, "Index: %d, Azimuth: %f, Elevation: %f, Radius: %f of SOFA file.\n",
979 m[i], *(s->sofa.sp_a + m[i]), *(s->sofa.sp_e + m[i]), *(s->sofa.sp_r + m[i]));
980 }
981
982 if (s->type == TIME_DOMAIN) {
983 /* copy IRs and delays to allocated memory in the SOFAlizerContext struct: */
984 memcpy(s->data_ir[0], data_ir_l, sizeof(float) * n_conv * FFALIGN(n_samples, 16));
985 memcpy(s->data_ir[1], data_ir_r, sizeof(float) * n_conv * FFALIGN(n_samples, 16));
986
987 av_freep(&data_ir_l); /* free temporary IR memory */
988 av_freep(&data_ir_r);
989 } else {
990 s->data_hrtf[0] = av_malloc_array(n_fft * s->n_conv, sizeof(FFTComplex));
991 s->data_hrtf[1] = av_malloc_array(n_fft * s->n_conv, sizeof(FFTComplex));
992 if (!s->data_hrtf[0] || !s->data_hrtf[1]) {
993 av_freep(&data_hrtf_l);
994 av_freep(&data_hrtf_r);
995 av_freep(&fft_in_l);
996 av_freep(&fft_in_r);
997 return AVERROR(ENOMEM); /* memory allocation failed */
998 }
999
1000 memcpy(s->data_hrtf[0], data_hrtf_l, /* copy HRTF data to */
1001 sizeof(FFTComplex) * n_conv * n_fft); /* filter struct */
1002 memcpy(s->data_hrtf[1], data_hrtf_r,
1003 sizeof(FFTComplex) * n_conv * n_fft);
1004
1005 av_freep(&data_hrtf_l); /* free temporary HRTF memory */
1006 av_freep(&data_hrtf_r);
1007
1008 av_freep(&fft_in_l); /* free temporary FFT memory */
1009 av_freep(&fft_in_r);
1010 }
1011
1012 memcpy(s->delay[0], &delay_l[0], sizeof(int) * s->n_conv);
1013 memcpy(s->delay[1], &delay_r[0], sizeof(int) * s->n_conv);
1014
1015 return 0;
1016}
1017
1018static av_cold int init(AVFilterContext *ctx)
1019{
1020 SOFAlizerContext *s = ctx->priv;
1021 int ret;
1022
1023 if (!s->filename) {
1024 av_log(ctx, AV_LOG_ERROR, "Valid SOFA filename must be set.\n");
1025 return AVERROR(EINVAL);
1026 }
1027
1028 /* load SOFA file, */
1029 /* initialize file IDs to 0 before attempting to load SOFA files,
1030 * this assures that in case of error, only the memory of already
1031 * loaded files is free'd */
1032 s->sofa.ncid = 0;
1033 ret = load_sofa(ctx, s->filename, &s->sample_rate);
1034 if (ret) {
1035 /* file loading error */
1036 av_log(ctx, AV_LOG_ERROR, "Error while loading SOFA file: '%s'\n", s->filename);
1037 } else { /* no file loading error, resampling not required */
1038 av_log(ctx, AV_LOG_DEBUG, "File '%s' loaded.\n", s->filename);
1039 }
1040
1041 if (ret) {
1042 av_log(ctx, AV_LOG_ERROR, "No valid SOFA file could be loaded. Please specify valid SOFA file.\n");
1043 return ret;
1044 }
1045
1046 s->fdsp = avpriv_float_dsp_alloc(0);
1047 if (!s->fdsp)
1048 return AVERROR(ENOMEM);
1049
1050 return 0;
1051}
1052
1053static int config_input(AVFilterLink *inlink)
1054{
1055 AVFilterContext *ctx = inlink->dst;
1056 SOFAlizerContext *s = ctx->priv;
1057 int nb_input_channels = inlink->channels; /* no. input channels */
1058 int n_max_ir = 0;
1059 int n_current;
1060 int n_max = 0;
1061 int ret;
1062
1063 if (s->type == FREQUENCY_DOMAIN) {
1064 inlink->partial_buf_size =
1065 inlink->min_samples =
1066 inlink->max_samples = inlink->sample_rate;
1067 }
1068
1069 /* gain -3 dB per channel, -6 dB to get LFE on a similar level */
1070 s->gain_lfe = expf((s->gain - 3 * inlink->channels - 6) / 20 * M_LN10);
1071
1072 s->n_conv = nb_input_channels;
1073
1074 /* get size of ringbuffer (longest IR plus max. delay) */
1075 /* then choose next power of 2 for performance optimization */
1076 n_current = s->sofa.n_samples + max_delay(&s->sofa);
1077 if (n_current > n_max) {
1078 /* length of longest IR plus max. delay (in all SOFA files) */
1079 n_max = n_current;
1080 /* length of longest IR (without delay, in all SOFA files) */
1081 n_max_ir = s->sofa.n_samples;
1082 }
1083 /* buffer length is longest IR plus max. delay -> next power of 2
1084 (32 - count leading zeros gives required exponent) */
1085 s->buffer_length = 1 << (32 - ff_clz(n_max));
1086 s->n_fft = 1 << (32 - ff_clz(n_max + inlink->sample_rate));
1087
1088 if (s->type == FREQUENCY_DOMAIN) {
1089 av_fft_end(s->fft[0]);
1090 av_fft_end(s->fft[1]);
1091 s->fft[0] = av_fft_init(log2(s->n_fft), 0);
1092 s->fft[1] = av_fft_init(log2(s->n_fft), 0);
1093 av_fft_end(s->ifft[0]);
1094 av_fft_end(s->ifft[1]);
1095 s->ifft[0] = av_fft_init(log2(s->n_fft), 1);
1096 s->ifft[1] = av_fft_init(log2(s->n_fft), 1);
1097
1098 if (!s->fft[0] || !s->fft[1] || !s->ifft[0] || !s->ifft[1]) {
1099 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT contexts of size %d.\n", s->n_fft);
1100 return AVERROR(ENOMEM);
1101 }
1102 }
1103
1104 /* Allocate memory for the impulse responses, delays and the ringbuffers */
1105 /* size: (longest IR) * (number of channels to convolute) */
1106 s->data_ir[0] = av_calloc(FFALIGN(n_max_ir, 16), sizeof(float) * s->n_conv);
1107 s->data_ir[1] = av_calloc(FFALIGN(n_max_ir, 16), sizeof(float) * s->n_conv);
1108 /* length: number of channels to convolute */
1109 s->delay[0] = av_malloc_array(s->n_conv, sizeof(float));
1110 s->delay[1] = av_malloc_array(s->n_conv, sizeof(float));
1111 /* length: (buffer length) * (number of input channels),
1112 * OR: buffer length (if frequency domain processing)
1113 * calloc zero-initializes the buffer */
1114
1115 if (s->type == TIME_DOMAIN) {
1116 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
1117 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float) * nb_input_channels);
1118 } else {
1119 s->ringbuffer[0] = av_calloc(s->buffer_length, sizeof(float));
1120 s->ringbuffer[1] = av_calloc(s->buffer_length, sizeof(float));
1121 s->temp_fft[0] = av_malloc_array(s->n_fft, sizeof(FFTComplex));
1122 s->temp_fft[1] = av_malloc_array(s->n_fft, sizeof(FFTComplex));
1123 if (!s->temp_fft[0] || !s->temp_fft[1])
1124 return AVERROR(ENOMEM);
1125 }
1126
1127 /* length: number of channels to convolute */
1128 s->speaker_azim = av_calloc(s->n_conv, sizeof(*s->speaker_azim));
1129 s->speaker_elev = av_calloc(s->n_conv, sizeof(*s->speaker_elev));
1130
1131 /* memory allocation failed: */
1132 if (!s->data_ir[0] || !s->data_ir[1] || !s->delay[1] ||
1133 !s->delay[0] || !s->ringbuffer[0] || !s->ringbuffer[1] ||
1134 !s->speaker_azim || !s->speaker_elev)
1135 return AVERROR(ENOMEM);
1136
1137 compensate_volume(ctx);
1138
1139 /* get speaker positions */
1140 if ((ret = get_speaker_pos(ctx, s->speaker_azim, s->speaker_elev)) < 0) {
1141 av_log(ctx, AV_LOG_ERROR, "Couldn't get speaker positions. Input channel configuration not supported.\n");
1142 return ret;
1143 }
1144
1145 /* load IRs to data_ir[0] and data_ir[1] for required directions */
1146 if ((ret = load_data(ctx, s->rotation, s->elevation, s->radius)) < 0)
1147 return ret;
1148
1149 av_log(ctx, AV_LOG_DEBUG, "Samplerate: %d Channels to convolute: %d, Length of ringbuffer: %d x %d\n",
1150 inlink->sample_rate, s->n_conv, nb_input_channels, s->buffer_length);
1151
1152 return 0;
1153}
1154
1155static av_cold void uninit(AVFilterContext *ctx)
1156{
1157 SOFAlizerContext *s = ctx->priv;
1158
1159 if (s->sofa.ncid) {
1160 av_freep(&s->sofa.sp_a);
1161 av_freep(&s->sofa.sp_e);
1162 av_freep(&s->sofa.sp_r);
1163 av_freep(&s->sofa.data_delay);
1164 av_freep(&s->sofa.data_ir);
1165 }
1166 av_fft_end(s->ifft[0]);
1167 av_fft_end(s->ifft[1]);
1168 av_fft_end(s->fft[0]);
1169 av_fft_end(s->fft[1]);
1170 av_freep(&s->delay[0]);
1171 av_freep(&s->delay[1]);
1172 av_freep(&s->data_ir[0]);
1173 av_freep(&s->data_ir[1]);
1174 av_freep(&s->ringbuffer[0]);
1175 av_freep(&s->ringbuffer[1]);
1176 av_freep(&s->speaker_azim);
1177 av_freep(&s->speaker_elev);
1178 av_freep(&s->temp_src[0]);
1179 av_freep(&s->temp_src[1]);
1180 av_freep(&s->temp_fft[0]);
1181 av_freep(&s->temp_fft[1]);
1182 av_freep(&s->data_hrtf[0]);
1183 av_freep(&s->data_hrtf[1]);
1184 av_freep(&s->fdsp);
1185}
1186
1187#define OFFSET(x) offsetof(SOFAlizerContext, x)
1188#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
1189
1190static const AVOption sofalizer_options[] = {
1191 { "sofa", "sofa filename", OFFSET(filename), AV_OPT_TYPE_STRING, {.str=NULL}, .flags = FLAGS },
1192 { "gain", "set gain in dB", OFFSET(gain), AV_OPT_TYPE_FLOAT, {.dbl=0}, -20, 40, .flags = FLAGS },
1193 { "rotation", "set rotation" , OFFSET(rotation), AV_OPT_TYPE_FLOAT, {.dbl=0}, -360, 360, .flags = FLAGS },
1194 { "elevation", "set elevation", OFFSET(elevation), AV_OPT_TYPE_FLOAT, {.dbl=0}, -90, 90, .flags = FLAGS },
1195 { "radius", "set radius", OFFSET(radius), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 3, .flags = FLAGS },
1196 { "type", "set processing", OFFSET(type), AV_OPT_TYPE_INT, {.i64=1}, 0, 1, .flags = FLAGS, "type" },
1197 { "time", "time domain", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, .flags = FLAGS, "type" },
1198 { "freq", "frequency domain", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, .flags = FLAGS, "type" },
1199 { "speakers", "set speaker custom positions", OFFSET(speakers_pos), AV_OPT_TYPE_STRING, {.str=0}, 0, 0, .flags = FLAGS },
1200 { NULL }
1201};
1202
1203AVFILTER_DEFINE_CLASS(sofalizer);
1204
1205static const AVFilterPad inputs[] = {
1206 {
1207 .name = "default",
1208 .type = AVMEDIA_TYPE_AUDIO,
1209 .config_props = config_input,
1210 .filter_frame = filter_frame,
1211 },
1212 { NULL }
1213};
1214
1215static const AVFilterPad outputs[] = {
1216 {
1217 .name = "default",
1218 .type = AVMEDIA_TYPE_AUDIO,
1219 },
1220 { NULL }
1221};
1222
1223AVFilter ff_af_sofalizer = {
1224 .name = "sofalizer",
1225 .description = NULL_IF_CONFIG_SMALL("SOFAlizer (Spatially Oriented Format for Acoustics)."),
1226 .priv_size = sizeof(SOFAlizerContext),
1227 .priv_class = &sofalizer_class,
1228 .init = init,
1229 .uninit = uninit,
1230 .query_formats = query_formats,
1231 .inputs = inputs,
1232 .outputs = outputs,
1233 .flags = AVFILTER_FLAG_SLICE_THREADS,
1234};
1235
generated by cgit at 2024-05-17 01:46:06 (GMT)