FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavfilter/af_atempo.c
Date: 2024-10-04 17:46:48
Exec Total Coverage
Lines: 366 419 87.4%
Functions: 21 23 91.3%
Branches: 129 296 43.6%

Line Branch Exec Source
1 /*
2 * Copyright (c) 2012 Pavel Koshevoy <pkoshevoy at gmail dot com>
3 *
4 * This file is part of FFmpeg.
5 *
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.
10 *
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.
15 *
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
19 */
20
21 /**
22 * @file
23 * tempo scaling audio filter -- an implementation of WSOLA algorithm
24 *
25 * Based on MIT licensed yaeAudioTempoFilter.h and yaeAudioFragment.h
26 * from Apprentice Video player by Pavel Koshevoy.
27 * https://sourceforge.net/projects/apprenticevideo/
28 *
29 * An explanation of SOLA algorithm is available at
30 * http://www.surina.net/article/time-and-pitch-scaling.html
31 *
32 * WSOLA is very similar to SOLA, only one major difference exists between
33 * these algorithms. SOLA shifts audio fragments along the output stream,
34 * where as WSOLA shifts audio fragments along the input stream.
35 *
36 * The advantage of WSOLA algorithm is that the overlap region size is
37 * always the same, therefore the blending function is constant and
38 * can be precomputed.
39 */
40
41 #include <float.h>
42 #include "libavutil/avassert.h"
43 #include "libavutil/channel_layout.h"
44 #include "libavutil/mem.h"
45 #include "libavutil/opt.h"
46 #include "libavutil/samplefmt.h"
47 #include "libavutil/tx.h"
48 #include "avfilter.h"
49 #include "audio.h"
50 #include "filters.h"
51
52 /**
53 * A fragment of audio waveform
54 */
55 typedef struct AudioFragment {
56 // index of the first sample of this fragment in the overall waveform;
57 // 0: input sample position
58 // 1: output sample position
59 int64_t position[2];
60
61 // original packed multi-channel samples:
62 uint8_t *data;
63
64 // number of samples in this fragment:
65 int nsamples;
66
67 // rDFT transform of the down-mixed mono fragment, used for
68 // fast waveform alignment via correlation in frequency domain:
69 float *xdat_in;
70 float *xdat;
71 } AudioFragment;
72
73 /**
74 * Filter state machine states
75 */
76 typedef enum {
77 YAE_LOAD_FRAGMENT,
78 YAE_ADJUST_POSITION,
79 YAE_RELOAD_FRAGMENT,
80 YAE_OUTPUT_OVERLAP_ADD,
81 YAE_FLUSH_OUTPUT,
82 } FilterState;
83
84 /**
85 * Filter state machine
86 */
87 typedef struct ATempoContext {
88 const AVClass *class;
89
90 // ring-buffer of input samples, necessary because some times
91 // input fragment position may be adjusted backwards:
92 uint8_t *buffer;
93
94 // ring-buffer maximum capacity, expressed in sample rate time base:
95 int ring;
96
97 // ring-buffer house keeping:
98 int size;
99 int head;
100 int tail;
101
102 // 0: input sample position corresponding to the ring buffer tail
103 // 1: output sample position
104 int64_t position[2];
105
106 // first input timestamp, all other timestamps are offset by this one
107 int64_t start_pts;
108
109 // sample format:
110 enum AVSampleFormat format;
111
112 // number of channels:
113 int channels;
114
115 // row of bytes to skip from one sample to next, across multple channels;
116 // stride = (number-of-channels * bits-per-sample-per-channel) / 8
117 int stride;
118
119 // fragment window size, power-of-two integer:
120 int window;
121
122 // Hann window coefficients, for feathering
123 // (blending) the overlapping fragment region:
124 float *hann;
125
126 // tempo scaling factor:
127 double tempo;
128
129 // a snapshot of previous fragment input and output position values
130 // captured when the tempo scale factor was set most recently:
131 int64_t origin[2];
132
133 // current/previous fragment ring-buffer:
134 AudioFragment frag[2];
135
136 // current fragment index:
137 uint64_t nfrag;
138
139 // current state:
140 FilterState state;
141
142 // for fast correlation calculation in frequency domain:
143 AVTXContext *real_to_complex;
144 AVTXContext *complex_to_real;
145 av_tx_fn r2c_fn, c2r_fn;
146 float *correlation_in;
147 float *correlation;
148
149 // for managing AVFilterPad.request_frame and AVFilterPad.filter_frame
150 AVFrame *dst_buffer;
151 uint8_t *dst;
152 uint8_t *dst_end;
153 uint64_t nsamples_in;
154 uint64_t nsamples_out;
155 } ATempoContext;
156
157 #define YAE_ATEMPO_MIN 0.5
158 #define YAE_ATEMPO_MAX 100.0
159
160 #define OFFSET(x) offsetof(ATempoContext, x)
161
162 static const AVOption atempo_options[] = {
163 { "tempo", "set tempo scale factor",
164 OFFSET(tempo), AV_OPT_TYPE_DOUBLE, { .dbl = 1.0 },
165 YAE_ATEMPO_MIN,
166 YAE_ATEMPO_MAX,
167 AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_RUNTIME_PARAM },
168 { NULL }
169 };
170
171 AVFILTER_DEFINE_CLASS(atempo);
172
173 1546 inline static AudioFragment *yae_curr_frag(ATempoContext *atempo)
174 {
175 1546 return &atempo->frag[atempo->nfrag % 2];
176 }
177
178 450 inline static AudioFragment *yae_prev_frag(ATempoContext *atempo)
179 {
180 450 return &atempo->frag[(atempo->nfrag + 1) % 2];
181 }
182
183 /**
184 * Reset filter to initial state, do not deallocate existing local buffers.
185 */
186 3 static void yae_clear(ATempoContext *atempo)
187 {
188 3 atempo->size = 0;
189 3 atempo->head = 0;
190 3 atempo->tail = 0;
191
192 3 atempo->nfrag = 0;
193 3 atempo->state = YAE_LOAD_FRAGMENT;
194 3 atempo->start_pts = AV_NOPTS_VALUE;
195
196 3 atempo->position[0] = 0;
197 3 atempo->position[1] = 0;
198
199 3 atempo->origin[0] = 0;
200 3 atempo->origin[1] = 0;
201
202 3 atempo->frag[0].position[0] = 0;
203 3 atempo->frag[0].position[1] = 0;
204 3 atempo->frag[0].nsamples = 0;
205
206 3 atempo->frag[1].position[0] = 0;
207 3 atempo->frag[1].position[1] = 0;
208 3 atempo->frag[1].nsamples = 0;
209
210 // shift left position of 1st fragment by half a window
211 // so that no re-normalization would be required for
212 // the left half of the 1st fragment:
213 3 atempo->frag[0].position[0] = -(int64_t)(atempo->window / 2);
214 3 atempo->frag[0].position[1] = -(int64_t)(atempo->window / 2);
215
216 3 av_frame_free(&atempo->dst_buffer);
217 3 atempo->dst = NULL;
218 3 atempo->dst_end = NULL;
219
220 3 atempo->nsamples_in = 0;
221 3 atempo->nsamples_out = 0;
222 3 }
223
224 /**
225 * Reset filter to initial state and deallocate all buffers.
226 */
227 3 static void yae_release_buffers(ATempoContext *atempo)
228 {
229 3 yae_clear(atempo);
230
231 3 av_freep(&atempo->frag[0].data);
232 3 av_freep(&atempo->frag[1].data);
233 3 av_freep(&atempo->frag[0].xdat_in);
234 3 av_freep(&atempo->frag[1].xdat_in);
235 3 av_freep(&atempo->frag[0].xdat);
236 3 av_freep(&atempo->frag[1].xdat);
237
238 3 av_freep(&atempo->buffer);
239 3 av_freep(&atempo->hann);
240 3 av_freep(&atempo->correlation_in);
241 3 av_freep(&atempo->correlation);
242
243 3 av_tx_uninit(&atempo->real_to_complex);
244 3 av_tx_uninit(&atempo->complex_to_real);
245 3 }
246
247 /**
248 * Prepare filter for processing audio data of given format,
249 * sample rate and number of channels.
250 */
251 1 static int yae_reset(ATempoContext *atempo,
252 enum AVSampleFormat format,
253 int sample_rate,
254 int channels)
255 {
256 1 const int sample_size = av_get_bytes_per_sample(format);
257 1 uint32_t nlevels = 0;
258 1 float scale = 1.f, iscale = 1.f;
259 uint32_t pot;
260 int ret;
261 int i;
262
263 1 atempo->format = format;
264 1 atempo->channels = channels;
265 1 atempo->stride = sample_size * channels;
266
267 // pick a segment window size:
268 1 atempo->window = sample_rate / 24;
269
270 // adjust window size to be a power-of-two integer:
271 1 nlevels = av_log2(atempo->window);
272 1 pot = 1 << nlevels;
273
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 av_assert0(pot <= atempo->window);
274
275
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
1 if (pot < atempo->window) {
276 1 atempo->window = pot * 2;
277 1 nlevels++;
278 }
279
280 /* av_realloc is not aligned enough, so simply discard all the old buffers
281 * (fortunately, their data does not need to be preserved) */
282 1 yae_release_buffers(atempo);
283
284 // initialize audio fragment buffers:
285
1/2
✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
1 if (!(atempo->frag[0].data = av_calloc(atempo->window, atempo->stride)) ||
286
1/2
✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
1 !(atempo->frag[1].data = av_calloc(atempo->window, atempo->stride)) ||
287
1/2
✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
1 !(atempo->frag[0].xdat_in = av_calloc(atempo->window + 1, sizeof(AVComplexFloat))) ||
288
1/2
✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
1 !(atempo->frag[1].xdat_in = av_calloc(atempo->window + 1, sizeof(AVComplexFloat))) ||
289
1/2
✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
1 !(atempo->frag[0].xdat = av_calloc(atempo->window + 1, sizeof(AVComplexFloat))) ||
290
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
1 !(atempo->frag[1].xdat = av_calloc(atempo->window + 1, sizeof(AVComplexFloat)))) {
291 ret = AVERROR(ENOMEM);
292 goto fail;
293 }
294
295 // initialize rDFT contexts:
296 1 ret = av_tx_init(&atempo->real_to_complex, &atempo->r2c_fn,
297 1 AV_TX_FLOAT_RDFT, 0, 1 << (nlevels + 1), &scale, 0);
298
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (ret < 0)
299 goto fail;
300
301 1 ret = av_tx_init(&atempo->complex_to_real, &atempo->c2r_fn,
302 1 AV_TX_FLOAT_RDFT, 1, 1 << (nlevels + 1), &iscale, 0);
303
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (ret < 0)
304 goto fail;
305
306
1/2
✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
1 if (!(atempo->correlation_in = av_calloc(atempo->window + 1, sizeof(AVComplexFloat))) ||
307
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
1 !(atempo->correlation = av_calloc(atempo->window, sizeof(AVComplexFloat)))) {
308 ret = AVERROR(ENOMEM);
309 goto fail;
310 }
311
312 1 atempo->ring = atempo->window * 3;
313 1 atempo->buffer = av_calloc(atempo->ring, atempo->stride);
314
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (!atempo->buffer) {
315 ret = AVERROR(ENOMEM);
316 goto fail;
317 }
318
319 // initialize the Hann window function:
320 1 atempo->hann = av_malloc_array(atempo->window, sizeof(float));
321
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (!atempo->hann) {
322 ret = AVERROR(ENOMEM);
323 goto fail;
324 }
325
326
2/2
✓ Branch 0 taken 2048 times.
✓ Branch 1 taken 1 times.
2049 for (i = 0; i < atempo->window; i++) {
327 2048 double t = (double)i / (double)(atempo->window - 1);
328 2048 double h = 0.5 * (1.0 - cos(2.0 * M_PI * t));
329 2048 atempo->hann[i] = (float)h;
330 }
331
332 1 return 0;
333 fail:
334 yae_release_buffers(atempo);
335 return ret;
336 }
337
338 static int yae_update(AVFilterContext *ctx)
339 {
340 const AudioFragment *prev;
341 ATempoContext *atempo = ctx->priv;
342
343 prev = yae_prev_frag(atempo);
344 atempo->origin[0] = prev->position[0] + atempo->window / 2;
345 atempo->origin[1] = prev->position[1] + atempo->window / 2;
346 return 0;
347 }
348
349 /**
350 * A helper macro for initializing complex data buffer with scalar data
351 * of a given type.
352 */
353 #define yae_init_xdat(scalar_type, scalar_max) \
354 do { \
355 const uint8_t *src_end = src + \
356 frag->nsamples * atempo->channels * sizeof(scalar_type); \
357 \
358 float *xdat = frag->xdat_in; \
359 scalar_type tmp; \
360 \
361 if (atempo->channels == 1) { \
362 for (; src < src_end; xdat++) { \
363 tmp = *(const scalar_type *)src; \
364 src += sizeof(scalar_type); \
365 \
366 *xdat = (float)tmp; \
367 } \
368 } else { \
369 float s, max, ti, si; \
370 int i; \
371 \
372 for (; src < src_end; xdat++) { \
373 tmp = *(const scalar_type *)src; \
374 src += sizeof(scalar_type); \
375 \
376 max = (float)tmp; \
377 s = FFMIN((float)scalar_max, \
378 (float)fabsf(max)); \
379 \
380 for (i = 1; i < atempo->channels; i++) { \
381 tmp = *(const scalar_type *)src; \
382 src += sizeof(scalar_type); \
383 \
384 ti = (float)tmp; \
385 si = FFMIN((float)scalar_max, \
386 (float)fabsf(ti)); \
387 \
388 if (s < si) { \
389 s = si; \
390 max = ti; \
391 } \
392 } \
393 \
394 *xdat = max; \
395 } \
396 } \
397 } while (0)
398
399 /**
400 * Initialize complex data buffer of a given audio fragment
401 * with down-mixed mono data of appropriate scalar type.
402 */
403 258 static void yae_downmix(ATempoContext *atempo, AudioFragment *frag)
404 {
405 // shortcuts:
406 258 const uint8_t *src = frag->data;
407
408 // init complex data buffer used for FFT and Correlation:
409 258 memset(frag->xdat_in, 0, sizeof(AVComplexFloat) * (atempo->window + 1));
410
411
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 258 times.
258 if (atempo->format == AV_SAMPLE_FMT_U8) {
412 yae_init_xdat(uint8_t, 127);
413
1/2
✓ Branch 0 taken 258 times.
✗ Branch 1 not taken.
258 } else if (atempo->format == AV_SAMPLE_FMT_S16) {
414
3/14
✓ Branch 0 taken 258 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 527650 times.
✓ Branch 3 taken 258 times.
✗ Branch 4 not taken.
✗ Branch 5 not taken.
✗ Branch 6 not taken.
✗ Branch 7 not taken.
✗ Branch 8 not taken.
✗ Branch 9 not taken.
✗ Branch 10 not taken.
✗ Branch 11 not taken.
✗ Branch 12 not taken.
✗ Branch 13 not taken.
527908 yae_init_xdat(int16_t, 32767);
415 } else if (atempo->format == AV_SAMPLE_FMT_S32) {
416 yae_init_xdat(int, 2147483647);
417 } else if (atempo->format == AV_SAMPLE_FMT_FLT) {
418 yae_init_xdat(float, 1);
419 } else if (atempo->format == AV_SAMPLE_FMT_DBL) {
420 yae_init_xdat(double, 1);
421 }
422 258 }
423
424 /**
425 * Populate the internal data buffer on as-needed basis.
426 *
427 * @return
428 * 0 if requested data was already available or was successfully loaded,
429 * AVERROR(EAGAIN) if more input data is required.
430 */
431 322 static int yae_load_data(ATempoContext *atempo,
432 const uint8_t **src_ref,
433 const uint8_t *src_end,
434 int64_t stop_here)
435 {
436 // shortcut:
437 322 const uint8_t *src = *src_ref;
438 322 const int read_size = stop_here - atempo->position[0];
439
440
2/2
✓ Branch 0 taken 53 times.
✓ Branch 1 taken 269 times.
322 if (stop_here <= atempo->position[0]) {
441 53 return 0;
442 }
443
444 // samples are not expected to be skipped, unless tempo is greater than 2:
445
1/4
✗ Branch 0 not taken.
✓ Branch 1 taken 269 times.
✗ Branch 2 not taken.
✗ Branch 3 not taken.
269 av_assert0(read_size <= atempo->ring || atempo->tempo > 2.0);
446
447
4/4
✓ Branch 0 taken 334 times.
✓ Branch 1 taken 204 times.
✓ Branch 2 taken 269 times.
✓ Branch 3 taken 65 times.
807 while (atempo->position[0] < stop_here && src < src_end) {
448 269 int src_samples = (src_end - src) / atempo->stride;
449
450 // load data piece-wise, in order to avoid complicating the logic:
451 269 int nsamples = FFMIN(read_size, src_samples);
452 int na;
453 int nb;
454
455 269 nsamples = FFMIN(nsamples, atempo->ring);
456 269 na = FFMIN(nsamples, atempo->ring - atempo->tail);
457 269 nb = FFMIN(nsamples - na, atempo->ring);
458
459
1/2
✓ Branch 0 taken 269 times.
✗ Branch 1 not taken.
269 if (na) {
460 269 uint8_t *a = atempo->buffer + atempo->tail * atempo->stride;
461 269 memcpy(a, src, na * atempo->stride);
462
463 269 src += na * atempo->stride;
464 269 atempo->position[0] += na;
465
466 269 atempo->size = FFMIN(atempo->size + na, atempo->ring);
467 269 atempo->tail = (atempo->tail + na) % atempo->ring;
468 269 atempo->head =
469 269 atempo->size < atempo->ring ?
470
2/2
✓ Branch 0 taken 5 times.
✓ Branch 1 taken 264 times.
269 atempo->tail - atempo->size :
471 atempo->tail;
472 }
473
474
2/2
✓ Branch 0 taken 247 times.
✓ Branch 1 taken 22 times.
269 if (nb) {
475 22 uint8_t *b = atempo->buffer;
476 22 memcpy(b, src, nb * atempo->stride);
477
478 22 src += nb * atempo->stride;
479 22 atempo->position[0] += nb;
480
481 22 atempo->size = FFMIN(atempo->size + nb, atempo->ring);
482 22 atempo->tail = (atempo->tail + nb) % atempo->ring;
483 22 atempo->head =
484 22 atempo->size < atempo->ring ?
485
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 22 times.
22 atempo->tail - atempo->size :
486 atempo->tail;
487 }
488 }
489
490 // pass back the updated source buffer pointer:
491 269 *src_ref = src;
492
493 // sanity check:
494
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 269 times.
269 av_assert0(atempo->position[0] <= stop_here);
495
496
2/2
✓ Branch 0 taken 204 times.
✓ Branch 1 taken 65 times.
269 return atempo->position[0] == stop_here ? 0 : AVERROR(EAGAIN);
497 }
498
499 /**
500 * Populate current audio fragment data buffer.
501 *
502 * @return
503 * 0 when the fragment is ready,
504 * AVERROR(EAGAIN) if more input data is required.
505 */
506 324 static int yae_load_frag(ATempoContext *atempo,
507 const uint8_t **src_ref,
508 const uint8_t *src_end)
509 {
510 // shortcuts:
511 324 AudioFragment *frag = yae_curr_frag(atempo);
512 uint8_t *dst;
513 int64_t missing, start, zeros;
514 uint32_t nsamples;
515 const uint8_t *a, *b;
516 int i0, i1, n0, n1, na, nb;
517
518 324 int64_t stop_here = frag->position[0] + atempo->window;
519
4/4
✓ Branch 0 taken 322 times.
✓ Branch 1 taken 2 times.
✓ Branch 3 taken 65 times.
✓ Branch 4 taken 257 times.
324 if (src_ref && yae_load_data(atempo, src_ref, src_end, stop_here) != 0) {
520 65 return AVERROR(EAGAIN);
521 }
522
523 // calculate the number of samples we don't have:
524 259 missing =
525 259 stop_here > atempo->position[0] ?
526
2/2
✓ Branch 0 taken 2 times.
✓ Branch 1 taken 257 times.
259 stop_here - atempo->position[0] : 0;
527
528 259 nsamples =
529 259 missing < (int64_t)atempo->window ?
530
1/2
✓ Branch 0 taken 259 times.
✗ Branch 1 not taken.
259 (uint32_t)(atempo->window - missing) : 0;
531
532 // setup the output buffer:
533 259 frag->nsamples = nsamples;
534 259 dst = frag->data;
535
536 259 start = atempo->position[0] - atempo->size;
537
538 // what we don't have we substitute with zeros:
539 259 zeros =
540 259 frag->position[0] < start ?
541
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 258 times.
259 FFMIN(start - frag->position[0], (int64_t)nsamples) : 0;
542
543
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 259 times.
259 if (zeros == nsamples) {
544 return 0;
545 }
546
547
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 258 times.
259 if (frag->position[0] < start) {
548 1 memset(dst, 0, zeros * atempo->stride);
549 1 dst += zeros * atempo->stride;
550 }
551
552 // get the remaining data from the ring buffer:
553 518 na = (atempo->head < atempo->tail ?
554
2/2
✓ Branch 0 taken 5 times.
✓ Branch 1 taken 254 times.
259 atempo->tail - atempo->head :
555 254 atempo->ring - atempo->head);
556
557
2/2
✓ Branch 0 taken 254 times.
✓ Branch 1 taken 5 times.
259 nb = atempo->head < atempo->tail ? 0 : atempo->tail;
558
559 // sanity check:
560
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 259 times.
259 av_assert0(nsamples <= zeros + na + nb);
561
562 259 a = atempo->buffer + atempo->head * atempo->stride;
563 259 b = atempo->buffer;
564
565 259 i0 = frag->position[0] + zeros - start;
566
2/2
✓ Branch 0 taken 168 times.
✓ Branch 1 taken 91 times.
259 i1 = i0 < na ? 0 : i0 - na;
567
568
2/2
✓ Branch 0 taken 91 times.
✓ Branch 1 taken 168 times.
259 n0 = i0 < na ? FFMIN(na - i0, (int)(nsamples - zeros)) : 0;
569 259 n1 = nsamples - zeros - n0;
570
571
2/2
✓ Branch 0 taken 91 times.
✓ Branch 1 taken 168 times.
259 if (n0) {
572 91 memcpy(dst, a + i0 * atempo->stride, n0 * atempo->stride);
573 91 dst += n0 * atempo->stride;
574 }
575
576
2/2
✓ Branch 0 taken 254 times.
✓ Branch 1 taken 5 times.
259 if (n1) {
577 254 memcpy(dst, b + i1 * atempo->stride, n1 * atempo->stride);
578 }
579
580 259 return 0;
581 }
582
583 /**
584 * Prepare for loading next audio fragment.
585 */
586 129 static void yae_advance_to_next_frag(ATempoContext *atempo)
587 {
588 129 const double fragment_step = atempo->tempo * (double)(atempo->window / 2);
589
590 const AudioFragment *prev;
591 AudioFragment *frag;
592
593 129 atempo->nfrag++;
594 129 prev = yae_prev_frag(atempo);
595 129 frag = yae_curr_frag(atempo);
596
597 129 frag->position[0] = prev->position[0] + (int64_t)fragment_step;
598 129 frag->position[1] = prev->position[1] + atempo->window / 2;
599 129 frag->nsamples = 0;
600 129 }
601
602 /**
603 * Calculate cross-correlation via rDFT.
604 *
605 * Multiply two vectors of complex numbers (result of real_to_complex rDFT)
606 * and transform back via complex_to_real rDFT.
607 */
608 129 static void yae_xcorr_via_rdft(float *xcorr_in,
609 float *xcorr,
610 AVTXContext *complex_to_real,
611 av_tx_fn c2r_fn,
612 const AVComplexFloat *xa,
613 const AVComplexFloat *xb,
614 const int window)
615 {
616 129 AVComplexFloat *xc = (AVComplexFloat *)xcorr_in;
617 int i;
618
619
2/2
✓ Branch 0 taken 264321 times.
✓ Branch 1 taken 129 times.
264450 for (i = 0; i <= window; i++, xa++, xb++, xc++) {
620 264321 xc->re = (xa->re * xb->re + xa->im * xb->im);
621 264321 xc->im = (xa->im * xb->re - xa->re * xb->im);
622 }
623
624 // apply inverse rDFT:
625 129 c2r_fn(complex_to_real, xcorr, xcorr_in, sizeof(*xc));
626 129 }
627
628 /**
629 * Calculate alignment offset for given fragment
630 * relative to the previous fragment.
631 *
632 * @return alignment offset of current fragment relative to previous.
633 */
634 129 static int yae_align(AudioFragment *frag,
635 const AudioFragment *prev,
636 const int window,
637 const int delta_max,
638 const int drift,
639 float *correlation_in,
640 float *correlation,
641 AVTXContext *complex_to_real,
642 av_tx_fn c2r_fn)
643 {
644 129 int best_offset = -drift;
645 129 float best_metric = -FLT_MAX;
646 float *xcorr;
647
648 int i0;
649 int i1;
650 int i;
651
652 129 yae_xcorr_via_rdft(correlation_in,
653 correlation,
654 complex_to_real,
655 c2r_fn,
656 129 (const AVComplexFloat *)prev->xdat,
657 129 (const AVComplexFloat *)frag->xdat,
658 window);
659
660 // identify search window boundaries:
661 129 i0 = FFMAX(window / 2 - delta_max - drift, 0);
662 129 i0 = FFMIN(i0, window);
663
664 129 i1 = FFMIN(window / 2 + delta_max - drift, window - window / 16);
665 129 i1 = FFMAX(i1, 0);
666
667 // identify cross-correlation peaks within search window:
668 129 xcorr = correlation + i0;
669
670
2/2
✓ Branch 0 taken 219329 times.
✓ Branch 1 taken 129 times.
219458 for (i = i0; i < i1; i++, xcorr++) {
671 219329 float metric = *xcorr;
672
673 // normalize:
674 219329 float drifti = (float)(drift + i);
675 219329 metric *= drifti * (float)(i - i0) * (float)(i1 - i);
676
677
2/2
✓ Branch 0 taken 11612 times.
✓ Branch 1 taken 207717 times.
219329 if (metric > best_metric) {
678 11612 best_metric = metric;
679 11612 best_offset = i - window / 2;
680 }
681 }
682
683 129 return best_offset;
684 }
685
686 /**
687 * Adjust current fragment position for better alignment
688 * with previous fragment.
689 *
690 * @return alignment correction.
691 */
692 129 static int yae_adjust_position(ATempoContext *atempo)
693 {
694 129 const AudioFragment *prev = yae_prev_frag(atempo);
695 129 AudioFragment *frag = yae_curr_frag(atempo);
696
697 129 const double prev_output_position =
698 129 (double)(prev->position[1] - atempo->origin[1] + atempo->window / 2) *
699 129 atempo->tempo;
700
701 129 const double ideal_output_position =
702 129 (double)(prev->position[0] - atempo->origin[0] + atempo->window / 2);
703
704 129 const int drift = (int)(prev_output_position - ideal_output_position);
705
706 129 const int delta_max = atempo->window / 2;
707 129 const int correction = yae_align(frag,
708 prev,
709 atempo->window,
710 delta_max,
711 drift,
712 atempo->correlation_in,
713 atempo->correlation,
714 atempo->complex_to_real,
715 atempo->c2r_fn);
716
717
1/2
✓ Branch 0 taken 129 times.
✗ Branch 1 not taken.
129 if (correction) {
718 // adjust fragment position:
719 129 frag->position[0] -= correction;
720
721 // clear so that the fragment can be reloaded:
722 129 frag->nsamples = 0;
723 }
724
725 129 return correction;
726 }
727
728 /**
729 * A helper macro for blending the overlap region of previous
730 * and current audio fragment.
731 */
732 #define yae_blend(scalar_type) \
733 do { \
734 const scalar_type *aaa = (const scalar_type *)a; \
735 const scalar_type *bbb = (const scalar_type *)b; \
736 \
737 scalar_type *out = (scalar_type *)dst; \
738 scalar_type *out_end = (scalar_type *)dst_end; \
739 int64_t i; \
740 \
741 for (i = 0; i < overlap && out < out_end; \
742 i++, atempo->position[1]++, wa++, wb++) { \
743 float w0 = *wa; \
744 float w1 = *wb; \
745 int j; \
746 \
747 for (j = 0; j < atempo->channels; \
748 j++, aaa++, bbb++, out++) { \
749 float t0 = (float)*aaa; \
750 float t1 = (float)*bbb; \
751 \
752 *out = \
753 frag->position[0] + i < 0 ? \
754 *aaa : \
755 (scalar_type)(t0 * w0 + t1 * w1); \
756 } \
757 } \
758 dst = (uint8_t *)out; \
759 } while (0)
760
761 /**
762 * Blend the overlap region of previous and current audio fragment
763 * and output the results to the given destination buffer.
764 *
765 * @return
766 * 0 if the overlap region was completely stored in the dst buffer,
767 * AVERROR(EAGAIN) if more destination buffer space is required.
768 */
769 192 static int yae_overlap_add(ATempoContext *atempo,
770 uint8_t **dst_ref,
771 uint8_t *dst_end)
772 {
773 // shortcuts:
774 192 const AudioFragment *prev = yae_prev_frag(atempo);
775 192 const AudioFragment *frag = yae_curr_frag(atempo);
776
777 192 const int64_t start_here = FFMAX(atempo->position[1],
778 frag->position[1]);
779
780 192 const int64_t stop_here = FFMIN(prev->position[1] + prev->nsamples,
781 frag->position[1] + frag->nsamples);
782
783 192 const int64_t overlap = stop_here - start_here;
784
785 192 const int64_t ia = start_here - prev->position[1];
786 192 const int64_t ib = start_here - frag->position[1];
787
788 192 const float *wa = atempo->hann + ia;
789 192 const float *wb = atempo->hann + ib;
790
791 192 const uint8_t *a = prev->data + ia * atempo->stride;
792 192 const uint8_t *b = frag->data + ib * atempo->stride;
793
794 192 uint8_t *dst = *dst_ref;
795
796
3/6
✓ Branch 0 taken 192 times.
✗ Branch 1 not taken.
✓ Branch 2 taken 192 times.
✗ Branch 3 not taken.
✗ Branch 4 not taken.
✓ Branch 5 taken 192 times.
192 av_assert0(start_here <= stop_here &&
797 frag->position[1] <= start_here &&
798 overlap <= frag->nsamples);
799
800
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 192 times.
192 if (atempo->format == AV_SAMPLE_FMT_U8) {
801 yae_blend(uint8_t);
802
1/2
✓ Branch 0 taken 192 times.
✗ Branch 1 not taken.
192 } else if (atempo->format == AV_SAMPLE_FMT_S16) {
803
7/8
✗ Branch 0 not taken.
✓ Branch 1 taken 132096 times.
✓ Branch 2 taken 132096 times.
✓ Branch 3 taken 132096 times.
✓ Branch 4 taken 132159 times.
✓ Branch 5 taken 129 times.
✓ Branch 6 taken 132096 times.
✓ Branch 7 taken 63 times.
264384 yae_blend(int16_t);
804 } else if (atempo->format == AV_SAMPLE_FMT_S32) {
805 yae_blend(int);
806 } else if (atempo->format == AV_SAMPLE_FMT_FLT) {
807 yae_blend(float);
808 } else if (atempo->format == AV_SAMPLE_FMT_DBL) {
809 yae_blend(double);
810 }
811
812 // pass-back the updated destination buffer pointer:
813 192 *dst_ref = dst;
814
815
2/2
✓ Branch 0 taken 129 times.
✓ Branch 1 taken 63 times.
192 return atempo->position[1] == stop_here ? 0 : AVERROR(EAGAIN);
816 }
817
818 /**
819 * Feed as much data to the filter as it is able to consume
820 * and receive as much processed data in the destination buffer
821 * as it is able to produce or store.
822 */
823 static void
824 128 yae_apply(ATempoContext *atempo,
825 const uint8_t **src_ref,
826 const uint8_t *src_end,
827 uint8_t **dst_ref,
828 uint8_t *dst_end)
829 {
830 while (1) {
831
2/2
✓ Branch 0 taken 194 times.
✓ Branch 1 taken 63 times.
257 if (atempo->state == YAE_LOAD_FRAGMENT) {
832 // load additional data for the current fragment:
833
2/2
✓ Branch 1 taken 65 times.
✓ Branch 2 taken 129 times.
194 if (yae_load_frag(atempo, src_ref, src_end) != 0) {
834 65 break;
835 }
836
837 // down-mix to mono:
838 129 yae_downmix(atempo, yae_curr_frag(atempo));
839
840 // apply rDFT:
841 129 atempo->r2c_fn(atempo->real_to_complex, yae_curr_frag(atempo)->xdat, yae_curr_frag(atempo)->xdat_in, sizeof(float));
842
843 // must load the second fragment before alignment can start:
844
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 128 times.
129 if (!atempo->nfrag) {
845 1 yae_advance_to_next_frag(atempo);
846 1 continue;
847 }
848
849 128 atempo->state = YAE_ADJUST_POSITION;
850 }
851
852
2/2
✓ Branch 0 taken 128 times.
✓ Branch 1 taken 63 times.
191 if (atempo->state == YAE_ADJUST_POSITION) {
853 // adjust position for better alignment:
854
1/2
✓ Branch 1 taken 128 times.
✗ Branch 2 not taken.
128 if (yae_adjust_position(atempo)) {
855 // reload the fragment at the corrected position, so that the
856 // Hann window blending would not require normalization:
857 128 atempo->state = YAE_RELOAD_FRAGMENT;
858 } else {
859 atempo->state = YAE_OUTPUT_OVERLAP_ADD;
860 }
861 }
862
863
2/2
✓ Branch 0 taken 128 times.
✓ Branch 1 taken 63 times.
191 if (atempo->state == YAE_RELOAD_FRAGMENT) {
864 // load additional data if necessary due to position adjustment:
865
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 128 times.
128 if (yae_load_frag(atempo, src_ref, src_end) != 0) {
866 break;
867 }
868
869 // down-mix to mono:
870 128 yae_downmix(atempo, yae_curr_frag(atempo));
871
872 // apply rDFT:
873 128 atempo->r2c_fn(atempo->real_to_complex, yae_curr_frag(atempo)->xdat, yae_curr_frag(atempo)->xdat_in, sizeof(float));
874
875 128 atempo->state = YAE_OUTPUT_OVERLAP_ADD;
876 }
877
878
1/2
✓ Branch 0 taken 191 times.
✗ Branch 1 not taken.
191 if (atempo->state == YAE_OUTPUT_OVERLAP_ADD) {
879 // overlap-add and output the result:
880
2/2
✓ Branch 1 taken 63 times.
✓ Branch 2 taken 128 times.
191 if (yae_overlap_add(atempo, dst_ref, dst_end) != 0) {
881 63 break;
882 }
883
884 // advance to the next fragment, repeat:
885 128 yae_advance_to_next_frag(atempo);
886 128 atempo->state = YAE_LOAD_FRAGMENT;
887 }
888 }
889 128 }
890
891 /**
892 * Flush any buffered data from the filter.
893 *
894 * @return
895 * 0 if all data was completely stored in the dst buffer,
896 * AVERROR(EAGAIN) if more destination buffer space is required.
897 */
898 1 static int yae_flush(ATempoContext *atempo,
899 uint8_t **dst_ref,
900 uint8_t *dst_end)
901 {
902 1 AudioFragment *frag = yae_curr_frag(atempo);
903 int64_t overlap_end;
904 int64_t start_here;
905 int64_t stop_here;
906 int64_t offset;
907
908 const uint8_t *src;
909 uint8_t *dst;
910
911 int src_size;
912 int dst_size;
913 int nbytes;
914
915 1 atempo->state = YAE_FLUSH_OUTPUT;
916
917
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (!atempo->nfrag) {
918 // there is nothing to flush:
919 return 0;
920 }
921
922
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (atempo->position[0] == frag->position[0] + frag->nsamples &&
923 atempo->position[1] == frag->position[1] + frag->nsamples) {
924 // the current fragment is already flushed:
925 return 0;
926 }
927
928
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
1 if (frag->position[0] + frag->nsamples < atempo->position[0]) {
929 // finish loading the current (possibly partial) fragment:
930 1 yae_load_frag(atempo, NULL, NULL);
931
932
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
1 if (atempo->nfrag) {
933 // down-mix to mono:
934 1 yae_downmix(atempo, frag);
935
936 // apply rDFT:
937 1 atempo->r2c_fn(atempo->real_to_complex, frag->xdat, frag->xdat_in, sizeof(float));
938
939 // align current fragment to previous fragment:
940
1/2
✓ Branch 1 taken 1 times.
✗ Branch 2 not taken.
1 if (yae_adjust_position(atempo)) {
941 // reload the current fragment due to adjusted position:
942 1 yae_load_frag(atempo, NULL, NULL);
943 }
944 }
945 }
946
947 // flush the overlap region:
948 1 overlap_end = frag->position[1] + FFMIN(atempo->window / 2,
949 frag->nsamples);
950
951
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
2 while (atempo->position[1] < overlap_end) {
952
1/2
✗ Branch 1 not taken.
✓ Branch 2 taken 1 times.
1 if (yae_overlap_add(atempo, dst_ref, dst_end) != 0) {
953 return AVERROR(EAGAIN);
954 }
955 }
956
957 // check whether all of the input samples have been consumed:
958
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (frag->position[0] + frag->nsamples < atempo->position[0]) {
959 yae_advance_to_next_frag(atempo);
960 return AVERROR(EAGAIN);
961 }
962
963 // flush the remainder of the current fragment:
964 1 start_here = FFMAX(atempo->position[1], overlap_end);
965 1 stop_here = frag->position[1] + frag->nsamples;
966 1 offset = start_here - frag->position[1];
967
2/4
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
✗ Branch 2 not taken.
✓ Branch 3 taken 1 times.
1 av_assert0(start_here <= stop_here && frag->position[1] <= start_here);
968
969 1 src = frag->data + offset * atempo->stride;
970 1 dst = (uint8_t *)*dst_ref;
971
972 1 src_size = (int)(stop_here - start_here) * atempo->stride;
973 1 dst_size = dst_end - dst;
974 1 nbytes = FFMIN(src_size, dst_size);
975
976 1 memcpy(dst, src, nbytes);
977 1 dst += nbytes;
978
979 1 atempo->position[1] += (nbytes / atempo->stride);
980
981 // pass-back the updated destination buffer pointer:
982 1 *dst_ref = (uint8_t *)dst;
983
984
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
1 return atempo->position[1] == stop_here ? 0 : AVERROR(EAGAIN);
985 }
986
987 2 static av_cold int init(AVFilterContext *ctx)
988 {
989 2 ATempoContext *atempo = ctx->priv;
990 2 atempo->format = AV_SAMPLE_FMT_NONE;
991 2 atempo->state = YAE_LOAD_FRAGMENT;
992 2 return 0;
993 }
994
995 2 static av_cold void uninit(AVFilterContext *ctx)
996 {
997 2 ATempoContext *atempo = ctx->priv;
998 2 yae_release_buffers(atempo);
999 2 }
1000
1001 // WSOLA necessitates an internal sliding window ring buffer
1002 // for incoming audio stream.
1003 //
1004 // Planar sample formats are too cumbersome to store in a ring buffer,
1005 // therefore planar sample formats are not supported.
1006 //
1007 static const enum AVSampleFormat sample_fmts[] = {
1008 AV_SAMPLE_FMT_U8,
1009 AV_SAMPLE_FMT_S16,
1010 AV_SAMPLE_FMT_S32,
1011 AV_SAMPLE_FMT_FLT,
1012 AV_SAMPLE_FMT_DBL,
1013 AV_SAMPLE_FMT_NONE
1014 };
1015
1016 1 static int config_props(AVFilterLink *inlink)
1017 {
1018 1 AVFilterContext *ctx = inlink->dst;
1019 1 ATempoContext *atempo = ctx->priv;
1020
1021 1 enum AVSampleFormat format = inlink->format;
1022 1 int sample_rate = (int)inlink->sample_rate;
1023
1024 1 return yae_reset(atempo, format, sample_rate, inlink->ch_layout.nb_channels);
1025 }
1026
1027 65 static int push_samples(ATempoContext *atempo,
1028 AVFilterLink *outlink,
1029 int n_out)
1030 {
1031 int ret;
1032
1033 65 atempo->dst_buffer->sample_rate = outlink->sample_rate;
1034 65 atempo->dst_buffer->nb_samples = n_out;
1035
1036 // adjust the PTS:
1037 65 atempo->dst_buffer->pts = atempo->start_pts +
1038 65 av_rescale_q(atempo->nsamples_out,
1039 65 (AVRational){ 1, outlink->sample_rate },
1040 outlink->time_base);
1041
1042 65 ret = ff_filter_frame(outlink, atempo->dst_buffer);
1043 65 atempo->dst_buffer = NULL;
1044 65 atempo->dst = NULL;
1045 65 atempo->dst_end = NULL;
1046
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 65 times.
65 if (ret < 0)
1047 return ret;
1048
1049 65 atempo->nsamples_out += n_out;
1050 65 return 0;
1051 }
1052
1053 65 static int filter_frame(AVFilterLink *inlink, AVFrame *src_buffer)
1054 {
1055 65 AVFilterContext *ctx = inlink->dst;
1056 65 ATempoContext *atempo = ctx->priv;
1057 65 AVFilterLink *outlink = ctx->outputs[0];
1058
1059 65 int ret = 0;
1060 65 int n_in = src_buffer->nb_samples;
1061 65 int n_out = (int)(0.5 + ((double)n_in) / atempo->tempo);
1062
1063 65 const uint8_t *src = src_buffer->data[0];
1064 65 const uint8_t *src_end = src + n_in * atempo->stride;
1065
1066
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 64 times.
65 if (atempo->start_pts == AV_NOPTS_VALUE)
1067 1 atempo->start_pts = av_rescale_q(src_buffer->pts,
1068 inlink->time_base,
1069 outlink->time_base);
1070
1071
2/2
✓ Branch 0 taken 128 times.
✓ Branch 1 taken 65 times.
258 while (src < src_end) {
1072
2/2
✓ Branch 0 taken 64 times.
✓ Branch 1 taken 64 times.
128 if (!atempo->dst_buffer) {
1073 64 atempo->dst_buffer = ff_get_audio_buffer(outlink, n_out);
1074
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 64 times.
64 if (!atempo->dst_buffer) {
1075 av_frame_free(&src_buffer);
1076 return AVERROR(ENOMEM);
1077 }
1078 64 av_frame_copy_props(atempo->dst_buffer, src_buffer);
1079
1080 64 atempo->dst = atempo->dst_buffer->data[0];
1081 64 atempo->dst_end = atempo->dst + n_out * atempo->stride;
1082 }
1083
1084 128 yae_apply(atempo, &src, src_end, &atempo->dst, atempo->dst_end);
1085
1086
2/2
✓ Branch 0 taken 64 times.
✓ Branch 1 taken 64 times.
128 if (atempo->dst == atempo->dst_end) {
1087 64 int n_samples = ((atempo->dst - atempo->dst_buffer->data[0]) /
1088 64 atempo->stride);
1089 64 ret = push_samples(atempo, outlink, n_samples);
1090
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 64 times.
64 if (ret < 0)
1091 goto end;
1092 }
1093 }
1094
1095 65 atempo->nsamples_in += n_in;
1096 65 end:
1097 65 av_frame_free(&src_buffer);
1098 65 return ret;
1099 }
1100
1101 66 static int request_frame(AVFilterLink *outlink)
1102 {
1103 66 AVFilterContext *ctx = outlink->src;
1104 66 ATempoContext *atempo = ctx->priv;
1105 int ret;
1106
1107 66 ret = ff_request_frame(ctx->inputs[0]);
1108
1109
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 65 times.
66 if (ret == AVERROR_EOF) {
1110 // flush the filter:
1111 1 int n_max = atempo->ring;
1112 int n_out;
1113 1 int err = AVERROR(EAGAIN);
1114
1115
2/2
✓ Branch 0 taken 1 times.
✓ Branch 1 taken 1 times.
2 while (err == AVERROR(EAGAIN)) {
1116
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
1 if (!atempo->dst_buffer) {
1117 1 atempo->dst_buffer = ff_get_audio_buffer(outlink, n_max);
1118
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (!atempo->dst_buffer)
1119 return AVERROR(ENOMEM);
1120
1121 1 atempo->dst = atempo->dst_buffer->data[0];
1122 1 atempo->dst_end = atempo->dst + n_max * atempo->stride;
1123 }
1124
1125 1 err = yae_flush(atempo, &atempo->dst, atempo->dst_end);
1126
1127 1 n_out = ((atempo->dst - atempo->dst_buffer->data[0]) /
1128 1 atempo->stride);
1129
1130
1/2
✓ Branch 0 taken 1 times.
✗ Branch 1 not taken.
1 if (n_out) {
1131 1 ret = push_samples(atempo, outlink, n_out);
1132
1/2
✗ Branch 0 not taken.
✓ Branch 1 taken 1 times.
1 if (ret < 0)
1133 return ret;
1134 }
1135 }
1136
1137 1 av_frame_free(&atempo->dst_buffer);
1138 1 atempo->dst = NULL;
1139 1 atempo->dst_end = NULL;
1140
1141 1 return AVERROR_EOF;
1142 }
1143
1144 65 return ret;
1145 }
1146
1147 static int process_command(AVFilterContext *ctx,
1148 const char *cmd,
1149 const char *arg,
1150 char *res,
1151 int res_len,
1152 int flags)
1153 {
1154 int ret = ff_filter_process_command(ctx, cmd, arg, res, res_len, flags);
1155
1156 if (ret < 0)
1157 return ret;
1158
1159 return yae_update(ctx);
1160 }
1161
1162 static const AVFilterPad atempo_inputs[] = {
1163 {
1164 .name = "default",
1165 .type = AVMEDIA_TYPE_AUDIO,
1166 .filter_frame = filter_frame,
1167 .config_props = config_props,
1168 },
1169 };
1170
1171 static const AVFilterPad atempo_outputs[] = {
1172 {
1173 .name = "default",
1174 .request_frame = request_frame,
1175 .type = AVMEDIA_TYPE_AUDIO,
1176 },
1177 };
1178
1179 const AVFilter ff_af_atempo = {
1180 .name = "atempo",
1181 .description = NULL_IF_CONFIG_SMALL("Adjust audio tempo."),
1182 .init = init,
1183 .uninit = uninit,
1184 .process_command = process_command,
1185 .priv_size = sizeof(ATempoContext),
1186 .priv_class = &atempo_class,
1187 FILTER_INPUTS(atempo_inputs),
1188 FILTER_OUTPUTS(atempo_outputs),
1189 FILTER_SAMPLEFMTS_ARRAY(sample_fmts),
1190 };
1191