FFmpeg coverage


Directory: ../../../ffmpeg/
File: src/libavfilter/af_adeclick.c
Date: 2026-09-25 17:25:24
Exec Total Coverage
Lines: 0 409 0.0%
Functions: 0 15 0.0%
Branches: 0 262 0.0%

Line Branch Exec Source
1 /*
2 * Copyright (c) 2018 Paul B Mahol
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 #include "libavutil/audio_fifo.h"
22 #include "libavutil/mem.h"
23 #include "libavutil/opt.h"
24 #include "libavutil/tx.h"
25 #include "avfilter.h"
26 #include "audio.h"
27 #include "filters.h"
28
29 typedef struct DeclickChannel {
30 double *auxiliary;
31 double *detection;
32 double *acoefficients;
33 double *acorrelation;
34 double *tmp;
35 double *interpolated;
36 double *matrix;
37 int matrix_size;
38 double *vector;
39 int vector_size;
40 double *y;
41 int y_size;
42 uint8_t *click;
43 int *index;
44 unsigned *histogram;
45 int histogram_size;
46 } DeclickChannel;
47
48 typedef struct AudioDeclickContext {
49 const AVClass *class;
50
51 double w;
52 double overlap;
53 double threshold;
54 double ar;
55 double burst;
56 int method;
57 int nb_hbins;
58
59 int is_declip;
60 int ar_order;
61 int nb_burst_samples;
62 int window_size;
63 int hop_size;
64 int overlap_skip;
65
66 AVFrame *enabled;
67 AVFrame *in;
68 AVFrame *out;
69 AVFrame *buffer;
70 AVFrame *is;
71
72 DeclickChannel *chan;
73
74 int64_t pts;
75 int nb_channels;
76 uint64_t nb_samples;
77 uint64_t detected_errors;
78 int samples_left;
79 int eof;
80
81 AVAudioFifo *efifo;
82 AVAudioFifo *fifo;
83 double *window_func_lut;
84
85 int (*detector)(struct AudioDeclickContext *s, DeclickChannel *c,
86 double sigmae, double *detection,
87 double *acoefficients, uint8_t *click, int *index,
88 const double *src, double *dst);
89 } AudioDeclickContext;
90
91 #define OFFSET(x) offsetof(AudioDeclickContext, x)
92 #define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
93
94 static const AVOption adeclick_options[] = {
95 { "window", "set window size", OFFSET(w), AV_OPT_TYPE_DOUBLE, {.dbl=55}, 10, 100, AF },
96 { "w", "set window size", OFFSET(w), AV_OPT_TYPE_DOUBLE, {.dbl=55}, 10, 100, AF },
97 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_DOUBLE, {.dbl=75}, 50, 95, AF },
98 { "o", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_DOUBLE, {.dbl=75}, 50, 95, AF },
99 { "arorder", "set autoregression order", OFFSET(ar), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 0, 25, AF },
100 { "a", "set autoregression order", OFFSET(ar), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 0, 25, AF },
101 { "threshold", "set threshold", OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 1, 100, AF },
102 { "t", "set threshold", OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 1, 100, AF },
103 { "burst", "set burst fusion", OFFSET(burst), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 0, 10, AF },
104 { "b", "set burst fusion", OFFSET(burst), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 0, 10, AF },
105 { "method", "set overlap method", OFFSET(method), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, .unit = "m" },
106 { "m", "set overlap method", OFFSET(method), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, .unit = "m" },
107 { "add", "overlap-add", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, .unit = "m" },
108 { "a", "overlap-add", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, .unit = "m" },
109 { "save", "overlap-save", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, .unit = "m" },
110 { "s", "overlap-save", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, .unit = "m" },
111 { NULL }
112 };
113
114 AVFILTER_DEFINE_CLASS(adeclick);
115
116 ✗ static int config_input(AVFilterLink *inlink)
117 {
118 ✗ AVFilterContext *ctx = inlink->dst;
119 ✗ AudioDeclickContext *s = ctx->priv;
120 int i;
121
122 ✗ s->pts = AV_NOPTS_VALUE;
123 ✗ s->window_size = FFMAX(100, inlink->sample_rate * s->w / 1000.);
124 ✗ s->ar_order = FFMAX(s->window_size * s->ar / 100., 1);
125 ✗ s->nb_burst_samples = s->window_size * s->burst / 1000.;
126 ✗ s->hop_size = FFMAX(1, s->window_size * (1. - (s->overlap / 100.)));
127
128 ✗ s->window_func_lut = av_calloc(s->window_size, sizeof(*s->window_func_lut));
129 ✗ if (!s->window_func_lut)
130 ✗ return AVERROR(ENOMEM);
131
132 {
133 ✗ double *tx_in[2] = { NULL }, *tx_out[2] = { NULL };
134 ✗ AVTXContext *tx = NULL, *itx = NULL;
135 av_tx_fn tx_fn, itx_fn;
136 int ret, tx_size;
137 double scale;
138
139 ✗ tx_size = 1 << (32 - ff_clz(s->window_size));
140
141 ✗ scale = 1.0;
142 ✗ ret = av_tx_init(&tx, &tx_fn, AV_TX_DOUBLE_RDFT, 0, tx_size, &scale, 0);
143 ✗ if (ret < 0)
144 ✗ goto tx_end;
145
146 ✗ scale = 1.0 / tx_size;
147 ✗ ret = av_tx_init(&itx, &itx_fn, AV_TX_DOUBLE_RDFT, 1, tx_size, &scale, 0);
148 ✗ if (ret < 0)
149 ✗ goto tx_end;
150
151 ✗ tx_in[0] = av_calloc(tx_size + 2, sizeof(*tx_in[0]));
152 ✗ tx_in[1] = av_calloc(tx_size + 2, sizeof(*tx_in[1]));
153 ✗ tx_out[0] = av_calloc(tx_size + 2, sizeof(*tx_out[0]));
154 ✗ tx_out[1] = av_calloc(tx_size + 2, sizeof(*tx_out[1]));
155 ✗ if (!tx_in[0] || !tx_in[1] || !tx_out[0] || !tx_out[1]) {
156 ✗ ret = AVERROR(ENOMEM);
157 ✗ goto tx_end;
158 }
159
160 ✗ for (int n = 0; n < s->window_size - s->hop_size; n++)
161 ✗ tx_in[0][n] = 1.0;
162
163 ✗ for (int n = 0; n < s->hop_size; n++)
164 ✗ tx_in[1][n] = 1.0;
165
166 ✗ tx_fn(tx, tx_out[0], tx_in[0], sizeof(double));
167 ✗ tx_fn(tx, tx_out[1], tx_in[1], sizeof(double));
168
169 ✗ for (int n = 0; n <= tx_size/2; n++) {
170 ✗ double re0 = tx_out[0][2*n];
171 ✗ double im0 = tx_out[0][2*n+1];
172 ✗ double re1 = tx_out[1][2*n];
173 ✗ double im1 = tx_out[1][2*n+1];
174
175 ✗ tx_in[0][2*n] = re0 * re1 - im0 * im1;
176 ✗ tx_in[0][2*n+1] = re0 * im1 + re1 * im0;
177 }
178
179 ✗ itx_fn(itx, tx_out[0], tx_in[0], sizeof(AVComplexDouble));
180
181 ✗ scale = 1.0 / (s->window_size - s->hop_size);
182 ✗ for (int n = 0; n < s->window_size; n++)
183 ✗ s->window_func_lut[n] = tx_out[0][n] * scale;
184
185 ✗ tx_end:
186 ✗ av_tx_uninit(&tx);
187 ✗ av_tx_uninit(&itx);
188
189 ✗ av_freep(&tx_in[0]);
190 ✗ av_freep(&tx_in[1]);
191 ✗ av_freep(&tx_out[0]);
192 ✗ av_freep(&tx_out[1]);
193
194 ✗ if (ret < 0)
195 ✗ return ret;
196 }
197
198 ✗ av_frame_free(&s->in);
199 ✗ av_frame_free(&s->out);
200 ✗ av_frame_free(&s->buffer);
201 ✗ av_frame_free(&s->is);
202 ✗ s->enabled = ff_get_audio_buffer(inlink, s->window_size);
203 ✗ s->in = ff_get_audio_buffer(inlink, s->window_size);
204 ✗ s->out = ff_get_audio_buffer(inlink, s->window_size);
205 ✗ s->buffer = ff_get_audio_buffer(inlink, s->window_size * 2);
206 ✗ s->is = ff_get_audio_buffer(inlink, s->window_size);
207 ✗ if (!s->in || !s->out || !s->buffer || !s->is || !s->enabled)
208 ✗ return AVERROR(ENOMEM);
209
210 ✗ s->efifo = av_audio_fifo_alloc(inlink->format, 1, s->window_size);
211 ✗ if (!s->efifo)
212 ✗ return AVERROR(ENOMEM);
213 ✗ s->fifo = av_audio_fifo_alloc(inlink->format, inlink->ch_layout.nb_channels, s->window_size);
214 ✗ if (!s->fifo)
215 ✗ return AVERROR(ENOMEM);
216 ✗ s->overlap_skip = s->method ? (s->window_size - s->hop_size) / 2 : 0;
217 ✗ if (s->overlap_skip > 0) {
218 ✗ av_audio_fifo_write(s->fifo, (void **)s->in->extended_data,
219 s->overlap_skip);
220 }
221
222 ✗ s->nb_channels = inlink->ch_layout.nb_channels;
223 ✗ s->chan = av_calloc(inlink->ch_layout.nb_channels, sizeof(*s->chan));
224 ✗ if (!s->chan)
225 ✗ return AVERROR(ENOMEM);
226
227 ✗ for (i = 0; i < inlink->ch_layout.nb_channels; i++) {
228 ✗ DeclickChannel *c = &s->chan[i];
229
230 ✗ c->detection = av_calloc(s->window_size, sizeof(*c->detection));
231 ✗ c->auxiliary = av_calloc(s->ar_order + 1, sizeof(*c->auxiliary));
232 ✗ c->acoefficients = av_calloc(s->ar_order + 1, sizeof(*c->acoefficients));
233 ✗ c->acorrelation = av_calloc(s->ar_order + 1, sizeof(*c->acorrelation));
234 ✗ c->tmp = av_calloc(s->ar_order, sizeof(*c->tmp));
235 ✗ c->click = av_calloc(s->window_size, sizeof(*c->click));
236 ✗ c->index = av_calloc(s->window_size, sizeof(*c->index));
237 ✗ c->interpolated = av_calloc(s->window_size, sizeof(*c->interpolated));
238 ✗ if (!c->auxiliary || !c->acoefficients || !c->detection || !c->click ||
239 ✗ !c->index || !c->interpolated || !c->acorrelation || !c->tmp)
240 ✗ return AVERROR(ENOMEM);
241 }
242
243 ✗ return 0;
244 }
245
246 ✗ static void autocorrelation(const double *input, int order, int size,
247 double *output, double scale)
248 {
249 int i, j;
250
251 ✗ for (i = 0; i <= order; i++) {
252 ✗ double value = 0.;
253
254 ✗ for (j = i; j < size; j++)
255 ✗ value += input[j] * input[j - i];
256
257 ✗ output[i] = value * scale;
258 }
259 ✗ }
260
261 ✗ static double autoregression(const double *samples, int ar_order,
262 int nb_samples, double *k, double *r, double *a)
263 {
264 double alpha;
265 int i, j;
266
267 ✗ memset(a, 0, ar_order * sizeof(*a));
268
269 ✗ autocorrelation(samples, ar_order, nb_samples, r, 1. / nb_samples);
270
271 /* Levinson-Durbin algorithm */
272 ✗ k[0] = a[0] = -r[1] / r[0];
273 ✗ alpha = r[0] * (1. - k[0] * k[0]);
274 ✗ for (i = 1; i < ar_order; i++) {
275 ✗ double epsilon = 0.;
276
277 ✗ for (j = 0; j < i; j++)
278 ✗ epsilon += a[j] * r[i - j];
279 ✗ epsilon += r[i + 1];
280
281 ✗ k[i] = -epsilon / alpha;
282 ✗ alpha *= (1. - k[i] * k[i]);
283 ✗ for (j = i - 1; j >= 0; j--)
284 ✗ k[j] = a[j] + k[i] * a[i - j - 1];
285 ✗ for (j = 0; j <= i; j++)
286 ✗ a[j] = k[j];
287 }
288
289 ✗ k[0] = 1.;
290 ✗ for (i = 1; i <= ar_order; i++)
291 ✗ k[i] = a[i - 1];
292
293 ✗ return sqrt(alpha);
294 }
295
296 ✗ static int isfinite_array(double *samples, int nb_samples)
297 {
298 int i;
299
300 ✗ for (i = 0; i < nb_samples; i++)
301 ✗ if (!isfinite(samples[i]))
302 ✗ return 0;
303
304 ✗ return 1;
305 }
306
307 ✗ static int find_index(int *index, int value, int size)
308 {
309 int i, start, end;
310
311 ✗ if ((value < index[0]) || (value > index[size - 1]))
312 ✗ return 1;
313
314 ✗ i = start = 0;
315 ✗ end = size - 1;
316
317 ✗ while (start <= end) {
318 ✗ i = (end + start) / 2;
319 ✗ if (index[i] == value)
320 ✗ return 0;
321 ✗ if (value < index[i])
322 ✗ end = i - 1;
323 ✗ if (value > index[i])
324 ✗ start = i + 1;
325 }
326
327 ✗ return 1;
328 }
329
330 ✗ static int factorization(double *matrix, int n)
331 {
332 int i, j, k;
333
334 ✗ for (i = 0; i < n; i++) {
335 ✗ const int in = i * n;
336 double value;
337
338 ✗ value = matrix[in + i];
339 ✗ for (j = 0; j < i; j++)
340 ✗ value -= matrix[j * n + j] * matrix[in + j] * matrix[in + j];
341
342 ✗ if (value == 0.) {
343 ✗ return -1;
344 }
345
346 ✗ matrix[in + i] = value;
347 ✗ for (j = i + 1; j < n; j++) {
348 ✗ const int jn = j * n;
349 double x;
350
351 ✗ x = matrix[jn + i];
352 ✗ for (k = 0; k < i; k++)
353 ✗ x -= matrix[k * n + k] * matrix[in + k] * matrix[jn + k];
354 ✗ matrix[jn + i] = x / matrix[in + i];
355 }
356 }
357
358 ✗ return 0;
359 }
360
361 ✗ static int do_interpolation(DeclickChannel *c, double *matrix,
362 double *vector, int n, double *out)
363 {
364 int i, j, ret;
365 double *y;
366
367 ✗ ret = factorization(matrix, n);
368 ✗ if (ret < 0)
369 ✗ return ret;
370
371 ✗ av_fast_malloc(&c->y, &c->y_size, n * sizeof(*c->y));
372 ✗ y = c->y;
373 ✗ if (!y)
374 ✗ return AVERROR(ENOMEM);
375
376 ✗ for (i = 0; i < n; i++) {
377 ✗ const int in = i * n;
378 double value;
379
380 ✗ value = vector[i];
381 ✗ for (j = 0; j < i; j++)
382 ✗ value -= matrix[in + j] * y[j];
383 ✗ y[i] = value;
384 }
385
386 ✗ for (i = n - 1; i >= 0; i--) {
387 ✗ out[i] = y[i] / matrix[i * n + i];
388 ✗ for (j = i + 1; j < n; j++)
389 ✗ out[i] -= matrix[j * n + i] * out[j];
390 }
391
392 ✗ return 0;
393 }
394
395 ✗ static int interpolation(DeclickChannel *c, const double *src, int ar_order,
396 double *acoefficients, int *index, int nb_errors,
397 double *auxiliary, double *interpolated)
398 {
399 double *vector, *matrix;
400 int i, j;
401
402 ✗ av_fast_malloc(&c->matrix, &c->matrix_size, nb_errors * nb_errors * sizeof(*c->matrix));
403 ✗ matrix = c->matrix;
404 ✗ if (!matrix)
405 ✗ return AVERROR(ENOMEM);
406
407 ✗ av_fast_malloc(&c->vector, &c->vector_size, nb_errors * sizeof(*c->vector));
408 ✗ vector = c->vector;
409 ✗ if (!vector)
410 ✗ return AVERROR(ENOMEM);
411
412 ✗ autocorrelation(acoefficients, ar_order, ar_order + 1, auxiliary, 1.);
413
414 ✗ for (i = 0; i < nb_errors; i++) {
415 ✗ const int im = i * nb_errors;
416
417 ✗ for (j = i; j < nb_errors; j++) {
418 ✗ if (abs(index[j] - index[i]) <= ar_order) {
419 ✗ matrix[j * nb_errors + i] = matrix[im + j] = auxiliary[abs(index[j] - index[i])];
420 } else {
421 ✗ matrix[j * nb_errors + i] = matrix[im + j] = 0;
422 }
423 }
424 }
425
426 ✗ for (i = 0; i < nb_errors; i++) {
427 ✗ double value = 0.;
428
429 ✗ for (j = -ar_order; j <= ar_order; j++)
430 ✗ if (find_index(index, index[i] - j, nb_errors))
431 ✗ value -= src[index[i] - j] * auxiliary[abs(j)];
432
433 ✗ vector[i] = value;
434 }
435
436 ✗ return do_interpolation(c, matrix, vector, nb_errors, interpolated);
437 }
438
439 ✗ static int detect_clips(AudioDeclickContext *s, DeclickChannel *c,
440 double unused0,
441 double *unused1, double *unused2,
442 uint8_t *clip, int *index,
443 const double *src, double *dst)
444 {
445 ✗ const double threshold = s->threshold;
446 ✗ double max_amplitude = 0;
447 unsigned *histogram;
448 ✗ int i, nb_clips = 0;
449
450 ✗ av_fast_malloc(&c->histogram, &c->histogram_size, s->nb_hbins * sizeof(*c->histogram));
451 ✗ if (!c->histogram)
452 ✗ return AVERROR(ENOMEM);
453 ✗ histogram = c->histogram;
454 ✗ memset(histogram, 0, sizeof(*histogram) * s->nb_hbins);
455
456 ✗ for (i = 0; i < s->window_size; i++) {
457 ✗ const unsigned index = fmin(fabs(src[i]), 1) * (s->nb_hbins - 1);
458
459 ✗ histogram[index]++;
460 ✗ dst[i] = src[i];
461 ✗ clip[i] = 0;
462 }
463
464 ✗ for (i = s->nb_hbins - 1; i > 1; i--) {
465 ✗ if (histogram[i]) {
466 ✗ if (histogram[i] / (double)FFMAX(histogram[i - 1], 1) > threshold) {
467 ✗ max_amplitude = i / (double)s->nb_hbins;
468 }
469 ✗ break;
470 }
471 }
472
473 ✗ if (max_amplitude > 0.) {
474 ✗ for (i = 0; i < s->window_size; i++) {
475 ✗ clip[i] = fabs(src[i]) >= max_amplitude;
476 }
477 }
478
479 ✗ memset(clip, 0, s->ar_order * sizeof(*clip));
480 ✗ memset(clip + (s->window_size - s->ar_order), 0, s->ar_order * sizeof(*clip));
481
482 ✗ for (i = s->ar_order; i < s->window_size - s->ar_order; i++)
483 ✗ if (clip[i])
484 ✗ index[nb_clips++] = i;
485
486 ✗ return nb_clips;
487 }
488
489 ✗ static int detect_clicks(AudioDeclickContext *s, DeclickChannel *c,
490 double sigmae,
491 double *detection, double *acoefficients,
492 uint8_t *click, int *index,
493 const double *src, double *dst)
494 {
495 ✗ const double threshold = s->threshold;
496 ✗ int i, j, nb_clicks = 0, prev = -1;
497
498 ✗ memset(detection, 0, s->window_size * sizeof(*detection));
499
500 ✗ for (i = s->ar_order; i < s->window_size; i++) {
501 ✗ for (j = 0; j <= s->ar_order; j++) {
502 ✗ detection[i] += acoefficients[j] * src[i - j];
503 }
504 }
505
506 ✗ for (i = 0; i < s->window_size; i++) {
507 ✗ click[i] = fabs(detection[i]) > sigmae * threshold;
508 ✗ dst[i] = src[i];
509 }
510
511 ✗ for (i = 0; i < s->window_size; i++) {
512 ✗ if (!click[i])
513 ✗ continue;
514
515 ✗ if (prev >= 0 && (i > prev + 1) && (i <= s->nb_burst_samples + prev))
516 ✗ for (j = prev + 1; j < i; j++)
517 ✗ click[j] = 1;
518 ✗ prev = i;
519 }
520
521 ✗ memset(click, 0, s->ar_order * sizeof(*click));
522 ✗ memset(click + (s->window_size - s->ar_order), 0, s->ar_order * sizeof(*click));
523
524 ✗ for (i = s->ar_order; i < s->window_size - s->ar_order; i++)
525 ✗ if (click[i])
526 ✗ index[nb_clicks++] = i;
527
528 ✗ return nb_clicks;
529 }
530
531 typedef struct ThreadData {
532 AVFrame *out;
533 } ThreadData;
534
535 ✗ static int filter_channel(AVFilterContext *ctx, void *arg, int ch, int nb_jobs)
536 {
537 ✗ AudioDeclickContext *s = ctx->priv;
538 ✗ ThreadData *td = arg;
539 ✗ AVFrame *out = td->out;
540 ✗ const double *src = (const double *)s->in->extended_data[ch];
541 ✗ double *is = (double *)s->is->extended_data[ch];
542 ✗ double *dst = (double *)s->out->extended_data[ch];
543 ✗ double *ptr = (double *)out->extended_data[ch];
544 ✗ double *buf = (double *)s->buffer->extended_data[ch];
545 ✗ const double *w = s->window_func_lut;
546 ✗ DeclickChannel *c = &s->chan[ch];
547 double sigmae;
548 int j, ret;
549
550 ✗ sigmae = autoregression(src, s->ar_order, s->window_size, c->acoefficients, c->acorrelation, c->tmp);
551
552 ✗ if (isfinite_array(c->acoefficients, s->ar_order + 1)) {
553 ✗ double *interpolated = c->interpolated;
554 ✗ int *index = c->index;
555 int nb_errors;
556
557 ✗ nb_errors = s->detector(s, c, sigmae, c->detection, c->acoefficients,
558 c->click, index, src, dst);
559 ✗ if (nb_errors > 0) {
560 ✗ double *enabled = (double *)s->enabled->extended_data[0];
561
562 ✗ ret = interpolation(c, src, s->ar_order, c->acoefficients, index,
563 nb_errors, c->auxiliary, interpolated);
564 ✗ if (ret < 0)
565 ✗ return ret;
566
567 ✗ av_audio_fifo_peek(s->efifo, (void**)s->enabled->extended_data, s->window_size);
568
569 ✗ for (j = 0; j < nb_errors; j++) {
570 ✗ if (enabled[index[j]]) {
571 ✗ dst[index[j]] = interpolated[j];
572 ✗ is[index[j]] = 1;
573 }
574 }
575 }
576 } else {
577 ✗ memcpy(dst, src, s->window_size * sizeof(*dst));
578 }
579
580 ✗ if (s->method == 0) {
581 ✗ for (j = 0; j < s->window_size; j++)
582 ✗ buf[j] += dst[j] * w[j];
583 } else {
584 ✗ const int skip = s->overlap_skip;
585
586 ✗ for (j = 0; j < s->hop_size; j++)
587 ✗ buf[j] = dst[skip + j];
588 }
589 ✗ for (j = 0; j < s->hop_size; j++)
590 ✗ ptr[j] = buf[j];
591
592 ✗ memmove(buf, buf + s->hop_size, (s->window_size * 2 - s->hop_size) * sizeof(*buf));
593 ✗ memmove(is, is + s->hop_size, (s->window_size - s->hop_size) * sizeof(*is));
594 ✗ memset(buf + s->window_size * 2 - s->hop_size, 0, s->hop_size * sizeof(*buf));
595 ✗ memset(is + s->window_size - s->hop_size, 0, s->hop_size * sizeof(*is));
596
597 ✗ return 0;
598 }
599
600 ✗ static int filter_frame(AVFilterLink *inlink)
601 {
602 ✗ AVFilterContext *ctx = inlink->dst;
603 ✗ AVFilterLink *outlink = ctx->outputs[0];
604 ✗ AudioDeclickContext *s = ctx->priv;
605 ✗ AVFrame *out = NULL;
606 ✗ int ret = 0, j, ch, detected_errors = 0;
607 ThreadData td;
608
609 ✗ out = ff_get_audio_buffer(outlink, s->hop_size);
610 ✗ if (!out)
611 ✗ return AVERROR(ENOMEM);
612
613 ✗ ret = av_audio_fifo_peek(s->fifo, (void **)s->in->extended_data,
614 s->window_size);
615 ✗ if (ret < 0)
616 ✗ goto fail;
617
618 ✗ td.out = out;
619 ✗ ret = ff_filter_execute(ctx, filter_channel, &td, NULL, inlink->ch_layout.nb_channels);
620 ✗ if (ret < 0)
621 ✗ goto fail;
622
623 ✗ for (ch = 0; ch < s->in->ch_layout.nb_channels; ch++) {
624 ✗ double *is = (double *)s->is->extended_data[ch];
625
626 ✗ for (j = 0; j < s->hop_size; j++) {
627 ✗ if (is[j])
628 ✗ detected_errors++;
629 }
630 }
631
632 ✗ av_audio_fifo_drain(s->fifo, s->hop_size);
633 ✗ av_audio_fifo_drain(s->efifo, s->hop_size);
634
635 ✗ if (s->samples_left > 0)
636 ✗ out->nb_samples = FFMIN(s->hop_size, s->samples_left);
637
638 ✗ out->pts = s->pts;
639 ✗ s->pts += av_rescale_q(s->hop_size, (AVRational){1, outlink->sample_rate}, outlink->time_base);
640
641 ✗ s->detected_errors += detected_errors;
642 ✗ s->nb_samples += out->nb_samples * inlink->ch_layout.nb_channels;
643
644 ✗ ret = ff_filter_frame(outlink, out);
645 ✗ if (ret < 0)
646 ✗ return ret;
647
648 ✗ if (s->samples_left > 0) {
649 ✗ s->samples_left -= s->hop_size;
650 ✗ if (s->samples_left <= 0)
651 ✗ av_audio_fifo_drain(s->fifo, av_audio_fifo_size(s->fifo));
652 }
653
654 ✗ fail:
655 ✗ if (ret < 0)
656 ✗ av_frame_free(&out);
657 ✗ return ret;
658 }
659
660 ✗ static int activate(AVFilterContext *ctx)
661 {
662 ✗ AVFilterLink *inlink = ctx->inputs[0];
663 ✗ AVFilterLink *outlink = ctx->outputs[0];
664 ✗ AudioDeclickContext *s = ctx->priv;
665 AVFrame *in;
666 int ret, status;
667 int64_t pts;
668
669 ✗ FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink);
670
671 ✗ ret = ff_inlink_consume_samples(inlink, s->window_size, s->window_size, &in);
672 ✗ if (ret < 0)
673 ✗ return ret;
674 ✗ if (ret > 0) {
675 ✗ double *e = (double *)s->enabled->extended_data[0];
676
677 ✗ if (s->pts == AV_NOPTS_VALUE)
678 ✗ s->pts = in->pts;
679
680 ✗ ret = av_audio_fifo_write(s->fifo, (void **)in->extended_data,
681 ✗ in->nb_samples);
682 ✗ for (int i = 0; i < in->nb_samples; i++)
683 ✗ e[i] = !ctx->is_disabled;
684
685 ✗ av_audio_fifo_write(s->efifo, (void**)s->enabled->extended_data, in->nb_samples);
686 ✗ av_frame_free(&in);
687 ✗ if (ret < 0)
688 ✗ return ret;
689 }
690
691 ✗ if (av_audio_fifo_size(s->fifo) >= s->window_size ||
692 ✗ s->samples_left > 0)
693 ✗ return filter_frame(inlink);
694
695 ✗ if (av_audio_fifo_size(s->fifo) >= s->window_size) {
696 ✗ ff_filter_set_ready(ctx, 100);
697 ✗ return 0;
698 }
699
700 ✗ if (!s->eof && ff_inlink_acknowledge_status(inlink, &status, &pts)) {
701 ✗ if (status == AVERROR_EOF) {
702 ✗ s->eof = 1;
703 ✗ s->samples_left = av_audio_fifo_size(s->fifo) - s->overlap_skip;
704 ✗ ff_filter_set_ready(ctx, 100);
705 ✗ return 0;
706 }
707 }
708
709 ✗ if (s->eof && s->samples_left <= 0) {
710 ✗ ff_outlink_set_status(outlink, AVERROR_EOF, s->pts);
711 ✗ return 0;
712 }
713
714 ✗ if (!s->eof)
715 ✗ FF_FILTER_FORWARD_WANTED(outlink, inlink);
716
717 ✗ return FFERROR_NOT_READY;
718 }
719
720 ✗ static av_cold int init(AVFilterContext *ctx)
721 {
722 ✗ AudioDeclickContext *s = ctx->priv;
723
724 ✗ s->is_declip = !strcmp(ctx->filter->name, "adeclip");
725 ✗ if (s->is_declip) {
726 ✗ s->detector = detect_clips;
727 } else {
728 ✗ s->detector = detect_clicks;
729 }
730
731 ✗ return 0;
732 }
733
734 ✗ static av_cold void uninit(AVFilterContext *ctx)
735 {
736 ✗ AudioDeclickContext *s = ctx->priv;
737 int i;
738
739 ✗ if (s->nb_samples > 0)
740 ✗ av_log(ctx, AV_LOG_INFO, "Detected %s in %"PRId64" of %"PRId64" samples (%g%%).\n",
741 ✗ s->is_declip ? "clips" : "clicks", s->detected_errors,
742 ✗ s->nb_samples, 100. * s->detected_errors / s->nb_samples);
743
744 ✗ av_audio_fifo_free(s->fifo);
745 ✗ av_audio_fifo_free(s->efifo);
746 ✗ av_freep(&s->window_func_lut);
747 ✗ av_frame_free(&s->enabled);
748 ✗ av_frame_free(&s->in);
749 ✗ av_frame_free(&s->out);
750 ✗ av_frame_free(&s->buffer);
751 ✗ av_frame_free(&s->is);
752
753 ✗ if (s->chan) {
754 ✗ for (i = 0; i < s->nb_channels; i++) {
755 ✗ DeclickChannel *c = &s->chan[i];
756
757 ✗ av_freep(&c->detection);
758 ✗ av_freep(&c->auxiliary);
759 ✗ av_freep(&c->acoefficients);
760 ✗ av_freep(&c->acorrelation);
761 ✗ av_freep(&c->tmp);
762 ✗ av_freep(&c->click);
763 ✗ av_freep(&c->index);
764 ✗ av_freep(&c->interpolated);
765 ✗ av_freep(&c->matrix);
766 ✗ c->matrix_size = 0;
767 ✗ av_freep(&c->histogram);
768 ✗ c->histogram_size = 0;
769 ✗ av_freep(&c->vector);
770 ✗ c->vector_size = 0;
771 ✗ av_freep(&c->y);
772 ✗ c->y_size = 0;
773 }
774 }
775 ✗ av_freep(&s->chan);
776 ✗ s->nb_channels = 0;
777 ✗ }
778
779 static const AVFilterPad inputs[] = {
780 {
781 .name = "default",
782 .type = AVMEDIA_TYPE_AUDIO,
783 .config_props = config_input,
784 },
785 };
786
787 const FFFilter ff_af_adeclick = {
788 .p.name = "adeclick",
789 .p.description = NULL_IF_CONFIG_SMALL("Remove impulsive noise from input audio."),
790 .p.priv_class = &adeclick_class,
791 .p.flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
792 .priv_size = sizeof(AudioDeclickContext),
793 .init = init,
794 .activate = activate,
795 .uninit = uninit,
796 FILTER_INPUTS(inputs),
797 FILTER_OUTPUTS(ff_audio_default_filterpad),
798 FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_DBLP),
799 };
800
801 static const AVOption adeclip_options[] = {
802 { "window", "set window size", OFFSET(w), AV_OPT_TYPE_DOUBLE, {.dbl=55}, 10, 100, AF },
803 { "w", "set window size", OFFSET(w), AV_OPT_TYPE_DOUBLE, {.dbl=55}, 10, 100, AF },
804 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_DOUBLE, {.dbl=75}, 50, 95, AF },
805 { "o", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_DOUBLE, {.dbl=75}, 50, 95, AF },
806 { "arorder", "set autoregression order", OFFSET(ar), AV_OPT_TYPE_DOUBLE, {.dbl=8}, 0, 25, AF },
807 { "a", "set autoregression order", OFFSET(ar), AV_OPT_TYPE_DOUBLE, {.dbl=8}, 0, 25, AF },
808 { "threshold", "set threshold", OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=10}, 1, 100, AF },
809 { "t", "set threshold", OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=10}, 1, 100, AF },
810 { "hsize", "set histogram size", OFFSET(nb_hbins), AV_OPT_TYPE_INT, {.i64=1000}, 100, 9999, AF },
811 { "n", "set histogram size", OFFSET(nb_hbins), AV_OPT_TYPE_INT, {.i64=1000}, 100, 9999, AF },
812 { "method", "set overlap method", OFFSET(method), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, .unit = "m" },
813 { "m", "set overlap method", OFFSET(method), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, .unit = "m" },
814 { "add", "overlap-add", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, .unit = "m" },
815 { "a", "overlap-add", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, .unit = "m" },
816 { "save", "overlap-save", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, .unit = "m" },
817 { "s", "overlap-save", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, .unit = "m" },
818 { NULL }
819 };
820
821 AVFILTER_DEFINE_CLASS(adeclip);
822
823 const FFFilter ff_af_adeclip = {
824 .p.name = "adeclip",
825 .p.description = NULL_IF_CONFIG_SMALL("Remove clipping from input audio."),
826 .p.priv_class = &adeclip_class,
827 .p.flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL,
828 .priv_size = sizeof(AudioDeclickContext),
829 .init = init,
830 .activate = activate,
831 .uninit = uninit,
832 FILTER_INPUTS(inputs),
833 FILTER_OUTPUTS(ff_audio_default_filterpad),
834 FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_DBLP),
835 };
836