aboutsummaryrefslogtreecommitdiffstats
path: root/libavfilter/af_dynaudnorm.c
blob: e58564871aefb313cdddd31c899ff3845458313b (plain) (blame)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
/*
 * Dynamic Audio Normalizer
 * Copyright (c) 2015 LoRd_MuldeR <mulder2@gmx.de>. Some rights reserved.
 *
 * This file is part of FFmpeg.
 *
 * FFmpeg is free software; you can redistribute it and/or
 * modify it under the terms of the GNU Lesser General Public
 * License as published by the Free Software Foundation; either
 * version 2.1 of the License, or (at your option) any later version.
 *
 * FFmpeg is distributed in the hope that it will be useful,
 * but WITHOUT ANY WARRANTY; without even the implied warranty of
 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
 * Lesser General Public License for more details.
 *
 * You should have received a copy of the GNU Lesser General Public
 * License along with FFmpeg; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
 */

/**
 * @file
 * Dynamic Audio Normalizer
 */

#include <float.h>

#include "libavutil/avassert.h"
#include "libavutil/opt.h"

#define FF_BUFQUEUE_SIZE 302
#include "libavfilter/bufferqueue.h"

#include "audio.h"
#include "avfilter.h"
#include "internal.h"

typedef struct cqueue {
    double *elements;
    int size;
    int nb_elements;
    int first;
} cqueue;

typedef struct DynamicAudioNormalizerContext {
    const AVClass *class;

    struct FFBufQueue queue;

    int frame_len;
    int frame_len_msec;
    int filter_size;
    int dc_correction;
    int channels_coupled;
    int alt_boundary_mode;

    double peak_value;
    double max_amplification;
    double target_rms;
    double compress_factor;
    double *prev_amplification_factor;
    double *dc_correction_value;
    double *compress_threshold;
    double *fade_factors[2];
    double *weights;

    int channels;
    int delay;

    cqueue **gain_history_original;
    cqueue **gain_history_minimum;
    cqueue **gain_history_smoothed;
} DynamicAudioNormalizerContext;

#define OFFSET(x) offsetof(DynamicAudioNormalizerContext, x)
#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM

static const AVOption dynaudnorm_options[] = {
    { "f", "set the frame length in msec",     OFFSET(frame_len_msec),    AV_OPT_TYPE_INT,    {.i64 = 500},   10,  8000, FLAGS },
    { "g", "set the filter size",              OFFSET(filter_size),       AV_OPT_TYPE_INT,    {.i64 = 31},     3,   301, FLAGS },
    { "p", "set the peak value",               OFFSET(peak_value),        AV_OPT_TYPE_DOUBLE, {.dbl = 0.95}, 0.0,   1.0, FLAGS },
    { "m", "set the max amplification",        OFFSET(max_amplification), AV_OPT_TYPE_DOUBLE, {.dbl = 10.0}, 1.0, 100.0, FLAGS },
    { "r", "set the target RMS",               OFFSET(target_rms),        AV_OPT_TYPE_DOUBLE, {.dbl = 0.0},  0.0,   1.0, FLAGS },
    { "n", "set channel coupling",             OFFSET(channels_coupled),  AV_OPT_TYPE_BOOL,   {.i64 = 1},      0,     1, FLAGS },
    { "c", "set DC correction",                OFFSET(dc_correction),     AV_OPT_TYPE_BOOL,   {.i64 = 0},      0,     1, FLAGS },
    { "b", "set alternative boundary mode",    OFFSET(alt_boundary_mode), AV_OPT_TYPE_BOOL,   {.i64 = 0},      0,     1, FLAGS },
    { "s", "set the compress factor",          OFFSET(compress_factor),   AV_OPT_TYPE_DOUBLE, {.dbl = 0.0},  0.0,  30.0, FLAGS },
    { NULL }
};

AVFILTER_DEFINE_CLASS(dynaudnorm);

static av_cold int init(AVFilterContext *ctx)
{
    DynamicAudioNormalizerContext *s = ctx->priv;

    if (!(s->filter_size & 1)) {
        av_log(ctx, AV_LOG_ERROR, "filter size %d is invalid. Must be an odd value.\n", s->filter_size);
        return AVERROR(EINVAL);
    }

    return 0;
}

static int query_formats(AVFilterContext *ctx)
{
    AVFilterFormats *formats;
    AVFilterChannelLayouts *layouts;
    static const enum AVSampleFormat sample_fmts[] = {
        AV_SAMPLE_FMT_DBLP,
        AV_SAMPLE_FMT_NONE
    };
    int ret;

    layouts = ff_all_channel_counts();
    if (!layouts)
        return AVERROR(ENOMEM);
    ret = ff_set_common_channel_layouts(ctx, layouts);
    if (ret < 0)
        return ret;

    formats = ff_make_format_list(sample_fmts);
    if (!formats)
        return AVERROR(ENOMEM);
    ret = ff_set_common_formats(ctx, formats);
    if (ret < 0)
        return ret;

    formats = ff_all_samplerates();
    if (!formats)
        return AVERROR(ENOMEM);
    return ff_set_common_samplerates(ctx, formats);
}

static inline int frame_size(int sample_rate, int frame_len_msec)
{
    const int frame_size = round((double)sample_rate * (frame_len_msec / 1000.0));
    return frame_size + (frame_size % 2);
}

static void precalculate_fade_factors(double *fade_factors[2], int frame_len)
{
    const double step_size = 1.0 / frame_len;
    int pos;

    for (pos = 0; pos < frame_len; pos++) {
        fade_factors[0][pos] = 1.0 - (step_size * (pos + 1.0));
        fade_factors[1][pos] = 1.0 - fade_factors[0][pos];
    }
}

static cqueue *cqueue_create(int size)
{
    cqueue *q;

    q = av_malloc(sizeof(cqueue));
    if (!q)
        return NULL;

    q->size = size;
    q->nb_elements = 0;
    q->first = 0;

    q->elements = av_malloc(sizeof(double) * size);
    if (!q->elements) {
        av_free(q);
        return NULL;
    }

    return q;
}

static void cqueue_free(cqueue *q)
{
    av_free(q->elements);
    av_free(q);
}

static int cqueue_size(cqueue *q)
{
    return q->nb_elements;
}

static int cqueue_empty(cqueue *q)
{
    return !q->nb_elements;
}

static int cqueue_enqueue(cqueue *q, double element)
{
    int i;

    av_assert2(q->nb_elements != q->size);

    i = (q->first + q->nb_elements) % q->size;
    q->elements[i] = element;
    q->nb_elements++;

    return 0;
}

static double cqueue_peek(cqueue *q, int index)
{
    av_assert2(index < q->nb_elements);
    return q->elements[(q->first + index) % q->size];
}

static int cqueue_dequeue(cqueue *q, double *element)
{
    av_assert2(!cqueue_empty(q));

    *element = q->elements[q->first];
    q->first = (q->first + 1) % q->size;
    q->nb_elements--;

    return 0;
}

static int cqueue_pop(cqueue *q)
{
    av_assert2(!cqueue_empty(q));

    q->first = (q->first + 1) % q->size;
    q->nb_elements--;

    return 0;
}

static const double s_pi = 3.1415926535897932384626433832795028841971693993751058209749445923078164062862089986280348253421170679;

static void init_gaussian_filter(DynamicAudioNormalizerContext *s)
{
    double total_weight = 0.0;
    const double sigma = (((s->filter_size / 2.0) - 1.0) / 3.0) + (1.0 / 3.0);
    double adjust;
    int i;

    // Pre-compute constants
    const int offset = s->filter_size / 2;
    const double c1 = 1.0 / (sigma * sqrt(2.0 * s_pi));
    const double c2 = 2.0 * pow(sigma, 2.0);

    // Compute weights
    for (i = 0; i < s->filter_size; i++) {
        const int x = i - offset;

        s->weights[i] = c1 * exp(-(pow(x, 2.0) / c2));
        total_weight += s->weights[i];
    }

    // Adjust weights
    adjust = 1.0 / total_weight;
    for (i = 0; i < s->filter_size; i++) {
        s->weights[i] *= adjust;
    }
}

static int config_input(AVFilterLink *inlink)
{
    AVFilterContext *ctx = inlink->dst;
    DynamicAudioNormalizerContext *s = ctx->priv;
    int c;

    s->frame_len =
    inlink->min_samples =
    inlink->max_samples =
    inlink->partial_buf_size = frame_size(inlink->sample_rate, s->frame_len_msec);
    av_log(ctx, AV_LOG_DEBUG, "frame len %d\n", s->frame_len);

    s->fade_factors[0] = av_malloc(s->frame_len * sizeof(*s->fade_factors[0]));
    s->fade_factors[1] = av_malloc(s->frame_len * sizeof(*s->fade_factors[1]));

    s->prev_amplification_factor = av_malloc(inlink->channels * sizeof(*s->prev_amplification_factor));
    s->dc_correction_value = av_calloc(inlink->channels, sizeof(*s->dc_correction_value));
    s->compress_threshold = av_calloc(inlink->channels, sizeof(*s->compress_threshold));
    s->gain_history_original = av_calloc(inlink->channels, sizeof(*s->gain_history_original));
    s->gain_history_minimum = av_calloc(inlink->channels, sizeof(*s->gain_history_minimum));
    s->gain_history_smoothed = av_calloc(inlink->channels, sizeof(*s->gain_history_smoothed));
    s->weights = av_malloc(s->filter_size * sizeof(*s->weights));
    if (!s->prev_amplification_factor || !s->dc_correction_value ||
        !s->compress_threshold || !s->fade_factors[0] || !s->fade_factors[1] ||
        !s->gain_history_original || !s->gain_history_minimum ||
        !s->gain_history_smoothed || !s->weights)
        return AVERROR(ENOMEM);

    for (c = 0; c < inlink->channels; c++) {
        s->prev_amplification_factor[c] = 1.0;

        s->gain_history_original[c] = cqueue_create(s->filter_size);
        s->gain_history_minimum[c]  = cqueue_create(s->filter_size);
        s->gain_history_smoothed[c] = cqueue_create(s->filter_size);

        if (!s->gain_history_original[c] || !s->gain_history_minimum[c] ||
            !s->gain_history_smoothed[c])
            return AVERROR(ENOMEM);
    }

    precalculate_fade_factors(s->fade_factors, s->frame_len);
    init_gaussian_filter(s);

    s->channels = inlink->channels;
    s->delay = s->filter_size;

    return 0;
}

static int config_output(AVFilterLink *outlink)
{
    outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
    return 0;
}

static inline double fade(double prev, double next, int pos,
                          double *fade_factors[2])
{
    return fade_factors[0][pos] * prev + fade_factors[1][pos] * next;
}

static inline double pow2(const double value)
{
    return value * value;
}

static inline double bound(const double threshold, const double val)
{
    const double CONST = 0.8862269254527580136490837416705725913987747280611935; //sqrt(PI) / 2.0
    return erf(CONST * (val / threshold)) * threshold;
}

static double find_peak_magnitude(AVFrame *frame, int channel)
{
    double max = DBL_EPSILON;
    int c, i;

    if (channel == -1) {
        for (c = 0; c < av_frame_get_channels(frame); c++) {
            double *data_ptr = (double *)frame->extended_data[c];

            for (i = 0; i < frame->nb_samples; i++)
                max = FFMAX(max, fabs(data_ptr[i]));
        }
    } else {
        double *data_ptr = (double *)frame->extended_data[channel];

        for (i = 0; i < frame->nb_samples; i++)
            max = FFMAX(max, fabs(data_ptr[i]));
    }

    return max;
}

static double compute_frame_rms(AVFrame *frame, int channel)
{
    double rms_value = 0.0;
    int c, i;

    if (channel == -1) {
        for (c = 0; c < av_frame_get_channels(frame); c++) {
            const double *data_ptr = (double *)frame->extended_data[c];

            for (i = 0; i < frame->nb_samples; i++) {
                rms_value += pow2(data_ptr[i]);
            }
        }

        rms_value /= frame->nb_samples * av_frame_get_channels(frame);
    } else {
        const double *data_ptr = (double *)frame->extended_data[channel];
        for (i = 0; i < frame->nb_samples; i++) {
            rms_value += pow2(data_ptr[i]);
        }

        rms_value /= frame->nb_samples;
    }

    return FFMAX(sqrt(rms_value), DBL_EPSILON);
}

static double get_max_local_gain(DynamicAudioNormalizerContext *s, AVFrame *frame,
                                 int channel)
{
    const double maximum_gain = s->peak_value / find_peak_magnitude(frame, channel);
    const double rms_gain = s->target_rms > DBL_EPSILON ? (s->target_rms / compute_frame_rms(frame, channel)) : DBL_MAX;
    return bound(s->max_amplification, FFMIN(maximum_gain, rms_gain));
}

static double minimum_filter(cqueue *q)
{
    double min = DBL_MAX;
    int i;

    for (i = 0; i < cqueue_size(q); i++) {
        min = FFMIN(min, cqueue_peek(q, i));
    }

    return min;
}

static double gaussian_filter(DynamicAudioNormalizerContext *s, cqueue *q)
{
    double result = 0.0;
    int i;

    for (i = 0; i < cqueue_size(q); i++) {
        result += cqueue_peek(q, i) * s->weights[i];
    }

    return result;
}

static void update_gain_history(DynamicAudioNormalizerContext *s, int channel,
                                double current_gain_factor)
{
    if (cqueue_empty(s->gain_history_original[channel]) ||
        cqueue_empty(s->gain_history_minimum[channel])) {
        const int pre_fill_size = s->filter_size / 2;

        s->prev_amplification_factor[channel] = s->alt_boundary_mode ? current_gain_factor : 1.0;

        while (cqueue_size(s->gain_history_original[channel]) < pre_fill_size) {
            cqueue_enqueue(s->gain_history_original[channel], s->alt_boundary_mode ? current_gain_factor : 1.0);
        }

        while (cqueue_size(s->gain_history_minimum[channel]) < pre_fill_size) {
            cqueue_enqueue(s->gain_history_minimum[channel], s->alt_boundary_mode ? current_gain_factor : 1.0);
        }
    }

    cqueue_enqueue(s->gain_history_original[channel], current_gain_factor);

    while (cqueue_size(s->gain_history_original[channel]) >= s->filter_size) {
        double minimum;
        av_assert0(cqueue_size(s->gain_history_original[channel]) == s->filter_size);
        minimum = minimum_filter(s->gain_history_original[channel]);

        cqueue_enqueue(s->gain_history_minimum[channel], minimum);

        cqueue_pop(s->gain_history_original[channel]);
    }

    while (cqueue_size(s->gain_history_minimum[channel]) >= s->filter_size) {
        double smoothed;
        av_assert0(cqueue_size(s->gain_history_minimum[channel]) == s->filter_size);
        smoothed = gaussian_filter(s, s->gain_history_minimum[channel]);

        cqueue_enqueue(s->gain_history_smoothed[channel], smoothed);

        cqueue_pop(s->gain_history_minimum[channel]);
    }
}

static inline double update_value(double new, double old, double aggressiveness)
{
    av_assert0((aggressiveness >= 0.0) && (aggressiveness <= 1.0));
    return aggressiveness * new + (1.0 - aggressiveness) * old;
}

static void perform_dc_correction(DynamicAudioNormalizerContext *s, AVFrame *frame)
{
    const double diff = 1.0 / frame->nb_samples;
    int is_first_frame = cqueue_empty(s->gain_history_original[0]);
    int c, i;

    for (c = 0; c < s->channels; c++) {
        double *dst_ptr = (double *)frame->extended_data[c];
        double current_average_value = 0.0;
        double prev_value;

        for (i = 0; i < frame->nb_samples; i++)
            current_average_value += dst_ptr[i] * diff;

        prev_value = is_first_frame ? current_average_value : s->dc_correction_value[c];
        s->dc_correction_value[c] = is_first_frame ? current_average_value : update_value(current_average_value, s->dc_correction_value[c], 0.1);

        for (i = 0; i < frame->nb_samples; i++) {
            dst_ptr[i] -= fade(prev_value, s->dc_correction_value[c], i, s->fade_factors);
        }
    }
}

static double setup_compress_thresh(double threshold)
{
    if ((threshold > DBL_EPSILON) && (threshold < (1.0 - DBL_EPSILON))) {
        double current_threshold = threshold;
        double step_size = 1.0;

        while (step_size > DBL_EPSILON) {
            while ((current_threshold + step_size > current_threshold) &&
                   (bound(current_threshold + step_size, 1.0) <= threshold)) {
                current_threshold += step_size;
            }

            step_size /= 2.0;
        }

        return current_threshold;
    } else {
        return threshold;
    }
}

static double compute_frame_std_dev(DynamicAudioNormalizerContext *s,
                                    AVFrame *frame, int channel)
{
    double variance = 0.0;
    int i, c;

    if (channel == -1) {
        for (c = 0; c < s->channels; c++) {
            const double *data_ptr = (double *)frame->extended_data[c];

            for (i = 0; i < frame->nb_samples; i++) {
                variance += pow2(data_ptr[i]);  // Assume that MEAN is *zero*
            }
        }
        variance /= (s->channels * frame->nb_samples) - 1;
    } else {
        const double *data_ptr = (double *)frame->extended_data[channel];

        for (i = 0; i < frame->nb_samples; i++) {
            variance += pow2(data_ptr[i]);      // Assume that MEAN is *zero*
        }
        variance /= frame->nb_samples - 1;
    }

    return FFMAX(sqrt(variance), DBL_EPSILON);
}

static void perform_compression(DynamicAudioNormalizerContext *s, AVFrame *frame)
{
    int is_first_frame = cqueue_empty(s->gain_history_original[0]);
    int c, i;

    if (s->channels_coupled) {
        const double standard_deviation = compute_frame_std_dev(s, frame, -1);
        const double current_threshold  = FFMIN(1.0, s->compress_factor * standard_deviation);

        const double prev_value = is_first_frame ? current_threshold : s->compress_threshold[0];
        double prev_actual_thresh, curr_actual_thresh;
        s->compress_threshold[0] = is_first_frame ? current_threshold : update_value(current_threshold, s->compress_threshold[0], (1.0/3.0));

        prev_actual_thresh = setup_compress_thresh(prev_value);
        curr_actual_thresh = setup_compress_thresh(s->compress_threshold[0]);

        for (c = 0; c < s->channels; c++) {
            double *const dst_ptr = (double *)frame->extended_data[c];
            for (i = 0; i < frame->nb_samples; i++) {
                const double localThresh = fade(prev_actual_thresh, curr_actual_thresh, i, s->fade_factors);
                dst_ptr[i] = copysign(bound(localThresh, fabs(dst_ptr[i])), dst_ptr[i]);
            }
        }
    } else {
        for (c = 0; c < s->channels; c++) {
            const double standard_deviation = compute_frame_std_dev(s, frame, c);
            const double current_threshold  = setup_compress_thresh(FFMIN(1.0, s->compress_factor * standard_deviation));

            const double prev_value = is_first_frame ? current_threshold : s->compress_threshold[c];
            double prev_actual_thresh, curr_actual_thresh;
            double *dst_ptr;
            s->compress_threshold[c] = is_first_frame ? current_threshold : update_value(current_threshold, s->compress_threshold[c], 1.0/3.0);

            prev_actual_thresh = setup_compress_thresh(prev_value);
            curr_actual_thresh = setup_compress_thresh(s->compress_threshold[c]);

            dst_ptr = (double *)frame->extended_data[c];
            for (i = 0; i < frame->nb_samples; i++) {
                const double localThresh = fade(prev_actual_thresh, curr_actual_thresh, i, s->fade_factors);
                dst_ptr[i] = copysign(bound(localThresh, fabs(dst_ptr[i])), dst_ptr[i]);
            }
        }
    }
}

static void analyze_frame(DynamicAudioNormalizerContext *s, AVFrame *frame)
{
    if (s->dc_correction) {
        perform_dc_correction(s, frame);
    }

    if (s->compress_factor > DBL_EPSILON) {
        perform_compression(s, frame);
    }

    if (s->channels_coupled) {
        const double current_gain_factor = get_max_local_gain(s, frame, -1);
        int c;

        for (c = 0; c < s->channels; c++)
            update_gain_history(s, c, current_gain_factor);
    } else {
        int c;

        for (c = 0; c < s->channels; c++)
            update_gain_history(s, c, get_max_local_gain(s, frame, c));
    }
}

static void amplify_frame(DynamicAudioNormalizerContext *s, AVFrame *frame)
{
    int c, i;

    for (c = 0; c < s->channels; c++) {
        double *dst_ptr = (double *)frame->extended_data[c];
        double current_amplification_factor;

        cqueue_dequeue(s->gain_history_smoothed[c], &current_amplification_factor);

        for (i = 0; i < frame->nb_samples; i++) {
            const double amplification_factor = fade(s->prev_amplification_factor[c],
                                                     current_amplification_factor, i,
                                                     s->fade_factors);

            dst_ptr[i] *= amplification_factor;

            if (fabs(dst_ptr[i]) > s->peak_value)
                dst_ptr[i] = copysign(s->peak_value, dst_ptr[i]);
        }

        s->prev_amplification_factor[c] = current_amplification_factor;
    }
}

static int filter_frame(AVFilterLink *inlink, AVFrame *in)
{
    AVFilterContext *ctx = inlink->dst;
    DynamicAudioNormalizerContext *s = ctx->priv;
    AVFilterLink *outlink = inlink->dst->outputs[0];
    int ret = 0;

    if (!cqueue_empty(s->gain_history_smoothed[0])) {
        AVFrame *out = ff_bufqueue_get(&s->queue);

        amplify_frame(s, out);
        ret = ff_filter_frame(outlink, out);
    }

    analyze_frame(s, in);
    ff_bufqueue_add(ctx, &s->queue, in);

    return ret;
}

static int flush_buffer(DynamicAudioNormalizerContext *s, AVFilterLink *inlink,
                        AVFilterLink *outlink)
{
    AVFrame *out = ff_get_audio_buffer(outlink, s->frame_len);
    int c, i;

    if (!out)
        return AVERROR(ENOMEM);

    for (c = 0; c < s->channels; c++) {
        double *dst_ptr = (double *)out->extended_data[c];

        for (i = 0; i < out->nb_samples; i++) {
            dst_ptr[i] = s->alt_boundary_mode ? DBL_EPSILON : ((s->target_rms > DBL_EPSILON) ? FFMIN(s->peak_value, s->target_rms) : s->peak_value);
            if (s->dc_correction) {
                dst_ptr[i] *= ((i % 2) == 1) ? -1 : 1;
                dst_ptr[i] += s->dc_correction_value[c];
            }
        }
    }

    s->delay--;
    return filter_frame(inlink, out);
}

static int request_frame(AVFilterLink *outlink)
{
    AVFilterContext *ctx = outlink->src;
    DynamicAudioNormalizerContext *s = ctx->priv;
    int ret = 0;

    ret = ff_request_frame(ctx->inputs[0]);

    if (ret == AVERROR_EOF && !ctx->is_disabled && s->delay)
        ret = flush_buffer(s, ctx->inputs[0], outlink);

    return ret;
}

static av_cold void uninit(AVFilterContext *ctx)
{
    DynamicAudioNormalizerContext *s = ctx->priv;
    int c;

    av_freep(&s->prev_amplification_factor);
    av_freep(&s->dc_correction_value);
    av_freep(&s->compress_threshold);
    av_freep(&s->fade_factors[0]);
    av_freep(&s->fade_factors[1]);

    for (c = 0; c < s->channels; c++) {
        cqueue_free(s->gain_history_original[c]);
        cqueue_free(s->gain_history_minimum[c]);
        cqueue_free(s->gain_history_smoothed[c]);
    }

    av_freep(&s->gain_history_original);
    av_freep(&s->gain_history_minimum);
    av_freep(&s->gain_history_smoothed);

    av_freep(&s->weights);

    ff_bufqueue_discard_all(&s->queue);
}

static const AVFilterPad avfilter_af_dynaudnorm_inputs[] = {
    {
        .name           = "default",
        .type           = AVMEDIA_TYPE_AUDIO,
        .filter_frame   = filter_frame,
        .config_props   = config_input,
        .needs_writable = 1,
    },
    { NULL }
};

static const AVFilterPad avfilter_af_dynaudnorm_outputs[] = {
    {
        .name          = "default",
        .type          = AVMEDIA_TYPE_AUDIO,
        .config_props  = config_output,
        .request_frame = request_frame,
    },
    { NULL }
};

AVFilter ff_af_dynaudnorm = {
    .name          = "dynaudnorm",
    .description   = NULL_IF_CONFIG_SMALL("Dynamic Audio Normalizer."),
    .query_formats = query_formats,
    .priv_size     = sizeof(DynamicAudioNormalizerContext),
    .init          = init,
    .uninit        = uninit,
    .inputs        = avfilter_af_dynaudnorm_inputs,
    .outputs       = avfilter_af_dynaudnorm_outputs,
    .priv_class    = &dynaudnorm_class,
};