上海昆剧团 40年春色正浓(传统文化，创新发展)

Line	Branch	Exec	Source
1			/*
2			* Copyright (c) 2022 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 <float.h>
22
23			#include "libavutil/eval.h"
24			#include "libavutil/ffmath.h"
25			#include "libavutil/mem.h"
26			#include "libavutil/opt.h"
27			#include "libavutil/tx.h"
28			#include "audio.h"
29			#include "avfilter.h"
30			#include "filters.h"
31
32			static const char * const var_names[] = {
33			"ch", ///< the value of the current channel
34			"sn", ///< number of samples
35			"nb_channels",
36			"t", ///< timestamp expressed in seconds
37			"sr", ///< sample rate
38			"p", ///< input power in dB for frequency bin
39			"f", ///< frequency in Hz
40			NULL
41			};
42
43			enum var_name {
44			VAR_CH,
45			VAR_SN,
46			VAR_NB_CHANNELS,
47			VAR_T,
48			VAR_SR,
49			VAR_P,
50			VAR_F,
51			VAR_VARS_NB
52			};
53
54			typedef struct AudioDRCContext {
55			const AVClass *class;
56
57			double attack_ms;
58			double release_ms;
59			char *expr_str;
60
61			double attack;
62			double release;
63
64			int fft_size;
65			int overlap;
66			int channels;
67
68			float fx;
69			float *window;
70
71			AVFrame *drc_frame;
72			AVFrame *energy;
73			AVFrame *envelope;
74			AVFrame *factors;
75			AVFrame *in;
76			AVFrame *in_buffer;
77			AVFrame *in_frame;
78			AVFrame *out_dist_frame;
79			AVFrame *spectrum_buf;
80			AVFrame *target_gain;
81			AVFrame *windowed_frame;
82
83			char *channels_to_filter;
84			AVChannelLayout ch_layout;
85
86			AVTXContext **tx_ctx;
87			av_tx_fn tx_fn;
88			AVTXContext **itx_ctx;
89			av_tx_fn itx_fn;
90
91			AVExpr *expr;
92			double var_values[VAR_VARS_NB];
93			} AudioDRCContext;
94
95			#define OFFSET(x) offsetof(AudioDRCContext, x)
96			#define FLAGS AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_FILTERING_PARAM | AV_OPT_FLAG_RUNTIME_PARAM
97
98			static const AVOption adrc_options[] = {
99			{ "transfer", "set the transfer expression", OFFSET(expr_str), AV_OPT_TYPE_STRING, {.str="p"}, 0, 0, FLAGS },
100			{ "attack", "set the attack", OFFSET(attack_ms), AV_OPT_TYPE_DOUBLE, {.dbl=50.}, 1, 1000, FLAGS },
101			{ "release", "set the release", OFFSET(release_ms), AV_OPT_TYPE_DOUBLE, {.dbl=100.}, 5, 2000, FLAGS },
102			{ "channels", "set channels to filter",OFFSET(channels_to_filter),AV_OPT_TYPE_STRING,{.str="all"},0, 0, FLAGS },
103			{NULL}
104			};
105
106			AVFILTER_DEFINE_CLASS(adrc);
107
108		✗	static void generate_hann_window(float *window, int size)
109			{
110		✗	for (int i = 0; i < size; i++) {
111		✗	float value = 0.5f * (1.f - cosf(2.f * M_PI * i / size));
112
113		✗	window[i] = value;
114			}
115		✗	}
116
117		✗	static int config_input(AVFilterLink *inlink)
118			{
119		✗	AVFilterContext *ctx = inlink->dst;
120		✗	AudioDRCContext *s = ctx->priv;
121			float scale;
122			int ret;
123
124		✗	s->fft_size = inlink->sample_rate > 100000 ? 1024 : inlink->sample_rate > 50000 ? 512 : 256;
125		✗	s->fx = inlink->sample_rate * 0.5f / (s->fft_size / 2 + 1);
126		✗	s->overlap = s->fft_size / 4;
127
128		✗	s->window = av_calloc(s->fft_size, sizeof(*s->window));
129		✗	if (!s->window)
130		✗	return AVERROR(ENOMEM);
131
132		✗	s->drc_frame = ff_get_audio_buffer(inlink, s->fft_size * 2);
133		✗	s->energy = ff_get_audio_buffer(inlink, s->fft_size / 2 + 1);
134		✗	s->envelope = ff_get_audio_buffer(inlink, s->fft_size / 2 + 1);
135		✗	s->factors = ff_get_audio_buffer(inlink, s->fft_size / 2 + 1);
136		✗	s->in_buffer = ff_get_audio_buffer(inlink, s->fft_size * 2);
137		✗	s->in_frame = ff_get_audio_buffer(inlink, s->fft_size * 2);
138		✗	s->out_dist_frame = ff_get_audio_buffer(inlink, s->fft_size * 2);
139		✗	s->spectrum_buf = ff_get_audio_buffer(inlink, s->fft_size * 2);
140		✗	s->target_gain = ff_get_audio_buffer(inlink, s->fft_size / 2 + 1);
141		✗	s->windowed_frame = ff_get_audio_buffer(inlink, s->fft_size * 2);
142		✗	if (!s->in_buffer || !s->in_frame || !s->target_gain ||
143		✗	!s->out_dist_frame || !s->windowed_frame || !s->envelope ||
144		✗	!s->drc_frame || !s->spectrum_buf || !s->energy || !s->factors)
145		✗	return AVERROR(ENOMEM);
146
147		✗	generate_hann_window(s->window, s->fft_size);
148
149		✗	s->channels = inlink->ch_layout.nb_channels;
150
151		✗	s->tx_ctx = av_calloc(s->channels, sizeof(*s->tx_ctx));
152		✗	s->itx_ctx = av_calloc(s->channels, sizeof(*s->itx_ctx));
153		✗	if (!s->tx_ctx || !s->itx_ctx)
154		✗	return AVERROR(ENOMEM);
155
156		✗	for (int ch = 0; ch < s->channels; ch++) {
157		✗	scale = 1.f / s->fft_size;
158		✗	ret = av_tx_init(&s->tx_ctx[ch], &s->tx_fn, AV_TX_FLOAT_RDFT, 0, s->fft_size, &scale, 0);
159		✗	if (ret < 0)
160		✗	return ret;
161
162		✗	scale = 1.f;
163		✗	ret = av_tx_init(&s->itx_ctx[ch], &s->itx_fn, AV_TX_FLOAT_RDFT, 1, s->fft_size, &scale, 0);
164		✗	if (ret < 0)
165		✗	return ret;
166			}
167
168		✗	s->var_values[VAR_SR] = inlink->sample_rate;
169		✗	s->var_values[VAR_NB_CHANNELS] = s->channels;
170
171		✗	return av_expr_parse(&s->expr, s->expr_str, var_names, NULL, NULL,
172			NULL, NULL, 0, ctx);
173			}
174
175		✗	static void apply_window(AudioDRCContext *s,
176			const float *in_frame, float *out_frame, const int add_to_out_frame)
177			{
178		✗	const float *window = s->window;
179		✗	const int fft_size = s->fft_size;
180
181		✗	if (add_to_out_frame) {
182		✗	for (int i = 0; i < fft_size; i++)
183		✗	out_frame[i] += in_frame[i] * window[i];
184			} else {
185		✗	for (int i = 0; i < fft_size; i++)
186		✗	out_frame[i] = in_frame[i] * window[i];
187			}
188		✗	}
189
190		✗	static float sqrf(float x)
191			{
192		✗	return x * x;
193			}
194
195		✗	static void get_energy(AVFilterContext *ctx,
196			int len,
197			float *energy,
198			const float *spectral)
199			{
200		✗	for (int n = 0; n < len; n++) {
201		✗	energy[n] = 10.f * log10f(sqrf(spectral[2 * n]) + sqrf(spectral[2 * n + 1]));
202		✗	if (!isnormal(energy[n]))
203		✗	energy[n] = -351.f;
204			}
205		✗	}
206
207		✗	static void get_target_gain(AVFilterContext *ctx,
208			int len,
209			float *gain,
210			const float *energy,
211			double *var_values,
212			float fx, int bypass)
213			{
214		✗	AudioDRCContext *s = ctx->priv;
215
216		✗	if (bypass) {
217		✗	memcpy(gain, energy, sizeof(*gain) * len);
218		✗	return;
219			}
220
221		✗	for (int n = 0; n < len; n++) {
222		✗	const float Xg = energy[n];
223
224		✗	var_values[VAR_P] = Xg;
225		✗	var_values[VAR_F] = n * fx;
226
227		✗	gain[n] = av_expr_eval(s->expr, var_values, s);
228			}
229			}
230
231		✗	static void get_envelope(AVFilterContext *ctx,
232			int len,
233			float *envelope,
234			const float *energy,
235			const float *gain)
236			{
237		✗	AudioDRCContext *s = ctx->priv;
238		✗	const float release = s->release;
239		✗	const float attack = s->attack;
240
241		✗	for (int n = 0; n < len; n++) {
242		✗	const float Bg = gain[n] - energy[n];
243		✗	const float Vg = envelope[n];
244
245		✗	if (Bg > Vg) {
246		✗	envelope[n] = attack * Vg + (1.f - attack) * Bg;
247		✗	} else if (Bg <= Vg) {
248		✗	envelope[n] = release * Vg + (1.f - release) * Bg;
249			} else {
250		✗	envelope[n] = 0.f;
251			}
252			}
253		✗	}
254
255		✗	static void get_factors(AVFilterContext *ctx,
256			int len,
257			float *factors,
258			const float *envelope)
259			{
260		✗	for (int n = 0; n < len; n++)
261		✗	factors[n] = sqrtf(ff_exp10f(envelope[n] / 10.f));
262		✗	}
263
264		✗	static void apply_factors(AVFilterContext *ctx,
265			int len,
266			float *spectrum,
267			const float *factors)
268			{
269		✗	for (int n = 0; n < len; n++) {
270		✗	spectrum[2*n+0] *= factors[n];
271		✗	spectrum[2*n+1] *= factors[n];
272			}
273		✗	}
274
275		✗	static void feed(AVFilterContext *ctx, int ch,
276			const float *in_samples, float *out_samples,
277			float *in_frame, float *out_dist_frame,
278			float *windowed_frame, float *drc_frame,
279			float *spectrum_buf, float *energy,
280			float *target_gain, float *envelope,
281			float *factors)
282			{
283		✗	AudioDRCContext *s = ctx->priv;
284			double var_values[VAR_VARS_NB];
285		✗	const int fft_size = s->fft_size;
286		✗	const int nb_coeffs = s->fft_size / 2 + 1;
287		✗	const int overlap = s->overlap;
288		✗	enum AVChannel channel = av_channel_layout_channel_from_index(&ctx->inputs[0]->ch_layout, ch);
289		✗	const int bypass = av_channel_layout_index_from_channel(&s->ch_layout, channel) < 0;
290
291		✗	memcpy(var_values, s->var_values, sizeof(var_values));
292
293		✗	var_values[VAR_CH] = ch;
294
295			// shift in/out buffers
296		✗	memmove(in_frame, in_frame + overlap, (fft_size - overlap) * sizeof(*in_frame));
297		✗	memmove(out_dist_frame, out_dist_frame + overlap, (fft_size - overlap) * sizeof(*out_dist_frame));
298
299		✗	memcpy(in_frame + fft_size - overlap, in_samples, sizeof(*in_frame) * overlap);
300		✗	memset(out_dist_frame + fft_size - overlap, 0, sizeof(*out_dist_frame) * overlap);
301
302		✗	apply_window(s, in_frame, windowed_frame, 0);
303		✗	s->tx_fn(s->tx_ctx[ch], spectrum_buf, windowed_frame, sizeof(float));
304
305		✗	get_energy(ctx, nb_coeffs, energy, spectrum_buf);
306		✗	get_target_gain(ctx, nb_coeffs, target_gain, energy, var_values, s->fx, bypass);
307		✗	get_envelope(ctx, nb_coeffs, envelope, energy, target_gain);
308		✗	get_factors(ctx, nb_coeffs, factors, envelope);
309		✗	apply_factors(ctx, nb_coeffs, spectrum_buf, factors);
310
311		✗	s->itx_fn(s->itx_ctx[ch], drc_frame, spectrum_buf, sizeof(AVComplexFloat));
312
313		✗	apply_window(s, drc_frame, out_dist_frame, 1);
314
315			// 4 times overlap with squared hanning window results in 1.5 time increase in amplitude
316		✗	if (!ctx->is_disabled) {
317		✗	for (int i = 0; i < overlap; i++)
318		✗	out_samples[i] = out_dist_frame[i] / 1.5f;
319			} else {
320		✗	memcpy(out_samples, in_frame, sizeof(*out_samples) * overlap);
321			}
322		✗	}
323
324		✗	static int drc_channel(AVFilterContext *ctx, AVFrame *in, AVFrame *out, int ch)
325			{
326		✗	AudioDRCContext *s = ctx->priv;
327		✗	const float *src = (const float *)in->extended_data[ch];
328		✗	float *in_buffer = (float *)s->in_buffer->extended_data[ch];
329		✗	float *dst = (float *)out->extended_data[ch];
330
331		✗	memcpy(in_buffer, src, sizeof(*in_buffer) * s->overlap);
332
333		✗	feed(ctx, ch, in_buffer, dst,
334		✗	(float *)(s->in_frame->extended_data[ch]),
335		✗	(float *)(s->out_dist_frame->extended_data[ch]),
336		✗	(float *)(s->windowed_frame->extended_data[ch]),
337		✗	(float *)(s->drc_frame->extended_data[ch]),
338		✗	(float *)(s->spectrum_buf->extended_data[ch]),
339		✗	(float *)(s->energy->extended_data[ch]),
340		✗	(float *)(s->target_gain->extended_data[ch]),
341		✗	(float *)(s->envelope->extended_data[ch]),
342		✗	(float *)(s->factors->extended_data[ch]));
343
344		✗	return 0;
345			}
346
347		✗	static int drc_channels(AVFilterContext *ctx, void *arg, int jobnr, int nb_jobs)
348			{
349		✗	AudioDRCContext *s = ctx->priv;
350		✗	AVFrame *in = s->in;
351		✗	AVFrame *out = arg;
352		✗	const int start = (out->ch_layout.nb_channels * jobnr) / nb_jobs;
353		✗	const int end = (out->ch_layout.nb_channels * (jobnr+1)) / nb_jobs;
354
355		✗	for (int ch = start; ch < end; ch++)
356		✗	drc_channel(ctx, in, out, ch);
357
358		✗	return 0;
359			}
360
361		✗	static int filter_frame(AVFilterLink *inlink, AVFrame *in)
362			{
363		✗	AVFilterContext *ctx = inlink->dst;
364		✗	AVFilterLink *outlink = ctx->outputs[0];
365		✗	FilterLink *outl = ff_filter_link(outlink);
366		✗	AudioDRCContext *s = ctx->priv;
367			AVFrame *out;
368			int ret;
369
370		✗	out = ff_get_audio_buffer(outlink, s->overlap);
371		✗	if (!out) {
372		✗	ret = AVERROR(ENOMEM);
373		✗	goto fail;
374			}
375
376		✗	s->var_values[VAR_SN] = outl->sample_count_in;
377		✗	s->var_values[VAR_T] = s->var_values[VAR_SN] * (double)1/outlink->sample_rate;
378
379		✗	s->in = in;
380		✗	av_frame_copy_props(out, in);
381		✗	ff_filter_execute(ctx, drc_channels, out, NULL,
382		✗	FFMIN(outlink->ch_layout.nb_channels, ff_filter_get_nb_threads(ctx)));
383
384		✗	out->pts = in->pts;
385		✗	out->nb_samples = in->nb_samples;
386		✗	ret = ff_filter_frame(outlink, out);
387		✗	fail:
388		✗	av_frame_free(&in);
389		✗	s->in = NULL;
390		✗	return ret < 0 ? ret : 0;
391			}
392
393		✗	static int activate(AVFilterContext *ctx)
394			{
395		✗	AVFilterLink *inlink = ctx->inputs[0];
396		✗	AVFilterLink *outlink = ctx->outputs[0];
397		✗	AudioDRCContext *s = ctx->priv;
398		✗	AVFrame *in = NULL;
399		✗	int ret = 0, status;
400			int64_t pts;
401
402		✗	ret = av_channel_layout_copy(&s->ch_layout, &inlink->ch_layout);
403		✗	if (ret < 0)
404		✗	return ret;
405		✗	if (strcmp(s->channels_to_filter, "all"))
406		✗	av_channel_layout_from_string(&s->ch_layout, s->channels_to_filter);
407
408		✗	FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink);
409
410		✗	ret = ff_inlink_consume_samples(inlink, s->overlap, s->overlap, &in);
411		✗	if (ret < 0)
412		✗	return ret;
413
414		✗	if (ret > 0) {
415		✗	s->attack = expf(-1.f / (s->attack_ms * inlink->sample_rate / 1000.f));
416		✗	s->release = expf(-1.f / (s->release_ms * inlink->sample_rate / 1000.f));
417
418		✗	return filter_frame(inlink, in);
419		✗	} else if (ff_inlink_acknowledge_status(inlink, &status, &pts)) {
420		✗	ff_outlink_set_status(outlink, status, pts);
421		✗	return 0;
422			} else {
423		✗	if (ff_inlink_queued_samples(inlink) >= s->overlap) {
424		✗	ff_filter_set_ready(ctx, 10);
425		✗	} else if (ff_outlink_frame_wanted(outlink)) {
426		✗	ff_inlink_request_frame(inlink);
427			}
428		✗	return 0;
429			}
430			}
431
432		✗	static av_cold void uninit(AVFilterContext *ctx)
433			{
434		✗	AudioDRCContext *s = ctx->priv;
435
436		✗	av_channel_layout_uninit(&s->ch_layout);
437
438		✗	av_expr_free(s->expr);
439		✗	s->expr = NULL;
440
441		✗	av_freep(&s->window);
442
443		✗	av_frame_free(&s->drc_frame);
444		✗	av_frame_free(&s->energy);
445		✗	av_frame_free(&s->envelope);
446		✗	av_frame_free(&s->factors);
447		✗	av_frame_free(&s->in_buffer);
448		✗	av_frame_free(&s->in_frame);
449		✗	av_frame_free(&s->out_dist_frame);
450		✗	av_frame_free(&s->spectrum_buf);
451		✗	av_frame_free(&s->target_gain);
452		✗	av_frame_free(&s->windowed_frame);
453
454		✗	for (int ch = 0; ch < s->channels; ch++) {
455		✗	if (s->tx_ctx)
456		✗	av_tx_uninit(&s->tx_ctx[ch]);
457		✗	if (s->itx_ctx)
458		✗	av_tx_uninit(&s->itx_ctx[ch]);
459			}
460
461		✗	av_freep(&s->tx_ctx);
462		✗	av_freep(&s->itx_ctx);
463		✗	}
464
465		✗	static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
466			char *res, int res_len, int flags)
467			{
468		✗	AudioDRCContext *s = ctx->priv;
469		✗	char *old_expr_str = av_strdup(s->expr_str);
470			int ret;
471
472		✗	ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
473		✗	if (ret >= 0 && strcmp(old_expr_str, s->expr_str)) {
474		✗	ret = av_expr_parse(&s->expr, s->expr_str, var_names, NULL, NULL,
475			NULL, NULL, 0, ctx);
476			}
477		✗	av_free(old_expr_str);
478		✗	return ret;
479			}
480
481			static const AVFilterPad inputs[] = {
482			{
483			.name = "default",
484			.type = AVMEDIA_TYPE_AUDIO,
485			.config_props = config_input,
486			},
487			};
488
489			const FFFilter ff_af_adrc = {
490			.p.name = "adrc",
491			.p.description = NULL_IF_CONFIG_SMALL("Audio Spectral Dynamic Range Controller."),
492			.p.priv_class = &adrc_class,
493			.p.flags = AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL |
494			AVFILTER_FLAG_SLICE_THREADS,
495			.priv_size = sizeof(AudioDRCContext),
496			.uninit = uninit,
497			FILTER_INPUTS(inputs),
498			FILTER_OUTPUTS(ff_audio_default_filterpad),
499			FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_FLTP),
500			.activate = activate,
501			.process_command = process_command,
502			};
503