Book a Demo!
CoCalc Logo Icon
StoreFeaturesDocsShareSupportNewsAboutPoliciesSign UpSign In
Download
52868 views
1
/*
2
* Copyright (c) 2016 Paul B Mahol
3
*
4
* This file is part of FFmpeg.
5
*
6
* FFmpeg is free software; you can redistribute it and/or modify it
7
* under the terms of the GNU Lesser General Public License as published
8
* by the Free Software Foundation; either version 2.1 of the License,
9
* 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/avstring.h"
23
#include "libavfilter/internal.h"
24
#include "libavutil/common.h"
25
#include "libavutil/opt.h"
26
#include "libavcodec/avfft.h"
27
#include "libavutil/eval.h"
28
#include "audio.h"
29
#include "window_func.h"
30
31
typedef struct AFFTFiltContext {
32
const AVClass *class;
33
char *real_str;
34
char *img_str;
35
int fft_bits;
36
37
FFTContext *fft, *ifft;
38
FFTComplex **fft_data;
39
int nb_exprs;
40
int window_size;
41
AVExpr **real;
42
AVExpr **imag;
43
AVAudioFifo *fifo;
44
int64_t pts;
45
int hop_size;
46
float overlap;
47
AVFrame *buffer;
48
int start, end;
49
int win_func;
50
float win_scale;
51
float *window_func_lut;
52
} AFFTFiltContext;
53
54
static const char *const var_names[] = { "sr", "b", "nb", "ch", "chs", "pts", NULL };
55
enum { VAR_SAMPLE_RATE, VAR_BIN, VAR_NBBINS, VAR_CHANNEL, VAR_CHANNELS, VAR_PTS, VAR_VARS_NB };
56
57
#define OFFSET(x) offsetof(AFFTFiltContext, x)
58
#define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
59
60
static const AVOption afftfilt_options[] = {
61
{ "real", "set channels real expressions", OFFSET(real_str), AV_OPT_TYPE_STRING, {.str = "1" }, 0, 0, A },
62
{ "imag", "set channels imaginary expressions", OFFSET(img_str), AV_OPT_TYPE_STRING, {.str = NULL }, 0, 0, A },
63
{ "win_size", "set window size", OFFSET(fft_bits), AV_OPT_TYPE_INT, {.i64=12}, 4, 16, A, "fft" },
64
{ "w16", 0, 0, AV_OPT_TYPE_CONST, {.i64=4}, 0, 0, A, "fft" },
65
{ "w32", 0, 0, AV_OPT_TYPE_CONST, {.i64=5}, 0, 0, A, "fft" },
66
{ "w64", 0, 0, AV_OPT_TYPE_CONST, {.i64=6}, 0, 0, A, "fft" },
67
{ "w128", 0, 0, AV_OPT_TYPE_CONST, {.i64=7}, 0, 0, A, "fft" },
68
{ "w256", 0, 0, AV_OPT_TYPE_CONST, {.i64=8}, 0, 0, A, "fft" },
69
{ "w512", 0, 0, AV_OPT_TYPE_CONST, {.i64=9}, 0, 0, A, "fft" },
70
{ "w1024", 0, 0, AV_OPT_TYPE_CONST, {.i64=10}, 0, 0, A, "fft" },
71
{ "w2048", 0, 0, AV_OPT_TYPE_CONST, {.i64=11}, 0, 0, A, "fft" },
72
{ "w4096", 0, 0, AV_OPT_TYPE_CONST, {.i64=12}, 0, 0, A, "fft" },
73
{ "w8192", 0, 0, AV_OPT_TYPE_CONST, {.i64=13}, 0, 0, A, "fft" },
74
{ "w16384", 0, 0, AV_OPT_TYPE_CONST, {.i64=14}, 0, 0, A, "fft" },
75
{ "w32768", 0, 0, AV_OPT_TYPE_CONST, {.i64=15}, 0, 0, A, "fft" },
76
{ "w65536", 0, 0, AV_OPT_TYPE_CONST, {.i64=16}, 0, 0, A, "fft" },
77
{ "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64 = WFUNC_HANNING}, 0, NB_WFUNC-1, A, "win_func" },
78
{ "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, A, "win_func" },
79
{ "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, A, "win_func" },
80
{ "hann", "Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
81
{ "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
82
{ "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, A, "win_func" },
83
{ "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, A, "win_func" },
84
{ "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=0.75}, 0, 1, A },
85
{ NULL },
86
};
87
88
AVFILTER_DEFINE_CLASS(afftfilt);
89
90
static int config_input(AVFilterLink *inlink)
91
{
92
AVFilterContext *ctx = inlink->dst;
93
AFFTFiltContext *s = ctx->priv;
94
char *saveptr = NULL;
95
int ret = 0, ch, i;
96
float overlap;
97
char *args;
98
const char *last_expr = "1";
99
100
s->fft = av_fft_init(s->fft_bits, 0);
101
s->ifft = av_fft_init(s->fft_bits, 1);
102
if (!s->fft || !s->ifft)
103
return AVERROR(ENOMEM);
104
105
s->window_size = 1 << s->fft_bits;
106
107
s->fft_data = av_calloc(inlink->channels, sizeof(*s->fft_data));
108
if (!s->fft_data)
109
return AVERROR(ENOMEM);
110
111
for (ch = 0; ch < inlink->channels; ch++) {
112
s->fft_data[ch] = av_calloc(s->window_size, sizeof(**s->fft_data));
113
if (!s->fft_data[ch])
114
return AVERROR(ENOMEM);
115
}
116
117
s->real = av_calloc(inlink->channels, sizeof(*s->real));
118
if (!s->real)
119
return AVERROR(ENOMEM);
120
121
s->imag = av_calloc(inlink->channels, sizeof(*s->imag));
122
if (!s->imag)
123
return AVERROR(ENOMEM);
124
125
args = av_strdup(s->real_str);
126
if (!args)
127
return AVERROR(ENOMEM);
128
129
for (ch = 0; ch < inlink->channels; ch++) {
130
char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
131
132
ret = av_expr_parse(&s->real[ch], arg ? arg : last_expr, var_names,
133
NULL, NULL, NULL, NULL, 0, ctx);
134
if (ret < 0)
135
break;
136
if (arg)
137
last_expr = arg;
138
s->nb_exprs++;
139
}
140
141
av_free(args);
142
143
args = av_strdup(s->img_str ? s->img_str : s->real_str);
144
if (!args)
145
return AVERROR(ENOMEM);
146
147
for (ch = 0; ch < inlink->channels; ch++) {
148
char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
149
150
ret = av_expr_parse(&s->imag[ch], arg ? arg : last_expr, var_names,
151
NULL, NULL, NULL, NULL, 0, ctx);
152
if (ret < 0)
153
break;
154
if (arg)
155
last_expr = arg;
156
}
157
158
av_free(args);
159
160
s->fifo = av_audio_fifo_alloc(inlink->format, inlink->channels, s->window_size);
161
if (!s->fifo)
162
return AVERROR(ENOMEM);
163
164
s->window_func_lut = av_realloc_f(s->window_func_lut, s->window_size,
165
sizeof(*s->window_func_lut));
166
if (!s->window_func_lut)
167
return AVERROR(ENOMEM);
168
ff_generate_window_func(s->window_func_lut, s->window_size, s->win_func, &overlap);
169
if (s->overlap == 1)
170
s->overlap = overlap;
171
172
for (s->win_scale = 0, i = 0; i < s->window_size; i++) {
173
s->win_scale += s->window_func_lut[i] * s->window_func_lut[i];
174
}
175
176
s->hop_size = s->window_size * (1 - s->overlap);
177
if (s->hop_size <= 0)
178
return AVERROR(EINVAL);
179
180
s->buffer = ff_get_audio_buffer(inlink, s->window_size * 2);
181
if (!s->buffer)
182
return AVERROR(ENOMEM);
183
184
return ret;
185
}
186
187
static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
188
{
189
AVFilterContext *ctx = inlink->dst;
190
AVFilterLink *outlink = ctx->outputs[0];
191
AFFTFiltContext *s = ctx->priv;
192
const int window_size = s->window_size;
193
const float f = 1. / s->win_scale;
194
double values[VAR_VARS_NB];
195
AVFrame *out, *in = NULL;
196
int ch, n, ret, i, j, k;
197
int start = s->start, end = s->end;
198
199
av_audio_fifo_write(s->fifo, (void **)frame->extended_data, frame->nb_samples);
200
av_frame_free(&frame);
201
202
while (av_audio_fifo_size(s->fifo) >= window_size) {
203
if (!in) {
204
in = ff_get_audio_buffer(outlink, window_size);
205
if (!in)
206
return AVERROR(ENOMEM);
207
}
208
209
ret = av_audio_fifo_peek(s->fifo, (void **)in->extended_data, window_size);
210
if (ret < 0)
211
break;
212
213
for (ch = 0; ch < inlink->channels; ch++) {
214
const float *src = (float *)in->extended_data[ch];
215
FFTComplex *fft_data = s->fft_data[ch];
216
217
for (n = 0; n < in->nb_samples; n++) {
218
fft_data[n].re = src[n] * s->window_func_lut[n];
219
fft_data[n].im = 0;
220
}
221
222
for (; n < window_size; n++) {
223
fft_data[n].re = 0;
224
fft_data[n].im = 0;
225
}
226
}
227
228
values[VAR_PTS] = s->pts;
229
values[VAR_SAMPLE_RATE] = inlink->sample_rate;
230
values[VAR_NBBINS] = window_size / 2;
231
values[VAR_CHANNELS] = inlink->channels;
232
233
for (ch = 0; ch < inlink->channels; ch++) {
234
FFTComplex *fft_data = s->fft_data[ch];
235
float *buf = (float *)s->buffer->extended_data[ch];
236
int x;
237
238
values[VAR_CHANNEL] = ch;
239
240
av_fft_permute(s->fft, fft_data);
241
av_fft_calc(s->fft, fft_data);
242
243
for (n = 0; n < window_size / 2; n++) {
244
float fr, fi;
245
246
values[VAR_BIN] = n;
247
248
fr = av_expr_eval(s->real[ch], values, s);
249
fi = av_expr_eval(s->imag[ch], values, s);
250
251
fft_data[n].re *= fr;
252
fft_data[n].im *= fi;
253
}
254
255
for (n = window_size / 2 + 1, x = window_size / 2 - 1; n < window_size; n++, x--) {
256
fft_data[n].re = fft_data[x].re;
257
fft_data[n].im = -fft_data[x].im;
258
}
259
260
av_fft_permute(s->ifft, fft_data);
261
av_fft_calc(s->ifft, fft_data);
262
263
start = s->start;
264
end = s->end;
265
k = end;
266
for (i = 0, j = start; j < k && i < window_size; i++, j++) {
267
buf[j] += s->fft_data[ch][i].re * f;
268
}
269
270
for (; i < window_size; i++, j++) {
271
buf[j] = s->fft_data[ch][i].re * f;
272
}
273
274
start += s->hop_size;
275
end = j;
276
}
277
278
s->start = start;
279
s->end = end;
280
281
if (start >= window_size) {
282
float *dst, *buf;
283
284
start -= window_size;
285
end -= window_size;
286
287
s->start = start;
288
s->end = end;
289
290
out = ff_get_audio_buffer(outlink, window_size);
291
if (!out) {
292
ret = AVERROR(ENOMEM);
293
break;
294
}
295
296
out->pts = s->pts;
297
s->pts += window_size;
298
299
for (ch = 0; ch < inlink->channels; ch++) {
300
dst = (float *)out->extended_data[ch];
301
buf = (float *)s->buffer->extended_data[ch];
302
303
for (n = 0; n < window_size; n++) {
304
dst[n] = buf[n] * (1 - s->overlap);
305
}
306
memmove(buf, buf + window_size, window_size * 4);
307
}
308
309
ret = ff_filter_frame(outlink, out);
310
if (ret < 0)
311
break;
312
}
313
314
av_audio_fifo_drain(s->fifo, s->hop_size);
315
}
316
317
av_frame_free(&in);
318
return ret;
319
}
320
321
static int query_formats(AVFilterContext *ctx)
322
{
323
AVFilterFormats *formats;
324
AVFilterChannelLayouts *layouts;
325
static const enum AVSampleFormat sample_fmts[] = {
326
AV_SAMPLE_FMT_FLTP,
327
AV_SAMPLE_FMT_NONE
328
};
329
int ret;
330
331
layouts = ff_all_channel_counts();
332
if (!layouts)
333
return AVERROR(ENOMEM);
334
ret = ff_set_common_channel_layouts(ctx, layouts);
335
if (ret < 0)
336
return ret;
337
338
formats = ff_make_format_list(sample_fmts);
339
if (!formats)
340
return AVERROR(ENOMEM);
341
ret = ff_set_common_formats(ctx, formats);
342
if (ret < 0)
343
return ret;
344
345
formats = ff_all_samplerates();
346
if (!formats)
347
return AVERROR(ENOMEM);
348
return ff_set_common_samplerates(ctx, formats);
349
}
350
351
static av_cold void uninit(AVFilterContext *ctx)
352
{
353
AFFTFiltContext *s = ctx->priv;
354
int i;
355
356
av_fft_end(s->fft);
357
av_fft_end(s->ifft);
358
359
for (i = 0; i < s->nb_exprs; i++) {
360
if (s->fft_data)
361
av_freep(&s->fft_data[i]);
362
}
363
av_freep(&s->fft_data);
364
365
for (i = 0; i < s->nb_exprs; i++) {
366
av_expr_free(s->real[i]);
367
av_expr_free(s->imag[i]);
368
}
369
370
av_freep(&s->real);
371
av_freep(&s->imag);
372
av_frame_free(&s->buffer);
373
}
374
375
static const AVFilterPad inputs[] = {
376
{
377
.name = "default",
378
.type = AVMEDIA_TYPE_AUDIO,
379
.config_props = config_input,
380
.filter_frame = filter_frame,
381
},
382
{ NULL }
383
};
384
385
static const AVFilterPad outputs[] = {
386
{
387
.name = "default",
388
.type = AVMEDIA_TYPE_AUDIO,
389
},
390
{ NULL }
391
};
392
393
AVFilter ff_af_afftfilt = {
394
.name = "afftfilt",
395
.description = NULL_IF_CONFIG_SMALL("Apply arbitrary expressions to samples in frequency domain."),
396
.priv_size = sizeof(AFFTFiltContext),
397
.priv_class = &afftfilt_class,
398
.inputs = inputs,
399
.outputs = outputs,
400
.query_formats = query_formats,
401
.uninit = uninit,
402
};
403
404