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 "filters.h"
30 #include "window_func.h"
31
32 typedef struct AFFTFiltContext {
33 const AVClass *class;
34 char *real_str;
35 char *img_str;
36 int fft_size;
37 int fft_bits;
38
39 FFTContext *fft, *ifft;
40 FFTComplex **fft_data;
41 FFTComplex **fft_temp;
42 int nb_exprs;
43 int channels;
44 int window_size;
45 AVExpr **real;
46 AVExpr **imag;
47 AVAudioFifo *fifo;
48 int64_t pts;
49 int hop_size;
50 float overlap;
51 AVFrame *buffer;
52 int eof;
53 int win_func;
54 float *window_func_lut;
55 } AFFTFiltContext;
56
57 static const char *const var_names[] = { "sr", "b", "nb", "ch", "chs", "pts", "re", "im", NULL };
58 enum { VAR_SAMPLE_RATE, VAR_BIN, VAR_NBBINS, VAR_CHANNEL, VAR_CHANNELS, VAR_PTS, VAR_REAL, VAR_IMAG, VAR_VARS_NB };
59
60 #define OFFSET(x) offsetof(AFFTFiltContext, x)
61 #define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
62
63 static const AVOption afftfilt_options[] = {
64 { "real", "set channels real expressions", OFFSET(real_str), AV_OPT_TYPE_STRING, {.str = "re" }, 0, 0, A },
65 { "imag", "set channels imaginary expressions", OFFSET(img_str), AV_OPT_TYPE_STRING, {.str = "im" }, 0, 0, A },
66 { "win_size", "set window size", OFFSET(fft_size), AV_OPT_TYPE_INT, {.i64=4096}, 16, 131072, A },
67 { "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64 = WFUNC_HANNING}, 0, NB_WFUNC-1, A, "win_func" },
68 { "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, A, "win_func" },
69 { "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, A, "win_func" },
70 { "hann", "Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
71 { "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
72 { "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, A, "win_func" },
73 { "blackman", "Blackman", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BLACKMAN}, 0, 0, A, "win_func" },
74 { "welch", "Welch", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_WELCH}, 0, 0, A, "win_func" },
75 { "flattop", "Flat-top", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_FLATTOP}, 0, 0, A, "win_func" },
76 { "bharris", "Blackman-Harris", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BHARRIS}, 0, 0, A, "win_func" },
77 { "bnuttall", "Blackman-Nuttall", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BNUTTALL}, 0, 0, A, "win_func" },
78 { "bhann", "Bartlett-Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BHANN}, 0, 0, A, "win_func" },
79 { "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, A, "win_func" },
80 { "nuttall", "Nuttall", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_NUTTALL}, 0, 0, A, "win_func" },
81 { "lanczos", "Lanczos", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_LANCZOS}, 0, 0, A, "win_func" },
82 { "gauss", "Gauss", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_GAUSS}, 0, 0, A, "win_func" },
83 { "tukey", "Tukey", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_TUKEY}, 0, 0, A, "win_func" },
84 { "dolph", "Dolph-Chebyshev", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_DOLPH}, 0, 0, A, "win_func" },
85 { "cauchy", "Cauchy", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_CAUCHY}, 0, 0, A, "win_func" },
86 { "parzen", "Parzen", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_PARZEN}, 0, 0, A, "win_func" },
87 { "poisson", "Poisson", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_POISSON}, 0, 0, A, "win_func" },
88 { "bohman", "Bohman", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BOHMAN}, 0, 0, A, "win_func" },
89 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=0.75}, 0, 1, A },
90 { NULL },
91 };
92
93 AVFILTER_DEFINE_CLASS(afftfilt);
94
getreal(void * priv,double x,double ch)95 static inline double getreal(void *priv, double x, double ch)
96 {
97 AFFTFiltContext *s = priv;
98 int ich, ix;
99
100 ich = av_clip(ch, 0, s->nb_exprs - 1);
101 ix = av_clip(x, 0, s->window_size / 2);
102
103 return s->fft_data[ich][ix].re;
104 }
105
getimag(void * priv,double x,double ch)106 static inline double getimag(void *priv, double x, double ch)
107 {
108 AFFTFiltContext *s = priv;
109 int ich, ix;
110
111 ich = av_clip(ch, 0, s->nb_exprs - 1);
112 ix = av_clip(x, 0, s->window_size / 2);
113
114 return s->fft_data[ich][ix].im;
115 }
116
realf(void * priv,double x,double ch)117 static double realf(void *priv, double x, double ch) { return getreal(priv, x, ch); }
imagf(void * priv,double x,double ch)118 static double imagf(void *priv, double x, double ch) { return getimag(priv, x, ch); }
119
120 static const char *const func2_names[] = { "real", "imag", NULL };
121 static double (*const func2[])(void *, double, double) = { realf, imagf, NULL };
122
config_input(AVFilterLink * inlink)123 static int config_input(AVFilterLink *inlink)
124 {
125 AVFilterContext *ctx = inlink->dst;
126 AFFTFiltContext *s = ctx->priv;
127 char *saveptr = NULL;
128 int ret = 0, ch;
129 float overlap;
130 char *args;
131 const char *last_expr = "1";
132
133 s->channels = inlink->channels;
134 s->pts = AV_NOPTS_VALUE;
135 s->fft_bits = av_log2(s->fft_size);
136 s->fft = av_fft_init(s->fft_bits, 0);
137 s->ifft = av_fft_init(s->fft_bits, 1);
138 if (!s->fft || !s->ifft)
139 return AVERROR(ENOMEM);
140
141 s->window_size = 1 << s->fft_bits;
142
143 s->fft_data = av_calloc(inlink->channels, sizeof(*s->fft_data));
144 if (!s->fft_data)
145 return AVERROR(ENOMEM);
146
147 s->fft_temp = av_calloc(inlink->channels, sizeof(*s->fft_temp));
148 if (!s->fft_temp)
149 return AVERROR(ENOMEM);
150
151 for (ch = 0; ch < inlink->channels; ch++) {
152 s->fft_data[ch] = av_calloc(s->window_size, sizeof(**s->fft_data));
153 if (!s->fft_data[ch])
154 return AVERROR(ENOMEM);
155 }
156
157 for (ch = 0; ch < inlink->channels; ch++) {
158 s->fft_temp[ch] = av_calloc(s->window_size, sizeof(**s->fft_temp));
159 if (!s->fft_temp[ch])
160 return AVERROR(ENOMEM);
161 }
162
163 s->real = av_calloc(inlink->channels, sizeof(*s->real));
164 if (!s->real)
165 return AVERROR(ENOMEM);
166
167 s->imag = av_calloc(inlink->channels, sizeof(*s->imag));
168 if (!s->imag)
169 return AVERROR(ENOMEM);
170
171 args = av_strdup(s->real_str);
172 if (!args)
173 return AVERROR(ENOMEM);
174
175 for (ch = 0; ch < inlink->channels; ch++) {
176 char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
177
178 ret = av_expr_parse(&s->real[ch], arg ? arg : last_expr, var_names,
179 NULL, NULL, func2_names, func2, 0, ctx);
180 if (ret < 0)
181 goto fail;
182 if (arg)
183 last_expr = arg;
184 s->nb_exprs++;
185 }
186
187 av_freep(&args);
188
189 args = av_strdup(s->img_str ? s->img_str : s->real_str);
190 if (!args)
191 return AVERROR(ENOMEM);
192
193 saveptr = NULL;
194 last_expr = "1";
195 for (ch = 0; ch < inlink->channels; ch++) {
196 char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
197
198 ret = av_expr_parse(&s->imag[ch], arg ? arg : last_expr, var_names,
199 NULL, NULL, func2_names, func2, 0, ctx);
200 if (ret < 0)
201 goto fail;
202 if (arg)
203 last_expr = arg;
204 }
205
206 av_freep(&args);
207
208 s->fifo = av_audio_fifo_alloc(inlink->format, inlink->channels, s->window_size);
209 if (!s->fifo)
210 return AVERROR(ENOMEM);
211
212 s->window_func_lut = av_realloc_f(s->window_func_lut, s->window_size,
213 sizeof(*s->window_func_lut));
214 if (!s->window_func_lut)
215 return AVERROR(ENOMEM);
216 generate_window_func(s->window_func_lut, s->window_size, s->win_func, &overlap);
217 if (s->overlap == 1)
218 s->overlap = overlap;
219
220 s->hop_size = s->window_size * (1 - s->overlap);
221 if (s->hop_size <= 0)
222 return AVERROR(EINVAL);
223
224 s->buffer = ff_get_audio_buffer(inlink, s->window_size * 2);
225 if (!s->buffer)
226 return AVERROR(ENOMEM);
227
228 fail:
229 av_freep(&args);
230
231 return ret;
232 }
233
filter_frame(AVFilterLink * inlink)234 static int filter_frame(AVFilterLink *inlink)
235 {
236 AVFilterContext *ctx = inlink->dst;
237 AVFilterLink *outlink = ctx->outputs[0];
238 AFFTFiltContext *s = ctx->priv;
239 const int window_size = s->window_size;
240 const float f = 1. / (s->window_size / 2);
241 double values[VAR_VARS_NB];
242 AVFrame *out, *in = NULL;
243 int ch, n, ret, i;
244
245 if (!in) {
246 in = ff_get_audio_buffer(outlink, window_size);
247 if (!in)
248 return AVERROR(ENOMEM);
249 }
250
251 ret = av_audio_fifo_peek(s->fifo, (void **)in->extended_data, window_size);
252 if (ret < 0)
253 goto fail;
254
255 for (ch = 0; ch < inlink->channels; ch++) {
256 const float *src = (float *)in->extended_data[ch];
257 FFTComplex *fft_data = s->fft_data[ch];
258
259 for (n = 0; n < in->nb_samples; n++) {
260 fft_data[n].re = src[n] * s->window_func_lut[n];
261 fft_data[n].im = 0;
262 }
263
264 for (; n < window_size; n++) {
265 fft_data[n].re = 0;
266 fft_data[n].im = 0;
267 }
268 }
269
270 values[VAR_PTS] = s->pts;
271 values[VAR_SAMPLE_RATE] = inlink->sample_rate;
272 values[VAR_NBBINS] = window_size / 2;
273 values[VAR_CHANNELS] = inlink->channels;
274
275 for (ch = 0; ch < inlink->channels; ch++) {
276 FFTComplex *fft_data = s->fft_data[ch];
277
278 av_fft_permute(s->fft, fft_data);
279 av_fft_calc(s->fft, fft_data);
280 }
281
282 for (ch = 0; ch < inlink->channels; ch++) {
283 FFTComplex *fft_data = s->fft_data[ch];
284 FFTComplex *fft_temp = s->fft_temp[ch];
285 float *buf = (float *)s->buffer->extended_data[ch];
286 int x;
287 values[VAR_CHANNEL] = ch;
288
289 for (n = 0; n <= window_size / 2; n++) {
290 float fr, fi;
291
292 values[VAR_BIN] = n;
293 values[VAR_REAL] = fft_data[n].re;
294 values[VAR_IMAG] = fft_data[n].im;
295
296 fr = av_expr_eval(s->real[ch], values, s);
297 fi = av_expr_eval(s->imag[ch], values, s);
298
299 fft_temp[n].re = fr;
300 fft_temp[n].im = fi;
301 }
302
303 for (n = window_size / 2 + 1, x = window_size / 2 - 1; n < window_size; n++, x--) {
304 fft_temp[n].re = fft_temp[x].re;
305 fft_temp[n].im = -fft_temp[x].im;
306 }
307
308 av_fft_permute(s->ifft, fft_temp);
309 av_fft_calc(s->ifft, fft_temp);
310
311 for (i = 0; i < window_size; i++) {
312 buf[i] += s->fft_temp[ch][i].re * f;
313 }
314 }
315
316 out = ff_get_audio_buffer(outlink, s->hop_size);
317 if (!out) {
318 ret = AVERROR(ENOMEM);
319 goto fail;
320 }
321
322 out->pts = s->pts;
323 s->pts += av_rescale_q(s->hop_size, (AVRational){1, outlink->sample_rate}, outlink->time_base);
324
325 for (ch = 0; ch < inlink->channels; ch++) {
326 float *dst = (float *)out->extended_data[ch];
327 float *buf = (float *)s->buffer->extended_data[ch];
328
329 for (n = 0; n < s->hop_size; n++)
330 dst[n] = buf[n] * (1.f - s->overlap);
331 memmove(buf, buf + s->hop_size, window_size * 4);
332 }
333
334 ret = ff_filter_frame(outlink, out);
335 if (ret < 0)
336 goto fail;
337
338 av_audio_fifo_drain(s->fifo, s->hop_size);
339
340 fail:
341 av_frame_free(&in);
342 return ret < 0 ? ret : 0;
343 }
344
activate(AVFilterContext * ctx)345 static int activate(AVFilterContext *ctx)
346 {
347 AVFilterLink *inlink = ctx->inputs[0];
348 AVFilterLink *outlink = ctx->outputs[0];
349 AFFTFiltContext *s = ctx->priv;
350 AVFrame *in = NULL;
351 int ret = 0, status;
352 int64_t pts;
353
354 FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink);
355
356 if (!s->eof && av_audio_fifo_size(s->fifo) < s->window_size) {
357 ret = ff_inlink_consume_frame(inlink, &in);
358 if (ret < 0)
359 return ret;
360
361 if (ret > 0) {
362 ret = av_audio_fifo_write(s->fifo, (void **)in->extended_data,
363 in->nb_samples);
364 if (ret >= 0 && s->pts == AV_NOPTS_VALUE)
365 s->pts = in->pts;
366
367 av_frame_free(&in);
368 if (ret < 0)
369 return ret;
370 }
371 }
372
373 if ((av_audio_fifo_size(s->fifo) >= s->window_size) ||
374 (av_audio_fifo_size(s->fifo) > 0 && s->eof)) {
375 ret = filter_frame(inlink);
376 if (av_audio_fifo_size(s->fifo) >= s->window_size)
377 ff_filter_set_ready(ctx, 100);
378 return ret;
379 }
380
381 if (!s->eof && ff_inlink_acknowledge_status(inlink, &status, &pts)) {
382 if (status == AVERROR_EOF) {
383 s->eof = 1;
384 if (av_audio_fifo_size(s->fifo) >= 0) {
385 ff_filter_set_ready(ctx, 100);
386 return 0;
387 }
388 }
389 }
390
391 if (s->eof && av_audio_fifo_size(s->fifo) <= 0) {
392 ff_outlink_set_status(outlink, AVERROR_EOF, s->pts);
393 return 0;
394 }
395
396 if (!s->eof)
397 FF_FILTER_FORWARD_WANTED(outlink, inlink);
398
399 return FFERROR_NOT_READY;
400 }
401
query_formats(AVFilterContext * ctx)402 static int query_formats(AVFilterContext *ctx)
403 {
404 AVFilterFormats *formats;
405 AVFilterChannelLayouts *layouts;
406 static const enum AVSampleFormat sample_fmts[] = {
407 AV_SAMPLE_FMT_FLTP,
408 AV_SAMPLE_FMT_NONE
409 };
410 int ret;
411
412 layouts = ff_all_channel_counts();
413 if (!layouts)
414 return AVERROR(ENOMEM);
415 ret = ff_set_common_channel_layouts(ctx, layouts);
416 if (ret < 0)
417 return ret;
418
419 formats = ff_make_format_list(sample_fmts);
420 if (!formats)
421 return AVERROR(ENOMEM);
422 ret = ff_set_common_formats(ctx, formats);
423 if (ret < 0)
424 return ret;
425
426 formats = ff_all_samplerates();
427 if (!formats)
428 return AVERROR(ENOMEM);
429 return ff_set_common_samplerates(ctx, formats);
430 }
431
uninit(AVFilterContext * ctx)432 static av_cold void uninit(AVFilterContext *ctx)
433 {
434 AFFTFiltContext *s = ctx->priv;
435 int i;
436
437 av_fft_end(s->fft);
438 av_fft_end(s->ifft);
439
440 for (i = 0; i < s->channels; i++) {
441 if (s->fft_data)
442 av_freep(&s->fft_data[i]);
443 if (s->fft_temp)
444 av_freep(&s->fft_temp[i]);
445 }
446 av_freep(&s->fft_data);
447 av_freep(&s->fft_temp);
448
449 for (i = 0; i < s->nb_exprs; i++) {
450 av_expr_free(s->real[i]);
451 av_expr_free(s->imag[i]);
452 }
453
454 av_freep(&s->real);
455 av_freep(&s->imag);
456 av_frame_free(&s->buffer);
457 av_freep(&s->window_func_lut);
458
459 av_audio_fifo_free(s->fifo);
460 }
461
462 static const AVFilterPad inputs[] = {
463 {
464 .name = "default",
465 .type = AVMEDIA_TYPE_AUDIO,
466 .config_props = config_input,
467 },
468 { NULL }
469 };
470
471 static const AVFilterPad outputs[] = {
472 {
473 .name = "default",
474 .type = AVMEDIA_TYPE_AUDIO,
475 },
476 { NULL }
477 };
478
479 AVFilter ff_af_afftfilt = {
480 .name = "afftfilt",
481 .description = NULL_IF_CONFIG_SMALL("Apply arbitrary expressions to samples in frequency domain."),
482 .priv_size = sizeof(AFFTFiltContext),
483 .priv_class = &afftfilt_class,
484 .inputs = inputs,
485 .outputs = outputs,
486 .activate = activate,
487 .query_formats = query_formats,
488 .uninit = uninit,
489 };
490