summaryrefslogtreecommitdiff
path: root/libavfilter/af_firequalizer.c (plain)
blob: 4243d66bd632e893775c832cbd037154ef5c2b34
1/*
2 * Copyright (c) 2016 Muhammad Faiz <mfcc64@gmail.com>
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/opt.h"
22#include "libavutil/eval.h"
23#include "libavutil/avassert.h"
24#include "libavcodec/avfft.h"
25#include "avfilter.h"
26#include "internal.h"
27#include "audio.h"
28
29#define RDFT_BITS_MIN 4
30#define RDFT_BITS_MAX 16
31
32enum WindowFunc {
33 WFUNC_RECTANGULAR,
34 WFUNC_HANN,
35 WFUNC_HAMMING,
36 WFUNC_BLACKMAN,
37 WFUNC_NUTTALL3,
38 WFUNC_MNUTTALL3,
39 WFUNC_NUTTALL,
40 WFUNC_BNUTTALL,
41 WFUNC_BHARRIS,
42 WFUNC_TUKEY,
43 NB_WFUNC
44};
45
46enum Scale {
47 SCALE_LINLIN,
48 SCALE_LINLOG,
49 SCALE_LOGLIN,
50 SCALE_LOGLOG,
51 NB_SCALE
52};
53
54#define NB_GAIN_ENTRY_MAX 4096
55typedef struct {
56 double freq;
57 double gain;
58} GainEntry;
59
60typedef struct {
61 int buf_idx;
62 int overlap_idx;
63} OverlapIndex;
64
65typedef struct {
66 const AVClass *class;
67
68 RDFTContext *analysis_rdft;
69 RDFTContext *analysis_irdft;
70 RDFTContext *rdft;
71 RDFTContext *irdft;
72 FFTContext *fft_ctx;
73 int analysis_rdft_len;
74 int rdft_len;
75
76 float *analysis_buf;
77 float *dump_buf;
78 float *kernel_tmp_buf;
79 float *kernel_buf;
80 float *conv_buf;
81 OverlapIndex *conv_idx;
82 int fir_len;
83 int nsamples_max;
84 int64_t next_pts;
85 int frame_nsamples_max;
86 int remaining;
87
88 char *gain_cmd;
89 char *gain_entry_cmd;
90 const char *gain;
91 const char *gain_entry;
92 double delay;
93 double accuracy;
94 int wfunc;
95 int fixed;
96 int multi;
97 int zero_phase;
98 int scale;
99 char *dumpfile;
100 int dumpscale;
101 int fft2;
102
103 int nb_gain_entry;
104 int gain_entry_err;
105 GainEntry gain_entry_tbl[NB_GAIN_ENTRY_MAX];
106} FIREqualizerContext;
107
108#define OFFSET(x) offsetof(FIREqualizerContext, x)
109#define FLAGS AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
110
111static const AVOption firequalizer_options[] = {
112 { "gain", "set gain curve", OFFSET(gain), AV_OPT_TYPE_STRING, { .str = "gain_interpolate(f)" }, 0, 0, FLAGS },
113 { "gain_entry", "set gain entry", OFFSET(gain_entry), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
114 { "delay", "set delay", OFFSET(delay), AV_OPT_TYPE_DOUBLE, { .dbl = 0.01 }, 0.0, 1e10, FLAGS },
115 { "accuracy", "set accuracy", OFFSET(accuracy), AV_OPT_TYPE_DOUBLE, { .dbl = 5.0 }, 0.0, 1e10, FLAGS },
116 { "wfunc", "set window function", OFFSET(wfunc), AV_OPT_TYPE_INT, { .i64 = WFUNC_HANN }, 0, NB_WFUNC-1, FLAGS, "wfunc" },
117 { "rectangular", "rectangular window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_RECTANGULAR }, 0, 0, FLAGS, "wfunc" },
118 { "hann", "hann window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_HANN }, 0, 0, FLAGS, "wfunc" },
119 { "hamming", "hamming window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_HAMMING }, 0, 0, FLAGS, "wfunc" },
120 { "blackman", "blackman window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_BLACKMAN }, 0, 0, FLAGS, "wfunc" },
121 { "nuttall3", "3-term nuttall window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_NUTTALL3 }, 0, 0, FLAGS, "wfunc" },
122 { "mnuttall3", "minimum 3-term nuttall window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_MNUTTALL3 }, 0, 0, FLAGS, "wfunc" },
123 { "nuttall", "nuttall window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_NUTTALL }, 0, 0, FLAGS, "wfunc" },
124 { "bnuttall", "blackman-nuttall window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_BNUTTALL }, 0, 0, FLAGS, "wfunc" },
125 { "bharris", "blackman-harris window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_BHARRIS }, 0, 0, FLAGS, "wfunc" },
126 { "tukey", "tukey window", 0, AV_OPT_TYPE_CONST, { .i64 = WFUNC_TUKEY }, 0, 0, FLAGS, "wfunc" },
127 { "fixed", "set fixed frame samples", OFFSET(fixed), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
128 { "multi", "set multi channels mode", OFFSET(multi), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
129 { "zero_phase", "set zero phase mode", OFFSET(zero_phase), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
130 { "scale", "set gain scale", OFFSET(scale), AV_OPT_TYPE_INT, { .i64 = SCALE_LINLOG }, 0, NB_SCALE-1, FLAGS, "scale" },
131 { "linlin", "linear-freq linear-gain", 0, AV_OPT_TYPE_CONST, { .i64 = SCALE_LINLIN }, 0, 0, FLAGS, "scale" },
132 { "linlog", "linear-freq logarithmic-gain", 0, AV_OPT_TYPE_CONST, { .i64 = SCALE_LINLOG }, 0, 0, FLAGS, "scale" },
133 { "loglin", "logarithmic-freq linear-gain", 0, AV_OPT_TYPE_CONST, { .i64 = SCALE_LOGLIN }, 0, 0, FLAGS, "scale" },
134 { "loglog", "logarithmic-freq logarithmic-gain", 0, AV_OPT_TYPE_CONST, { .i64 = SCALE_LOGLOG }, 0, 0, FLAGS, "scale" },
135 { "dumpfile", "set dump file", OFFSET(dumpfile), AV_OPT_TYPE_STRING, { .str = NULL }, 0, 0, FLAGS },
136 { "dumpscale", "set dump scale", OFFSET(dumpscale), AV_OPT_TYPE_INT, { .i64 = SCALE_LINLOG }, 0, NB_SCALE-1, FLAGS, "scale" },
137 { "fft2", "set 2-channels fft", OFFSET(fft2), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
138 { NULL }
139};
140
141AVFILTER_DEFINE_CLASS(firequalizer);
142
143static void common_uninit(FIREqualizerContext *s)
144{
145 av_rdft_end(s->analysis_rdft);
146 av_rdft_end(s->analysis_irdft);
147 av_rdft_end(s->rdft);
148 av_rdft_end(s->irdft);
149 av_fft_end(s->fft_ctx);
150 s->analysis_rdft = s->analysis_irdft = s->rdft = s->irdft = NULL;
151 s->fft_ctx = NULL;
152
153 av_freep(&s->analysis_buf);
154 av_freep(&s->dump_buf);
155 av_freep(&s->kernel_tmp_buf);
156 av_freep(&s->kernel_buf);
157 av_freep(&s->conv_buf);
158 av_freep(&s->conv_idx);
159}
160
161static av_cold void uninit(AVFilterContext *ctx)
162{
163 FIREqualizerContext *s = ctx->priv;
164
165 common_uninit(s);
166 av_freep(&s->gain_cmd);
167 av_freep(&s->gain_entry_cmd);
168}
169
170static int query_formats(AVFilterContext *ctx)
171{
172 AVFilterChannelLayouts *layouts;
173 AVFilterFormats *formats;
174 static const enum AVSampleFormat sample_fmts[] = {
175 AV_SAMPLE_FMT_FLTP,
176 AV_SAMPLE_FMT_NONE
177 };
178 int ret;
179
180 layouts = ff_all_channel_counts();
181 if (!layouts)
182 return AVERROR(ENOMEM);
183 ret = ff_set_common_channel_layouts(ctx, layouts);
184 if (ret < 0)
185 return ret;
186
187 formats = ff_make_format_list(sample_fmts);
188 if (!formats)
189 return AVERROR(ENOMEM);
190 ret = ff_set_common_formats(ctx, formats);
191 if (ret < 0)
192 return ret;
193
194 formats = ff_all_samplerates();
195 if (!formats)
196 return AVERROR(ENOMEM);
197 return ff_set_common_samplerates(ctx, formats);
198}
199
200static void fast_convolute(FIREqualizerContext *av_restrict s, const float *av_restrict kernel_buf, float *av_restrict conv_buf,
201 OverlapIndex *av_restrict idx, float *av_restrict data, int nsamples)
202{
203 if (nsamples <= s->nsamples_max) {
204 float *buf = conv_buf + idx->buf_idx * s->rdft_len;
205 float *obuf = conv_buf + !idx->buf_idx * s->rdft_len + idx->overlap_idx;
206 int center = s->fir_len/2;
207 int k;
208
209 memset(buf, 0, center * sizeof(*data));
210 memcpy(buf + center, data, nsamples * sizeof(*data));
211 memset(buf + center + nsamples, 0, (s->rdft_len - nsamples - center) * sizeof(*data));
212 av_rdft_calc(s->rdft, buf);
213
214 buf[0] *= kernel_buf[0];
215 buf[1] *= kernel_buf[s->rdft_len/2];
216 for (k = 1; k < s->rdft_len/2; k++) {
217 buf[2*k] *= kernel_buf[k];
218 buf[2*k+1] *= kernel_buf[k];
219 }
220
221 av_rdft_calc(s->irdft, buf);
222 for (k = 0; k < s->rdft_len - idx->overlap_idx; k++)
223 buf[k] += obuf[k];
224 memcpy(data, buf, nsamples * sizeof(*data));
225 idx->buf_idx = !idx->buf_idx;
226 idx->overlap_idx = nsamples;
227 } else {
228 while (nsamples > s->nsamples_max * 2) {
229 fast_convolute(s, kernel_buf, conv_buf, idx, data, s->nsamples_max);
230 data += s->nsamples_max;
231 nsamples -= s->nsamples_max;
232 }
233 fast_convolute(s, kernel_buf, conv_buf, idx, data, nsamples/2);
234 fast_convolute(s, kernel_buf, conv_buf, idx, data + nsamples/2, nsamples - nsamples/2);
235 }
236}
237
238static void fast_convolute2(FIREqualizerContext *av_restrict s, const float *av_restrict kernel_buf, FFTComplex *av_restrict conv_buf,
239 OverlapIndex *av_restrict idx, float *av_restrict data0, float *av_restrict data1, int nsamples)
240{
241 if (nsamples <= s->nsamples_max) {
242 FFTComplex *buf = conv_buf + idx->buf_idx * s->rdft_len;
243 FFTComplex *obuf = conv_buf + !idx->buf_idx * s->rdft_len + idx->overlap_idx;
244 int center = s->fir_len/2;
245 int k;
246 float tmp;
247
248 memset(buf, 0, center * sizeof(*buf));
249 for (k = 0; k < nsamples; k++) {
250 buf[center+k].re = data0[k];
251 buf[center+k].im = data1[k];
252 }
253 memset(buf + center + nsamples, 0, (s->rdft_len - nsamples - center) * sizeof(*buf));
254 av_fft_permute(s->fft_ctx, buf);
255 av_fft_calc(s->fft_ctx, buf);
256
257 /* swap re <-> im, do backward fft using forward fft_ctx */
258 /* normalize with 0.5f */
259 tmp = buf[0].re;
260 buf[0].re = 0.5f * kernel_buf[0] * buf[0].im;
261 buf[0].im = 0.5f * kernel_buf[0] * tmp;
262 for (k = 1; k < s->rdft_len/2; k++) {
263 int m = s->rdft_len - k;
264 tmp = buf[k].re;
265 buf[k].re = 0.5f * kernel_buf[k] * buf[k].im;
266 buf[k].im = 0.5f * kernel_buf[k] * tmp;
267 tmp = buf[m].re;
268 buf[m].re = 0.5f * kernel_buf[k] * buf[m].im;
269 buf[m].im = 0.5f * kernel_buf[k] * tmp;
270 }
271 tmp = buf[k].re;
272 buf[k].re = 0.5f * kernel_buf[k] * buf[k].im;
273 buf[k].im = 0.5f * kernel_buf[k] * tmp;
274
275 av_fft_permute(s->fft_ctx, buf);
276 av_fft_calc(s->fft_ctx, buf);
277
278 for (k = 0; k < s->rdft_len - idx->overlap_idx; k++) {
279 buf[k].re += obuf[k].re;
280 buf[k].im += obuf[k].im;
281 }
282
283 /* swapped re <-> im */
284 for (k = 0; k < nsamples; k++) {
285 data0[k] = buf[k].im;
286 data1[k] = buf[k].re;
287 }
288 idx->buf_idx = !idx->buf_idx;
289 idx->overlap_idx = nsamples;
290 } else {
291 while (nsamples > s->nsamples_max * 2) {
292 fast_convolute2(s, kernel_buf, conv_buf, idx, data0, data1, s->nsamples_max);
293 data0 += s->nsamples_max;
294 data1 += s->nsamples_max;
295 nsamples -= s->nsamples_max;
296 }
297 fast_convolute2(s, kernel_buf, conv_buf, idx, data0, data1, nsamples/2);
298 fast_convolute2(s, kernel_buf, conv_buf, idx, data0 + nsamples/2, data1 + nsamples/2, nsamples - nsamples/2);
299 }
300}
301
302static void dump_fir(AVFilterContext *ctx, FILE *fp, int ch)
303{
304 FIREqualizerContext *s = ctx->priv;
305 int rate = ctx->inputs[0]->sample_rate;
306 int xlog = s->dumpscale == SCALE_LOGLIN || s->dumpscale == SCALE_LOGLOG;
307 int ylog = s->dumpscale == SCALE_LINLOG || s->dumpscale == SCALE_LOGLOG;
308 int x;
309 int center = s->fir_len / 2;
310 double delay = s->zero_phase ? 0.0 : (double) center / rate;
311 double vx, ya, yb;
312
313 s->analysis_buf[0] *= s->rdft_len/2;
314 for (x = 1; x <= center; x++) {
315 s->analysis_buf[x] *= s->rdft_len/2;
316 s->analysis_buf[s->analysis_rdft_len - x] *= s->rdft_len/2;
317 }
318
319 if (ch)
320 fprintf(fp, "\n\n");
321
322 fprintf(fp, "# time[%d] (time amplitude)\n", ch);
323
324 for (x = center; x > 0; x--)
325 fprintf(fp, "%15.10f %15.10f\n", delay - (double) x / rate, (double) s->analysis_buf[s->analysis_rdft_len - x]);
326
327 for (x = 0; x <= center; x++)
328 fprintf(fp, "%15.10f %15.10f\n", delay + (double)x / rate , (double) s->analysis_buf[x]);
329
330 av_rdft_calc(s->analysis_rdft, s->analysis_buf);
331
332 fprintf(fp, "\n\n# freq[%d] (frequency desired_gain actual_gain)\n", ch);
333
334 for (x = 0; x <= s->analysis_rdft_len/2; x++) {
335 int i = (x == s->analysis_rdft_len/2) ? 1 : 2 * x;
336 vx = (double)x * rate / s->analysis_rdft_len;
337 if (xlog)
338 vx = log2(0.05*vx);
339 ya = s->dump_buf[i];
340 yb = s->analysis_buf[i];
341 if (ylog) {
342 ya = 20.0 * log10(fabs(ya));
343 yb = 20.0 * log10(fabs(yb));
344 }
345 fprintf(fp, "%17.10f %17.10f %17.10f\n", vx, ya, yb);
346 }
347}
348
349static double entry_func(void *p, double freq, double gain)
350{
351 AVFilterContext *ctx = p;
352 FIREqualizerContext *s = ctx->priv;
353
354 if (s->nb_gain_entry >= NB_GAIN_ENTRY_MAX) {
355 av_log(ctx, AV_LOG_ERROR, "entry table overflow.\n");
356 s->gain_entry_err = AVERROR(EINVAL);
357 return 0;
358 }
359
360 if (isnan(freq)) {
361 av_log(ctx, AV_LOG_ERROR, "nan frequency (%g, %g).\n", freq, gain);
362 s->gain_entry_err = AVERROR(EINVAL);
363 return 0;
364 }
365
366 if (s->nb_gain_entry > 0 && freq <= s->gain_entry_tbl[s->nb_gain_entry - 1].freq) {
367 av_log(ctx, AV_LOG_ERROR, "unsorted frequency (%g, %g).\n", freq, gain);
368 s->gain_entry_err = AVERROR(EINVAL);
369 return 0;
370 }
371
372 s->gain_entry_tbl[s->nb_gain_entry].freq = freq;
373 s->gain_entry_tbl[s->nb_gain_entry].gain = gain;
374 s->nb_gain_entry++;
375 return 0;
376}
377
378static int gain_entry_compare(const void *key, const void *memb)
379{
380 const double *freq = key;
381 const GainEntry *entry = memb;
382
383 if (*freq < entry[0].freq)
384 return -1;
385 if (*freq > entry[1].freq)
386 return 1;
387 return 0;
388}
389
390static double gain_interpolate_func(void *p, double freq)
391{
392 AVFilterContext *ctx = p;
393 FIREqualizerContext *s = ctx->priv;
394 GainEntry *res;
395 double d0, d1, d;
396
397 if (isnan(freq))
398 return freq;
399
400 if (!s->nb_gain_entry)
401 return 0;
402
403 if (freq <= s->gain_entry_tbl[0].freq)
404 return s->gain_entry_tbl[0].gain;
405
406 if (freq >= s->gain_entry_tbl[s->nb_gain_entry-1].freq)
407 return s->gain_entry_tbl[s->nb_gain_entry-1].gain;
408
409 res = bsearch(&freq, &s->gain_entry_tbl, s->nb_gain_entry - 1, sizeof(*res), gain_entry_compare);
410 av_assert0(res);
411
412 d = res[1].freq - res[0].freq;
413 d0 = freq - res[0].freq;
414 d1 = res[1].freq - freq;
415
416 if (d0 && d1)
417 return (d0 * res[1].gain + d1 * res[0].gain) / d;
418
419 if (d0)
420 return res[1].gain;
421
422 return res[0].gain;
423}
424
425static double cubic_interpolate_func(void *p, double freq)
426{
427 AVFilterContext *ctx = p;
428 FIREqualizerContext *s = ctx->priv;
429 GainEntry *res;
430 double x, x2, x3;
431 double a, b, c, d;
432 double m0, m1, m2, msum, unit;
433
434 if (!s->nb_gain_entry)
435 return 0;
436
437 if (freq <= s->gain_entry_tbl[0].freq)
438 return s->gain_entry_tbl[0].gain;
439
440 if (freq >= s->gain_entry_tbl[s->nb_gain_entry-1].freq)
441 return s->gain_entry_tbl[s->nb_gain_entry-1].gain;
442
443 res = bsearch(&freq, &s->gain_entry_tbl, s->nb_gain_entry - 1, sizeof(*res), gain_entry_compare);
444 av_assert0(res);
445
446 unit = res[1].freq - res[0].freq;
447 m0 = res != s->gain_entry_tbl ?
448 unit * (res[0].gain - res[-1].gain) / (res[0].freq - res[-1].freq) : 0;
449 m1 = res[1].gain - res[0].gain;
450 m2 = res != s->gain_entry_tbl + s->nb_gain_entry - 2 ?
451 unit * (res[2].gain - res[1].gain) / (res[2].freq - res[1].freq) : 0;
452
453 msum = fabs(m0) + fabs(m1);
454 m0 = msum > 0 ? (fabs(m0) * m1 + fabs(m1) * m0) / msum : 0;
455 msum = fabs(m1) + fabs(m2);
456 m1 = msum > 0 ? (fabs(m1) * m2 + fabs(m2) * m1) / msum : 0;
457
458 d = res[0].gain;
459 c = m0;
460 b = 3 * res[1].gain - m1 - 2 * c - 3 * d;
461 a = res[1].gain - b - c - d;
462
463 x = (freq - res[0].freq) / unit;
464 x2 = x * x;
465 x3 = x2 * x;
466
467 return a * x3 + b * x2 + c * x + d;
468}
469
470static const char *const var_names[] = {
471 "f",
472 "sr",
473 "ch",
474 "chid",
475 "chs",
476 "chlayout",
477 NULL
478};
479
480enum VarOffset {
481 VAR_F,
482 VAR_SR,
483 VAR_CH,
484 VAR_CHID,
485 VAR_CHS,
486 VAR_CHLAYOUT,
487 VAR_NB
488};
489
490static int generate_kernel(AVFilterContext *ctx, const char *gain, const char *gain_entry)
491{
492 FIREqualizerContext *s = ctx->priv;
493 AVFilterLink *inlink = ctx->inputs[0];
494 const char *gain_entry_func_names[] = { "entry", NULL };
495 const char *gain_func_names[] = { "gain_interpolate", "cubic_interpolate", NULL };
496 double (*gain_entry_funcs[])(void *, double, double) = { entry_func, NULL };
497 double (*gain_funcs[])(void *, double) = { gain_interpolate_func, cubic_interpolate_func, NULL };
498 double vars[VAR_NB];
499 AVExpr *gain_expr;
500 int ret, k, center, ch;
501 int xlog = s->scale == SCALE_LOGLIN || s->scale == SCALE_LOGLOG;
502 int ylog = s->scale == SCALE_LINLOG || s->scale == SCALE_LOGLOG;
503 FILE *dump_fp = NULL;
504
505 s->nb_gain_entry = 0;
506 s->gain_entry_err = 0;
507 if (gain_entry) {
508 double result = 0.0;
509 ret = av_expr_parse_and_eval(&result, gain_entry, NULL, NULL, NULL, NULL,
510 gain_entry_func_names, gain_entry_funcs, ctx, 0, ctx);
511 if (ret < 0)
512 return ret;
513 if (s->gain_entry_err < 0)
514 return s->gain_entry_err;
515 }
516
517 av_log(ctx, AV_LOG_DEBUG, "nb_gain_entry = %d.\n", s->nb_gain_entry);
518
519 ret = av_expr_parse(&gain_expr, gain, var_names,
520 gain_func_names, gain_funcs, NULL, NULL, 0, ctx);
521 if (ret < 0)
522 return ret;
523
524 if (s->dumpfile && (!s->dump_buf || !s->analysis_rdft || !(dump_fp = fopen(s->dumpfile, "w"))))
525 av_log(ctx, AV_LOG_WARNING, "dumping failed.\n");
526
527 vars[VAR_CHS] = inlink->channels;
528 vars[VAR_CHLAYOUT] = inlink->channel_layout;
529 vars[VAR_SR] = inlink->sample_rate;
530 for (ch = 0; ch < inlink->channels; ch++) {
531 float *rdft_buf = s->kernel_tmp_buf + ch * s->rdft_len;
532 double result;
533 vars[VAR_CH] = ch;
534 vars[VAR_CHID] = av_channel_layout_extract_channel(inlink->channel_layout, ch);
535 vars[VAR_F] = 0.0;
536 if (xlog)
537 vars[VAR_F] = log2(0.05 * vars[VAR_F]);
538 result = av_expr_eval(gain_expr, vars, ctx);
539 s->analysis_buf[0] = ylog ? pow(10.0, 0.05 * result) : result;
540
541 vars[VAR_F] = 0.5 * inlink->sample_rate;
542 if (xlog)
543 vars[VAR_F] = log2(0.05 * vars[VAR_F]);
544 result = av_expr_eval(gain_expr, vars, ctx);
545 s->analysis_buf[1] = ylog ? pow(10.0, 0.05 * result) : result;
546
547 for (k = 1; k < s->analysis_rdft_len/2; k++) {
548 vars[VAR_F] = k * ((double)inlink->sample_rate /(double)s->analysis_rdft_len);
549 if (xlog)
550 vars[VAR_F] = log2(0.05 * vars[VAR_F]);
551 result = av_expr_eval(gain_expr, vars, ctx);
552 s->analysis_buf[2*k] = ylog ? pow(10.0, 0.05 * result) : result;
553 s->analysis_buf[2*k+1] = 0.0;
554 }
555
556 if (s->dump_buf)
557 memcpy(s->dump_buf, s->analysis_buf, s->analysis_rdft_len * sizeof(*s->analysis_buf));
558
559 av_rdft_calc(s->analysis_irdft, s->analysis_buf);
560 center = s->fir_len / 2;
561
562 for (k = 0; k <= center; k++) {
563 double u = k * (M_PI/center);
564 double win;
565 switch (s->wfunc) {
566 case WFUNC_RECTANGULAR:
567 win = 1.0;
568 break;
569 case WFUNC_HANN:
570 win = 0.5 + 0.5 * cos(u);
571 break;
572 case WFUNC_HAMMING:
573 win = 0.53836 + 0.46164 * cos(u);
574 break;
575 case WFUNC_BLACKMAN:
576 win = 0.42 + 0.5 * cos(u) + 0.08 * cos(2*u);
577 break;
578 case WFUNC_NUTTALL3:
579 win = 0.40897 + 0.5 * cos(u) + 0.09103 * cos(2*u);
580 break;
581 case WFUNC_MNUTTALL3:
582 win = 0.4243801 + 0.4973406 * cos(u) + 0.0782793 * cos(2*u);
583 break;
584 case WFUNC_NUTTALL:
585 win = 0.355768 + 0.487396 * cos(u) + 0.144232 * cos(2*u) + 0.012604 * cos(3*u);
586 break;
587 case WFUNC_BNUTTALL:
588 win = 0.3635819 + 0.4891775 * cos(u) + 0.1365995 * cos(2*u) + 0.0106411 * cos(3*u);
589 break;
590 case WFUNC_BHARRIS:
591 win = 0.35875 + 0.48829 * cos(u) + 0.14128 * cos(2*u) + 0.01168 * cos(3*u);
592 break;
593 case WFUNC_TUKEY:
594 win = (u <= 0.5 * M_PI) ? 1.0 : (0.5 + 0.5 * cos(2*u - M_PI));
595 break;
596 default:
597 av_assert0(0);
598 }
599 s->analysis_buf[k] *= (2.0/s->analysis_rdft_len) * (2.0/s->rdft_len) * win;
600 if (k)
601 s->analysis_buf[s->analysis_rdft_len - k] = s->analysis_buf[k];
602 }
603
604 memset(s->analysis_buf + center + 1, 0, (s->analysis_rdft_len - s->fir_len) * sizeof(*s->analysis_buf));
605 memcpy(rdft_buf, s->analysis_buf, s->rdft_len/2 * sizeof(*s->analysis_buf));
606 memcpy(rdft_buf + s->rdft_len/2, s->analysis_buf + s->analysis_rdft_len - s->rdft_len/2, s->rdft_len/2 * sizeof(*s->analysis_buf));
607 av_rdft_calc(s->rdft, rdft_buf);
608
609 for (k = 0; k < s->rdft_len; k++) {
610 if (isnan(rdft_buf[k]) || isinf(rdft_buf[k])) {
611 av_log(ctx, AV_LOG_ERROR, "filter kernel contains nan or infinity.\n");
612 av_expr_free(gain_expr);
613 if (dump_fp)
614 fclose(dump_fp);
615 return AVERROR(EINVAL);
616 }
617 }
618
619 rdft_buf[s->rdft_len-1] = rdft_buf[1];
620 for (k = 0; k < s->rdft_len/2; k++)
621 rdft_buf[k] = rdft_buf[2*k];
622 rdft_buf[s->rdft_len/2] = rdft_buf[s->rdft_len-1];
623
624 if (dump_fp)
625 dump_fir(ctx, dump_fp, ch);
626
627 if (!s->multi)
628 break;
629 }
630
631 memcpy(s->kernel_buf, s->kernel_tmp_buf, (s->multi ? inlink->channels : 1) * s->rdft_len * sizeof(*s->kernel_buf));
632 av_expr_free(gain_expr);
633 if (dump_fp)
634 fclose(dump_fp);
635 return 0;
636}
637
638#define SELECT_GAIN(s) (s->gain_cmd ? s->gain_cmd : s->gain)
639#define SELECT_GAIN_ENTRY(s) (s->gain_entry_cmd ? s->gain_entry_cmd : s->gain_entry)
640
641static int config_input(AVFilterLink *inlink)
642{
643 AVFilterContext *ctx = inlink->dst;
644 FIREqualizerContext *s = ctx->priv;
645 int rdft_bits;
646
647 common_uninit(s);
648
649 s->next_pts = 0;
650 s->frame_nsamples_max = 0;
651
652 s->fir_len = FFMAX(2 * (int)(inlink->sample_rate * s->delay) + 1, 3);
653 s->remaining = s->fir_len - 1;
654
655 for (rdft_bits = RDFT_BITS_MIN; rdft_bits <= RDFT_BITS_MAX; rdft_bits++) {
656 s->rdft_len = 1 << rdft_bits;
657 s->nsamples_max = s->rdft_len - s->fir_len + 1;
658 if (s->nsamples_max * 2 >= s->fir_len)
659 break;
660 }
661
662 if (rdft_bits > RDFT_BITS_MAX) {
663 av_log(ctx, AV_LOG_ERROR, "too large delay, please decrease it.\n");
664 return AVERROR(EINVAL);
665 }
666
667 if (!(s->rdft = av_rdft_init(rdft_bits, DFT_R2C)) || !(s->irdft = av_rdft_init(rdft_bits, IDFT_C2R)))
668 return AVERROR(ENOMEM);
669
670 if (s->fft2 && !s->multi && inlink->channels > 1 && !(s->fft_ctx = av_fft_init(rdft_bits, 0)))
671 return AVERROR(ENOMEM);
672
673 for ( ; rdft_bits <= RDFT_BITS_MAX; rdft_bits++) {
674 s->analysis_rdft_len = 1 << rdft_bits;
675 if (inlink->sample_rate <= s->accuracy * s->analysis_rdft_len)
676 break;
677 }
678
679 if (rdft_bits > RDFT_BITS_MAX) {
680 av_log(ctx, AV_LOG_ERROR, "too small accuracy, please increase it.\n");
681 return AVERROR(EINVAL);
682 }
683
684 if (!(s->analysis_irdft = av_rdft_init(rdft_bits, IDFT_C2R)))
685 return AVERROR(ENOMEM);
686
687 if (s->dumpfile) {
688 s->analysis_rdft = av_rdft_init(rdft_bits, DFT_R2C);
689 s->dump_buf = av_malloc_array(s->analysis_rdft_len, sizeof(*s->dump_buf));
690 }
691
692 s->analysis_buf = av_malloc_array(s->analysis_rdft_len, sizeof(*s->analysis_buf));
693 s->kernel_tmp_buf = av_malloc_array(s->rdft_len * (s->multi ? inlink->channels : 1), sizeof(*s->kernel_tmp_buf));
694 s->kernel_buf = av_malloc_array(s->rdft_len * (s->multi ? inlink->channels : 1), sizeof(*s->kernel_buf));
695 s->conv_buf = av_calloc(2 * s->rdft_len * inlink->channels, sizeof(*s->conv_buf));
696 s->conv_idx = av_calloc(inlink->channels, sizeof(*s->conv_idx));
697 if (!s->analysis_buf || !s->kernel_tmp_buf || !s->kernel_buf || !s->conv_buf || !s->conv_idx)
698 return AVERROR(ENOMEM);
699
700 av_log(ctx, AV_LOG_DEBUG, "sample_rate = %d, channels = %d, analysis_rdft_len = %d, rdft_len = %d, fir_len = %d, nsamples_max = %d.\n",
701 inlink->sample_rate, inlink->channels, s->analysis_rdft_len, s->rdft_len, s->fir_len, s->nsamples_max);
702
703 if (s->fixed)
704 inlink->min_samples = inlink->max_samples = inlink->partial_buf_size = s->nsamples_max;
705
706 return generate_kernel(ctx, SELECT_GAIN(s), SELECT_GAIN_ENTRY(s));
707}
708
709static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
710{
711 AVFilterContext *ctx = inlink->dst;
712 FIREqualizerContext *s = ctx->priv;
713 int ch;
714
715 for (ch = 0; ch + 1 < inlink->channels && s->fft_ctx; ch += 2) {
716 fast_convolute2(s, s->kernel_buf, (FFTComplex *)(s->conv_buf + 2 * ch * s->rdft_len),
717 s->conv_idx + ch, (float *) frame->extended_data[ch],
718 (float *) frame->extended_data[ch+1], frame->nb_samples);
719 }
720
721 for ( ; ch < inlink->channels; ch++) {
722 fast_convolute(s, s->kernel_buf + (s->multi ? ch * s->rdft_len : 0),
723 s->conv_buf + 2 * ch * s->rdft_len, s->conv_idx + ch,
724 (float *) frame->extended_data[ch], frame->nb_samples);
725 }
726
727 s->next_pts = AV_NOPTS_VALUE;
728 if (frame->pts != AV_NOPTS_VALUE) {
729 s->next_pts = frame->pts + av_rescale_q(frame->nb_samples, av_make_q(1, inlink->sample_rate), inlink->time_base);
730 if (s->zero_phase)
731 frame->pts -= av_rescale_q(s->fir_len/2, av_make_q(1, inlink->sample_rate), inlink->time_base);
732 }
733 s->frame_nsamples_max = FFMAX(s->frame_nsamples_max, frame->nb_samples);
734 return ff_filter_frame(ctx->outputs[0], frame);
735}
736
737static int request_frame(AVFilterLink *outlink)
738{
739 AVFilterContext *ctx = outlink->src;
740 FIREqualizerContext *s= ctx->priv;
741 int ret;
742
743 ret = ff_request_frame(ctx->inputs[0]);
744 if (ret == AVERROR_EOF && s->remaining > 0 && s->frame_nsamples_max > 0) {
745 AVFrame *frame = ff_get_audio_buffer(outlink, FFMIN(s->remaining, s->frame_nsamples_max));
746
747 if (!frame)
748 return AVERROR(ENOMEM);
749
750 av_samples_set_silence(frame->extended_data, 0, frame->nb_samples, outlink->channels, frame->format);
751 frame->pts = s->next_pts;
752 s->remaining -= frame->nb_samples;
753 ret = filter_frame(ctx->inputs[0], frame);
754 }
755
756 return ret;
757}
758
759static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
760 char *res, int res_len, int flags)
761{
762 FIREqualizerContext *s = ctx->priv;
763 int ret = AVERROR(ENOSYS);
764
765 if (!strcmp(cmd, "gain")) {
766 char *gain_cmd;
767
768 if (SELECT_GAIN(s) && !strcmp(SELECT_GAIN(s), args)) {
769 av_log(ctx, AV_LOG_DEBUG, "equal gain, do not rebuild.\n");
770 return 0;
771 }
772
773 gain_cmd = av_strdup(args);
774 if (!gain_cmd)
775 return AVERROR(ENOMEM);
776
777 ret = generate_kernel(ctx, gain_cmd, SELECT_GAIN_ENTRY(s));
778 if (ret >= 0) {
779 av_freep(&s->gain_cmd);
780 s->gain_cmd = gain_cmd;
781 } else {
782 av_freep(&gain_cmd);
783 }
784 } else if (!strcmp(cmd, "gain_entry")) {
785 char *gain_entry_cmd;
786
787 if (SELECT_GAIN_ENTRY(s) && !strcmp(SELECT_GAIN_ENTRY(s), args)) {
788 av_log(ctx, AV_LOG_DEBUG, "equal gain_entry, do not rebuild.\n");
789 return 0;
790 }
791
792 gain_entry_cmd = av_strdup(args);
793 if (!gain_entry_cmd)
794 return AVERROR(ENOMEM);
795
796 ret = generate_kernel(ctx, SELECT_GAIN(s), gain_entry_cmd);
797 if (ret >= 0) {
798 av_freep(&s->gain_entry_cmd);
799 s->gain_entry_cmd = gain_entry_cmd;
800 } else {
801 av_freep(&gain_entry_cmd);
802 }
803 }
804
805 return ret;
806}
807
808static const AVFilterPad firequalizer_inputs[] = {
809 {
810 .name = "default",
811 .config_props = config_input,
812 .filter_frame = filter_frame,
813 .type = AVMEDIA_TYPE_AUDIO,
814 .needs_writable = 1,
815 },
816 { NULL }
817};
818
819static const AVFilterPad firequalizer_outputs[] = {
820 {
821 .name = "default",
822 .request_frame = request_frame,
823 .type = AVMEDIA_TYPE_AUDIO,
824 },
825 { NULL }
826};
827
828AVFilter ff_af_firequalizer = {
829 .name = "firequalizer",
830 .description = NULL_IF_CONFIG_SMALL("Finite Impulse Response Equalizer."),
831 .uninit = uninit,
832 .query_formats = query_formats,
833 .process_command = process_command,
834 .priv_size = sizeof(FIREqualizerContext),
835 .inputs = firequalizer_inputs,
836 .outputs = firequalizer_outputs,
837 .priv_class = &firequalizer_class,
838};
839
generated by cgit at 2024-05-17 02:03:13 (GMT)