summaryrefslogtreecommitdiff
path: root/libavfilter/vaf_spectrumsynth.c (plain)
blob: ed191a3e34b9654e9bd526d5f62a5dc7eb070cda
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
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/**
22 * @file
23 * SpectrumSynth filter
24 * @todo support float pixel format
25 */
26
27#include "libavcodec/avfft.h"
28#include "libavutil/avassert.h"
29#include "libavutil/channel_layout.h"
30#include "libavutil/ffmath.h"
31#include "libavutil/opt.h"
32#include "libavutil/parseutils.h"
33#include "avfilter.h"
34#include "formats.h"
35#include "audio.h"
36#include "video.h"
37#include "internal.h"
38#include "window_func.h"
39
40enum MagnitudeScale { LINEAR, LOG, NB_SCALES };
41enum SlideMode { REPLACE, SCROLL, FULLFRAME, RSCROLL, NB_SLIDES };
42enum Orientation { VERTICAL, HORIZONTAL, NB_ORIENTATIONS };
43
44typedef struct SpectrumSynthContext {
45 const AVClass *class;
46 int sample_rate;
47 int channels;
48 int scale;
49 int sliding;
50 int win_func;
51 float overlap;
52 int orientation;
53
54 AVFrame *magnitude, *phase;
55 FFTContext *fft; ///< Fast Fourier Transform context
56 int fft_bits; ///< number of bits (FFT window size = 1<<fft_bits)
57 FFTComplex **fft_data; ///< bins holder for each (displayed) channels
58 int win_size;
59 int size;
60 int nb_freq;
61 int hop_size;
62 int start, end;
63 int xpos;
64 int xend;
65 int64_t pts;
66 float factor;
67 AVFrame *buffer;
68 float *window_func_lut; ///< Window function LUT
69} SpectrumSynthContext;
70
71#define OFFSET(x) offsetof(SpectrumSynthContext, x)
72#define A AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_AUDIO_PARAM
73#define V AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
74
75static const AVOption spectrumsynth_options[] = {
76 { "sample_rate", "set sample rate", OFFSET(sample_rate), AV_OPT_TYPE_INT, {.i64 = 44100}, 15, INT_MAX, A },
77 { "channels", "set channels", OFFSET(channels), AV_OPT_TYPE_INT, {.i64 = 1}, 1, 8, A },
78 { "scale", "set input amplitude scale", OFFSET(scale), AV_OPT_TYPE_INT, {.i64 = LOG}, 0, NB_SCALES-1, V, "scale" },
79 { "lin", "linear", 0, AV_OPT_TYPE_CONST, {.i64=LINEAR}, 0, 0, V, "scale" },
80 { "log", "logarithmic", 0, AV_OPT_TYPE_CONST, {.i64=LOG}, 0, 0, V, "scale" },
81 { "slide", "set input sliding mode", OFFSET(sliding), AV_OPT_TYPE_INT, {.i64 = FULLFRAME}, 0, NB_SLIDES-1, V, "slide" },
82 { "replace", "consume old columns with new", 0, AV_OPT_TYPE_CONST, {.i64=REPLACE}, 0, 0, V, "slide" },
83 { "scroll", "consume only most right column", 0, AV_OPT_TYPE_CONST, {.i64=SCROLL}, 0, 0, V, "slide" },
84 { "fullframe", "consume full frames", 0, AV_OPT_TYPE_CONST, {.i64=FULLFRAME}, 0, 0, V, "slide" },
85 { "rscroll", "consume only most left column", 0, AV_OPT_TYPE_CONST, {.i64=RSCROLL}, 0, 0, V, "slide" },
86 { "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64 = 0}, 0, NB_WFUNC-1, A, "win_func" },
87 { "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, A, "win_func" },
88 { "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, A, "win_func" },
89 { "hann", "Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
90 { "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
91 { "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, A, "win_func" },
92 { "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, A, "win_func" },
93 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, A },
94 { "orientation", "set orientation", OFFSET(orientation), AV_OPT_TYPE_INT, {.i64=VERTICAL}, 0, NB_ORIENTATIONS-1, V, "orientation" },
95 { "vertical", NULL, 0, AV_OPT_TYPE_CONST, {.i64=VERTICAL}, 0, 0, V, "orientation" },
96 { "horizontal", NULL, 0, AV_OPT_TYPE_CONST, {.i64=HORIZONTAL}, 0, 0, V, "orientation" },
97 { NULL }
98};
99
100AVFILTER_DEFINE_CLASS(spectrumsynth);
101
102static int query_formats(AVFilterContext *ctx)
103{
104 SpectrumSynthContext *s = ctx->priv;
105 AVFilterFormats *formats = NULL;
106 AVFilterChannelLayouts *layout = NULL;
107 AVFilterLink *magnitude = ctx->inputs[0];
108 AVFilterLink *phase = ctx->inputs[1];
109 AVFilterLink *outlink = ctx->outputs[0];
110 static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE };
111 static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY16,
112 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVJ444P,
113 AV_PIX_FMT_YUV444P16, AV_PIX_FMT_NONE };
114 int ret, sample_rates[] = { 48000, -1 };
115
116 formats = ff_make_format_list(sample_fmts);
117 if ((ret = ff_formats_ref (formats, &outlink->in_formats )) < 0 ||
118 (ret = ff_add_channel_layout (&layout, FF_COUNT2LAYOUT(s->channels))) < 0 ||
119 (ret = ff_channel_layouts_ref (layout , &outlink->in_channel_layouts)) < 0)
120 return ret;
121
122 sample_rates[0] = s->sample_rate;
123 formats = ff_make_format_list(sample_rates);
124 if (!formats)
125 return AVERROR(ENOMEM);
126 if ((ret = ff_formats_ref(formats, &outlink->in_samplerates)) < 0)
127 return ret;
128
129 formats = ff_make_format_list(pix_fmts);
130 if (!formats)
131 return AVERROR(ENOMEM);
132 if ((ret = ff_formats_ref(formats, &magnitude->out_formats)) < 0)
133 return ret;
134
135 formats = ff_make_format_list(pix_fmts);
136 if (!formats)
137 return AVERROR(ENOMEM);
138 if ((ret = ff_formats_ref(formats, &phase->out_formats)) < 0)
139 return ret;
140
141 return 0;
142}
143
144static int config_output(AVFilterLink *outlink)
145{
146 AVFilterContext *ctx = outlink->src;
147 SpectrumSynthContext *s = ctx->priv;
148 int width = ctx->inputs[0]->w;
149 int height = ctx->inputs[0]->h;
150 AVRational time_base = ctx->inputs[0]->time_base;
151 AVRational frame_rate = ctx->inputs[0]->frame_rate;
152 int i, ch, fft_bits;
153 float factor, overlap;
154
155 outlink->sample_rate = s->sample_rate;
156 outlink->time_base = (AVRational){1, s->sample_rate};
157
158 if (width != ctx->inputs[1]->w ||
159 height != ctx->inputs[1]->h) {
160 av_log(ctx, AV_LOG_ERROR,
161 "Magnitude and Phase sizes differ (%dx%d vs %dx%d).\n",
162 width, height,
163 ctx->inputs[1]->w, ctx->inputs[1]->h);
164 return AVERROR_INVALIDDATA;
165 } else if (av_cmp_q(time_base, ctx->inputs[1]->time_base) != 0) {
166 av_log(ctx, AV_LOG_ERROR,
167 "Magnitude and Phase time bases differ (%d/%d vs %d/%d).\n",
168 time_base.num, time_base.den,
169 ctx->inputs[1]->time_base.num,
170 ctx->inputs[1]->time_base.den);
171 return AVERROR_INVALIDDATA;
172 } else if (av_cmp_q(frame_rate, ctx->inputs[1]->frame_rate) != 0) {
173 av_log(ctx, AV_LOG_ERROR,
174 "Magnitude and Phase framerates differ (%d/%d vs %d/%d).\n",
175 frame_rate.num, frame_rate.den,
176 ctx->inputs[1]->frame_rate.num,
177 ctx->inputs[1]->frame_rate.den);
178 return AVERROR_INVALIDDATA;
179 }
180
181 s->size = s->orientation == VERTICAL ? height / s->channels : width / s->channels;
182 s->xend = s->orientation == VERTICAL ? width : height;
183
184 for (fft_bits = 1; 1 << fft_bits < 2 * s->size; fft_bits++);
185
186 s->win_size = 1 << fft_bits;
187 s->nb_freq = 1 << (fft_bits - 1);
188
189 s->fft = av_fft_init(fft_bits, 1);
190 if (!s->fft) {
191 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT context. "
192 "The window size might be too high.\n");
193 return AVERROR(EINVAL);
194 }
195 s->fft_data = av_calloc(s->channels, sizeof(*s->fft_data));
196 if (!s->fft_data)
197 return AVERROR(ENOMEM);
198 for (ch = 0; ch < s->channels; ch++) {
199 s->fft_data[ch] = av_calloc(s->win_size, sizeof(**s->fft_data));
200 if (!s->fft_data[ch])
201 return AVERROR(ENOMEM);
202 }
203
204 s->buffer = ff_get_audio_buffer(outlink, s->win_size * 2);
205 if (!s->buffer)
206 return AVERROR(ENOMEM);
207
208 /* pre-calc windowing function */
209 s->window_func_lut = av_realloc_f(s->window_func_lut, s->win_size,
210 sizeof(*s->window_func_lut));
211 if (!s->window_func_lut)
212 return AVERROR(ENOMEM);
213 ff_generate_window_func(s->window_func_lut, s->win_size, s->win_func, &overlap);
214 if (s->overlap == 1)
215 s->overlap = overlap;
216 s->hop_size = (1 - s->overlap) * s->win_size;
217 for (factor = 0, i = 0; i < s->win_size; i++) {
218 factor += s->window_func_lut[i] * s->window_func_lut[i];
219 }
220 s->factor = (factor / s->win_size) / FFMAX(1 / (1 - s->overlap) - 1, 1);
221
222 return 0;
223}
224
225static int request_frame(AVFilterLink *outlink)
226{
227 AVFilterContext *ctx = outlink->src;
228 SpectrumSynthContext *s = ctx->priv;
229 int ret;
230
231 if (!s->magnitude) {
232 ret = ff_request_frame(ctx->inputs[0]);
233 if (ret < 0)
234 return ret;
235 }
236 if (!s->phase) {
237 ret = ff_request_frame(ctx->inputs[1]);
238 if (ret < 0)
239 return ret;
240 }
241 return 0;
242}
243
244static void read16_fft_bin(SpectrumSynthContext *s,
245 int x, int y, int f, int ch)
246{
247 const int m_linesize = s->magnitude->linesize[0];
248 const int p_linesize = s->phase->linesize[0];
249 const uint16_t *m = (uint16_t *)(s->magnitude->data[0] + y * m_linesize);
250 const uint16_t *p = (uint16_t *)(s->phase->data[0] + y * p_linesize);
251 float magnitude, phase;
252
253 switch (s->scale) {
254 case LINEAR:
255 magnitude = m[x] / (double)UINT16_MAX;
256 break;
257 case LOG:
258 magnitude = ff_exp10(((m[x] / (double)UINT16_MAX) - 1.) * 6.);
259 break;
260 default:
261 av_assert0(0);
262 }
263 phase = ((p[x] / (double)UINT16_MAX) * 2. - 1.) * M_PI;
264
265 s->fft_data[ch][f].re = magnitude * cos(phase);
266 s->fft_data[ch][f].im = magnitude * sin(phase);
267}
268
269static void read8_fft_bin(SpectrumSynthContext *s,
270 int x, int y, int f, int ch)
271{
272 const int m_linesize = s->magnitude->linesize[0];
273 const int p_linesize = s->phase->linesize[0];
274 const uint8_t *m = (uint8_t *)(s->magnitude->data[0] + y * m_linesize);
275 const uint8_t *p = (uint8_t *)(s->phase->data[0] + y * p_linesize);
276 float magnitude, phase;
277
278 switch (s->scale) {
279 case LINEAR:
280 magnitude = m[x] / (double)UINT8_MAX;
281 break;
282 case LOG:
283 magnitude = ff_exp10(((m[x] / (double)UINT8_MAX) - 1.) * 6.);
284 break;
285 default:
286 av_assert0(0);
287 }
288 phase = ((p[x] / (double)UINT8_MAX) * 2. - 1.) * M_PI;
289
290 s->fft_data[ch][f].re = magnitude * cos(phase);
291 s->fft_data[ch][f].im = magnitude * sin(phase);
292}
293
294static void read_fft_data(AVFilterContext *ctx, int x, int h, int ch)
295{
296 SpectrumSynthContext *s = ctx->priv;
297 AVFilterLink *inlink = ctx->inputs[0];
298 int start = h * (s->channels - ch) - 1;
299 int end = h * (s->channels - ch - 1);
300 int y, f;
301
302 switch (s->orientation) {
303 case VERTICAL:
304 switch (inlink->format) {
305 case AV_PIX_FMT_YUV444P16:
306 case AV_PIX_FMT_GRAY16:
307 for (y = start, f = 0; y >= end; y--, f++) {
308 read16_fft_bin(s, x, y, f, ch);
309 }
310 break;
311 case AV_PIX_FMT_YUVJ444P:
312 case AV_PIX_FMT_YUV444P:
313 case AV_PIX_FMT_GRAY8:
314 for (y = start, f = 0; y >= end; y--, f++) {
315 read8_fft_bin(s, x, y, f, ch);
316 }
317 break;
318 }
319 break;
320 case HORIZONTAL:
321 switch (inlink->format) {
322 case AV_PIX_FMT_YUV444P16:
323 case AV_PIX_FMT_GRAY16:
324 for (y = end, f = 0; y <= start; y++, f++) {
325 read16_fft_bin(s, y, x, f, ch);
326 }
327 break;
328 case AV_PIX_FMT_YUVJ444P:
329 case AV_PIX_FMT_YUV444P:
330 case AV_PIX_FMT_GRAY8:
331 for (y = end, f = 0; y <= start; y++, f++) {
332 read8_fft_bin(s, y, x, f, ch);
333 }
334 break;
335 }
336 break;
337 }
338}
339
340static void synth_window(AVFilterContext *ctx, int x)
341{
342 SpectrumSynthContext *s = ctx->priv;
343 const int h = s->size;
344 int nb = s->win_size;
345 int y, f, ch;
346
347 for (ch = 0; ch < s->channels; ch++) {
348 read_fft_data(ctx, x, h, ch);
349
350 for (y = h; y <= s->nb_freq; y++) {
351 s->fft_data[ch][y].re = 0;
352 s->fft_data[ch][y].im = 0;
353 }
354
355 for (y = s->nb_freq + 1, f = s->nb_freq - 1; y < nb; y++, f--) {
356 s->fft_data[ch][y].re = s->fft_data[ch][f].re;
357 s->fft_data[ch][y].im = -s->fft_data[ch][f].im;
358 }
359
360 av_fft_permute(s->fft, s->fft_data[ch]);
361 av_fft_calc(s->fft, s->fft_data[ch]);
362 }
363}
364
365static int try_push_frame(AVFilterContext *ctx, int x)
366{
367 SpectrumSynthContext *s = ctx->priv;
368 AVFilterLink *outlink = ctx->outputs[0];
369 const float factor = s->factor;
370 int ch, n, i, ret;
371 int start, end;
372 AVFrame *out;
373
374 synth_window(ctx, x);
375
376 for (ch = 0; ch < s->channels; ch++) {
377 float *buf = (float *)s->buffer->extended_data[ch];
378 int j, k;
379
380 start = s->start;
381 end = s->end;
382 k = end;
383 for (i = 0, j = start; j < k && i < s->win_size; i++, j++) {
384 buf[j] += s->fft_data[ch][i].re;
385 }
386
387 for (; i < s->win_size; i++, j++) {
388 buf[j] = s->fft_data[ch][i].re;
389 }
390
391 start += s->hop_size;
392 end = j;
393
394 if (start >= s->win_size) {
395 start -= s->win_size;
396 end -= s->win_size;
397
398 if (ch == s->channels - 1) {
399 float *dst;
400 int c;
401
402 out = ff_get_audio_buffer(outlink, s->win_size);
403 if (!out) {
404 av_frame_free(&s->magnitude);
405 av_frame_free(&s->phase);
406 return AVERROR(ENOMEM);
407 }
408
409 out->pts = s->pts;
410 s->pts += s->win_size;
411 for (c = 0; c < s->channels; c++) {
412 dst = (float *)out->extended_data[c];
413 buf = (float *)s->buffer->extended_data[c];
414
415 for (n = 0; n < s->win_size; n++) {
416 dst[n] = buf[n] * factor;
417 }
418 memmove(buf, buf + s->win_size, s->win_size * 4);
419 }
420
421 ret = ff_filter_frame(outlink, out);
422 if (ret < 0)
423 return ret;
424 }
425 }
426 }
427
428 s->start = start;
429 s->end = end;
430
431 return 0;
432}
433
434static int try_push_frames(AVFilterContext *ctx)
435{
436 SpectrumSynthContext *s = ctx->priv;
437 int ret, x;
438
439 if (!(s->magnitude && s->phase))
440 return 0;
441
442 switch (s->sliding) {
443 case REPLACE:
444 ret = try_push_frame(ctx, s->xpos);
445 s->xpos++;
446 if (s->xpos >= s->xend)
447 s->xpos = 0;
448 break;
449 case SCROLL:
450 s->xpos = s->xend - 1;
451 ret = try_push_frame(ctx, s->xpos);
452 break;
453 case RSCROLL:
454 s->xpos = 0;
455 ret = try_push_frame(ctx, s->xpos);
456 break;
457 case FULLFRAME:
458 for (x = 0; x < s->xend; x++) {
459 ret = try_push_frame(ctx, x);
460 if (ret < 0)
461 break;
462 }
463 break;
464 default:
465 av_assert0(0);
466 }
467
468 av_frame_free(&s->magnitude);
469 av_frame_free(&s->phase);
470 return ret;
471}
472
473static int filter_frame_magnitude(AVFilterLink *inlink, AVFrame *magnitude)
474{
475 AVFilterContext *ctx = inlink->dst;
476 SpectrumSynthContext *s = ctx->priv;
477
478 s->magnitude = magnitude;
479 return try_push_frames(ctx);
480}
481
482static int filter_frame_phase(AVFilterLink *inlink, AVFrame *phase)
483{
484 AVFilterContext *ctx = inlink->dst;
485 SpectrumSynthContext *s = ctx->priv;
486
487 s->phase = phase;
488 return try_push_frames(ctx);
489}
490
491static av_cold void uninit(AVFilterContext *ctx)
492{
493 SpectrumSynthContext *s = ctx->priv;
494 int i;
495
496 av_frame_free(&s->magnitude);
497 av_frame_free(&s->phase);
498 av_frame_free(&s->buffer);
499 av_fft_end(s->fft);
500 if (s->fft_data) {
501 for (i = 0; i < s->channels; i++)
502 av_freep(&s->fft_data[i]);
503 }
504 av_freep(&s->fft_data);
505 av_freep(&s->window_func_lut);
506}
507
508static const AVFilterPad spectrumsynth_inputs[] = {
509 {
510 .name = "magnitude",
511 .type = AVMEDIA_TYPE_VIDEO,
512 .filter_frame = filter_frame_magnitude,
513 .needs_fifo = 1,
514 },
515 {
516 .name = "phase",
517 .type = AVMEDIA_TYPE_VIDEO,
518 .filter_frame = filter_frame_phase,
519 .needs_fifo = 1,
520 },
521 { NULL }
522};
523
524static const AVFilterPad spectrumsynth_outputs[] = {
525 {
526 .name = "default",
527 .type = AVMEDIA_TYPE_AUDIO,
528 .config_props = config_output,
529 .request_frame = request_frame,
530 },
531 { NULL }
532};
533
534AVFilter ff_vaf_spectrumsynth = {
535 .name = "spectrumsynth",
536 .description = NULL_IF_CONFIG_SMALL("Convert input spectrum videos to audio output."),
537 .uninit = uninit,
538 .query_formats = query_formats,
539 .priv_size = sizeof(SpectrumSynthContext),
540 .inputs = spectrumsynth_inputs,
541 .outputs = spectrumsynth_outputs,
542 .priv_class = &spectrumsynth_class,
543};
544
generated by cgit at 2024-05-17 01:45:52 (GMT)