2 * Copyright (c) 2016 Paul B Mahol
4 * This file is part of FFmpeg.
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.
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.
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
23 * SpectrumSynth filter
24 * @todo support float pixel format
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"
39 #include "window_func.h"
41 enum MagnitudeScale { LINEAR, LOG, NB_SCALES };
42 enum SlideMode { REPLACE, SCROLL, FULLFRAME, RSCROLL, NB_SLIDES };
43 enum Orientation { VERTICAL, HORIZONTAL, NB_ORIENTATIONS };
45 typedef struct SpectrumSynthContext {
55 AVFrame *magnitude, *phase;
56 FFTContext *fft; ///< Fast Fourier Transform context
57 int fft_bits; ///< number of bits (FFT window size = 1<<fft_bits)
58 FFTComplex **fft_data; ///< bins holder for each (displayed) channels
69 float *window_func_lut; ///< Window function LUT
70 } SpectrumSynthContext;
72 #define OFFSET(x) offsetof(SpectrumSynthContext, x)
73 #define A AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_AUDIO_PARAM
74 #define V AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
76 static const AVOption spectrumsynth_options[] = {
77 { "sample_rate", "set sample rate", OFFSET(sample_rate), AV_OPT_TYPE_INT, {.i64 = 44100}, 15, INT_MAX, A },
78 { "channels", "set channels", OFFSET(channels), AV_OPT_TYPE_INT, {.i64 = 1}, 1, 8, A },
79 { "scale", "set input amplitude scale", OFFSET(scale), AV_OPT_TYPE_INT, {.i64 = LOG}, 0, NB_SCALES-1, V, "scale" },
80 { "lin", "linear", 0, AV_OPT_TYPE_CONST, {.i64=LINEAR}, 0, 0, V, "scale" },
81 { "log", "logarithmic", 0, AV_OPT_TYPE_CONST, {.i64=LOG}, 0, 0, V, "scale" },
82 { "slide", "set input sliding mode", OFFSET(sliding), AV_OPT_TYPE_INT, {.i64 = FULLFRAME}, 0, NB_SLIDES-1, V, "slide" },
83 { "replace", "consume old columns with new", 0, AV_OPT_TYPE_CONST, {.i64=REPLACE}, 0, 0, V, "slide" },
84 { "scroll", "consume only most right column", 0, AV_OPT_TYPE_CONST, {.i64=SCROLL}, 0, 0, V, "slide" },
85 { "fullframe", "consume full frames", 0, AV_OPT_TYPE_CONST, {.i64=FULLFRAME}, 0, 0, V, "slide" },
86 { "rscroll", "consume only most left column", 0, AV_OPT_TYPE_CONST, {.i64=RSCROLL}, 0, 0, V, "slide" },
87 { "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64 = 0}, 0, NB_WFUNC-1, A, "win_func" },
88 { "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, A, "win_func" },
89 { "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, A, "win_func" },
90 { "hann", "Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
91 { "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
92 { "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, A, "win_func" },
93 { "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, A, "win_func" },
94 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, A },
95 { "orientation", "set orientation", OFFSET(orientation), AV_OPT_TYPE_INT, {.i64=VERTICAL}, 0, NB_ORIENTATIONS-1, V, "orientation" },
96 { "vertical", NULL, 0, AV_OPT_TYPE_CONST, {.i64=VERTICAL}, 0, 0, V, "orientation" },
97 { "horizontal", NULL, 0, AV_OPT_TYPE_CONST, {.i64=HORIZONTAL}, 0, 0, V, "orientation" },
101 AVFILTER_DEFINE_CLASS(spectrumsynth);
103 static int query_formats(AVFilterContext *ctx)
105 SpectrumSynthContext *s = ctx->priv;
106 AVFilterFormats *formats = NULL;
107 AVFilterChannelLayouts *layout = NULL;
108 AVFilterLink *magnitude = ctx->inputs[0];
109 AVFilterLink *phase = ctx->inputs[1];
110 AVFilterLink *outlink = ctx->outputs[0];
111 static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE };
112 static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY16,
113 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVJ444P,
114 AV_PIX_FMT_YUV444P16, AV_PIX_FMT_NONE };
115 int ret, sample_rates[] = { 48000, -1 };
117 formats = ff_make_format_list(sample_fmts);
118 if ((ret = ff_formats_ref (formats, &outlink->incfg.formats )) < 0 ||
119 (ret = ff_add_channel_layout (&layout, FF_COUNT2LAYOUT(s->channels))) < 0 ||
120 (ret = ff_channel_layouts_ref (layout , &outlink->incfg.channel_layouts)) < 0)
123 sample_rates[0] = s->sample_rate;
124 formats = ff_make_format_list(sample_rates);
126 return AVERROR(ENOMEM);
127 if ((ret = ff_formats_ref(formats, &outlink->incfg.samplerates)) < 0)
130 formats = ff_make_format_list(pix_fmts);
132 return AVERROR(ENOMEM);
133 if ((ret = ff_formats_ref(formats, &magnitude->outcfg.formats)) < 0)
136 formats = ff_make_format_list(pix_fmts);
138 return AVERROR(ENOMEM);
139 if ((ret = ff_formats_ref(formats, &phase->outcfg.formats)) < 0)
145 static int config_output(AVFilterLink *outlink)
147 AVFilterContext *ctx = outlink->src;
148 SpectrumSynthContext *s = ctx->priv;
149 int width = ctx->inputs[0]->w;
150 int height = ctx->inputs[0]->h;
151 AVRational time_base = ctx->inputs[0]->time_base;
152 AVRational frame_rate = ctx->inputs[0]->frame_rate;
154 float factor, overlap;
156 outlink->sample_rate = s->sample_rate;
157 outlink->time_base = (AVRational){1, s->sample_rate};
159 if (width != ctx->inputs[1]->w ||
160 height != ctx->inputs[1]->h) {
161 av_log(ctx, AV_LOG_ERROR,
162 "Magnitude and Phase sizes differ (%dx%d vs %dx%d).\n",
164 ctx->inputs[1]->w, ctx->inputs[1]->h);
165 return AVERROR_INVALIDDATA;
166 } else if (av_cmp_q(time_base, ctx->inputs[1]->time_base) != 0) {
167 av_log(ctx, AV_LOG_ERROR,
168 "Magnitude and Phase time bases differ (%d/%d vs %d/%d).\n",
169 time_base.num, time_base.den,
170 ctx->inputs[1]->time_base.num,
171 ctx->inputs[1]->time_base.den);
172 return AVERROR_INVALIDDATA;
173 } else if (av_cmp_q(frame_rate, ctx->inputs[1]->frame_rate) != 0) {
174 av_log(ctx, AV_LOG_ERROR,
175 "Magnitude and Phase framerates differ (%d/%d vs %d/%d).\n",
176 frame_rate.num, frame_rate.den,
177 ctx->inputs[1]->frame_rate.num,
178 ctx->inputs[1]->frame_rate.den);
179 return AVERROR_INVALIDDATA;
182 s->size = s->orientation == VERTICAL ? height / s->channels : width / s->channels;
183 s->xend = s->orientation == VERTICAL ? width : height;
185 for (fft_bits = 1; 1 << fft_bits < 2 * s->size; fft_bits++);
187 s->win_size = 1 << fft_bits;
188 s->nb_freq = 1 << (fft_bits - 1);
190 s->fft = av_fft_init(fft_bits, 1);
192 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT context. "
193 "The window size might be too high.\n");
194 return AVERROR(EINVAL);
196 s->fft_data = av_calloc(s->channels, sizeof(*s->fft_data));
198 return AVERROR(ENOMEM);
199 for (ch = 0; ch < s->channels; ch++) {
200 s->fft_data[ch] = av_calloc(s->win_size, sizeof(**s->fft_data));
201 if (!s->fft_data[ch])
202 return AVERROR(ENOMEM);
205 s->buffer = ff_get_audio_buffer(outlink, s->win_size * 2);
207 return AVERROR(ENOMEM);
209 /* pre-calc windowing function */
210 s->window_func_lut = av_realloc_f(s->window_func_lut, s->win_size,
211 sizeof(*s->window_func_lut));
212 if (!s->window_func_lut)
213 return AVERROR(ENOMEM);
214 generate_window_func(s->window_func_lut, s->win_size, s->win_func, &overlap);
216 s->overlap = overlap;
217 s->hop_size = (1 - s->overlap) * s->win_size;
218 for (factor = 0, i = 0; i < s->win_size; i++) {
219 factor += s->window_func_lut[i] * s->window_func_lut[i];
221 s->factor = (factor / s->win_size) / FFMAX(1 / (1 - s->overlap) - 1, 1);
226 static void read16_fft_bin(SpectrumSynthContext *s,
227 int x, int y, int f, int ch)
229 const int m_linesize = s->magnitude->linesize[0];
230 const int p_linesize = s->phase->linesize[0];
231 const uint16_t *m = (uint16_t *)(s->magnitude->data[0] + y * m_linesize);
232 const uint16_t *p = (uint16_t *)(s->phase->data[0] + y * p_linesize);
233 float magnitude, phase;
237 magnitude = m[x] / (double)UINT16_MAX;
240 magnitude = ff_exp10(((m[x] / (double)UINT16_MAX) - 1.) * 6.);
245 phase = ((p[x] / (double)UINT16_MAX) * 2. - 1.) * M_PI;
247 s->fft_data[ch][f].re = magnitude * cos(phase);
248 s->fft_data[ch][f].im = magnitude * sin(phase);
251 static void read8_fft_bin(SpectrumSynthContext *s,
252 int x, int y, int f, int ch)
254 const int m_linesize = s->magnitude->linesize[0];
255 const int p_linesize = s->phase->linesize[0];
256 const uint8_t *m = (uint8_t *)(s->magnitude->data[0] + y * m_linesize);
257 const uint8_t *p = (uint8_t *)(s->phase->data[0] + y * p_linesize);
258 float magnitude, phase;
262 magnitude = m[x] / (double)UINT8_MAX;
265 magnitude = ff_exp10(((m[x] / (double)UINT8_MAX) - 1.) * 6.);
270 phase = ((p[x] / (double)UINT8_MAX) * 2. - 1.) * M_PI;
272 s->fft_data[ch][f].re = magnitude * cos(phase);
273 s->fft_data[ch][f].im = magnitude * sin(phase);
276 static void read_fft_data(AVFilterContext *ctx, int x, int h, int ch)
278 SpectrumSynthContext *s = ctx->priv;
279 AVFilterLink *inlink = ctx->inputs[0];
280 int start = h * (s->channels - ch) - 1;
281 int end = h * (s->channels - ch - 1);
284 switch (s->orientation) {
286 switch (inlink->format) {
287 case AV_PIX_FMT_YUV444P16:
288 case AV_PIX_FMT_GRAY16:
289 for (y = start, f = 0; y >= end; y--, f++) {
290 read16_fft_bin(s, x, y, f, ch);
293 case AV_PIX_FMT_YUVJ444P:
294 case AV_PIX_FMT_YUV444P:
295 case AV_PIX_FMT_GRAY8:
296 for (y = start, f = 0; y >= end; y--, f++) {
297 read8_fft_bin(s, x, y, f, ch);
303 switch (inlink->format) {
304 case AV_PIX_FMT_YUV444P16:
305 case AV_PIX_FMT_GRAY16:
306 for (y = end, f = 0; y <= start; y++, f++) {
307 read16_fft_bin(s, y, x, f, ch);
310 case AV_PIX_FMT_YUVJ444P:
311 case AV_PIX_FMT_YUV444P:
312 case AV_PIX_FMT_GRAY8:
313 for (y = end, f = 0; y <= start; y++, f++) {
314 read8_fft_bin(s, y, x, f, ch);
322 static void synth_window(AVFilterContext *ctx, int x)
324 SpectrumSynthContext *s = ctx->priv;
325 const int h = s->size;
326 int nb = s->win_size;
329 for (ch = 0; ch < s->channels; ch++) {
330 read_fft_data(ctx, x, h, ch);
332 for (y = h; y <= s->nb_freq; y++) {
333 s->fft_data[ch][y].re = 0;
334 s->fft_data[ch][y].im = 0;
337 for (y = s->nb_freq + 1, f = s->nb_freq - 1; y < nb; y++, f--) {
338 s->fft_data[ch][y].re = s->fft_data[ch][f].re;
339 s->fft_data[ch][y].im = -s->fft_data[ch][f].im;
342 av_fft_permute(s->fft, s->fft_data[ch]);
343 av_fft_calc(s->fft, s->fft_data[ch]);
347 static int try_push_frame(AVFilterContext *ctx, int x)
349 SpectrumSynthContext *s = ctx->priv;
350 AVFilterLink *outlink = ctx->outputs[0];
351 const float factor = s->factor;
356 synth_window(ctx, x);
358 for (ch = 0; ch < s->channels; ch++) {
359 float *buf = (float *)s->buffer->extended_data[ch];
365 for (i = 0, j = start; j < k && i < s->win_size; i++, j++) {
366 buf[j] += s->fft_data[ch][i].re;
369 for (; i < s->win_size; i++, j++) {
370 buf[j] = s->fft_data[ch][i].re;
373 start += s->hop_size;
376 if (start >= s->win_size) {
377 start -= s->win_size;
380 if (ch == s->channels - 1) {
384 out = ff_get_audio_buffer(outlink, s->win_size);
386 av_frame_free(&s->magnitude);
387 av_frame_free(&s->phase);
388 return AVERROR(ENOMEM);
392 s->pts += s->win_size;
393 for (c = 0; c < s->channels; c++) {
394 dst = (float *)out->extended_data[c];
395 buf = (float *)s->buffer->extended_data[c];
397 for (n = 0; n < s->win_size; n++) {
398 dst[n] = buf[n] * factor;
400 memmove(buf, buf + s->win_size, s->win_size * 4);
403 ret = ff_filter_frame(outlink, out);
416 static int try_push_frames(AVFilterContext *ctx)
418 SpectrumSynthContext *s = ctx->priv;
421 if (!(s->magnitude && s->phase))
424 switch (s->sliding) {
426 ret = try_push_frame(ctx, s->xpos);
428 if (s->xpos >= s->xend)
432 s->xpos = s->xend - 1;
433 ret = try_push_frame(ctx, s->xpos);
437 ret = try_push_frame(ctx, s->xpos);
440 for (x = 0; x < s->xend; x++) {
441 ret = try_push_frame(ctx, x);
450 av_frame_free(&s->magnitude);
451 av_frame_free(&s->phase);
455 static int activate(AVFilterContext *ctx)
457 SpectrumSynthContext *s = ctx->priv;
458 AVFrame **staging[2] = { &s->magnitude, &s->phase };
462 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
464 for (i = 0; i < 2; i++) {
467 ret = ff_inlink_consume_frame(ctx->inputs[i], staging[i]);
471 ff_filter_set_ready(ctx, 10);
472 return try_push_frames(ctx);
476 for (i = 0; i < 2; i++) {
477 if (ff_inlink_acknowledge_status(ctx->inputs[i], &ret, &pts)) {
478 ff_outlink_set_status(ctx->outputs[0], ret, pts);
479 ff_inlink_set_status(ctx->inputs[1 - i], ret);
484 if (ff_outlink_frame_wanted(ctx->outputs[0])) {
485 for (i = 0; i < 2; i++) {
487 ff_inlink_request_frame(ctx->inputs[i]);
491 return FFERROR_NOT_READY;
494 static av_cold void uninit(AVFilterContext *ctx)
496 SpectrumSynthContext *s = ctx->priv;
499 av_frame_free(&s->magnitude);
500 av_frame_free(&s->phase);
501 av_frame_free(&s->buffer);
504 for (i = 0; i < s->channels; i++)
505 av_freep(&s->fft_data[i]);
507 av_freep(&s->fft_data);
508 av_freep(&s->window_func_lut);
511 static const AVFilterPad spectrumsynth_inputs[] = {
514 .type = AVMEDIA_TYPE_VIDEO,
518 .type = AVMEDIA_TYPE_VIDEO,
523 static const AVFilterPad spectrumsynth_outputs[] = {
526 .type = AVMEDIA_TYPE_AUDIO,
527 .config_props = config_output,
532 AVFilter ff_vaf_spectrumsynth = {
533 .name = "spectrumsynth",
534 .description = NULL_IF_CONFIG_SMALL("Convert input spectrum videos to audio output."),
536 .query_formats = query_formats,
537 .activate = activate,
538 .priv_size = sizeof(SpectrumSynthContext),
539 .inputs = spectrumsynth_inputs,
540 .outputs = spectrumsynth_outputs,
541 .priv_class = &spectrumsynth_class,