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/opt.h"
31 #include "libavutil/parseutils.h"
37 #include "window_func.h"
39 enum MagnitudeScale { LINEAR, LOG, NB_SCALES };
40 enum SlideMode { REPLACE, SCROLL, FULLFRAME, RSCROLL, NB_SLIDES };
41 enum Orientation { VERTICAL, HORIZONTAL, NB_ORIENTATIONS };
43 typedef struct SpectrumSynthContext {
53 AVFrame *magnitude, *phase;
54 FFTContext *fft; ///< Fast Fourier Transform context
55 int fft_bits; ///< number of bits (FFT window size = 1<<fft_bits)
56 FFTComplex **fft_data; ///< bins holder for each (displayed) channels
67 float *window_func_lut; ///< Window function LUT
68 } SpectrumSynthContext;
70 #define OFFSET(x) offsetof(SpectrumSynthContext, x)
71 #define A AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_AUDIO_PARAM
72 #define V AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
74 static const AVOption spectrumsynth_options[] = {
75 { "sample_rate", "set sample rate", OFFSET(sample_rate), AV_OPT_TYPE_INT, {.i64 = 44100}, 15, INT_MAX, A },
76 { "channels", "set channels", OFFSET(channels), AV_OPT_TYPE_INT, {.i64 = 1}, 1, 8, A },
77 { "scale", "set input amplitude scale", OFFSET(scale), AV_OPT_TYPE_INT, {.i64 = LOG}, 0, NB_SCALES-1, V, "scale" },
78 { "lin", "linear", 0, AV_OPT_TYPE_CONST, {.i64=LINEAR}, 0, 0, V, "scale" },
79 { "log", "logarithmic", 0, AV_OPT_TYPE_CONST, {.i64=LOG}, 0, 0, V, "scale" },
80 { "slide", "set input sliding mode", OFFSET(sliding), AV_OPT_TYPE_INT, {.i64 = FULLFRAME}, 0, NB_SLIDES-1, V, "slide" },
81 { "replace", "consume old columns with new", 0, AV_OPT_TYPE_CONST, {.i64=REPLACE}, 0, 0, V, "slide" },
82 { "scroll", "consume only most right column", 0, AV_OPT_TYPE_CONST, {.i64=SCROLL}, 0, 0, V, "slide" },
83 { "fullframe", "consume full frames", 0, AV_OPT_TYPE_CONST, {.i64=FULLFRAME}, 0, 0, V, "slide" },
84 { "rscroll", "consume only most left column", 0, AV_OPT_TYPE_CONST, {.i64=RSCROLL}, 0, 0, V, "slide" },
85 { "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64 = 0}, 0, NB_WFUNC-1, A, "win_func" },
86 { "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, A, "win_func" },
87 { "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, A, "win_func" },
88 { "hann", "Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
89 { "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
90 { "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, A, "win_func" },
91 { "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, A, "win_func" },
92 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=1}, 0, 1, A },
93 { "orientation", "set orientation", OFFSET(orientation), AV_OPT_TYPE_INT, {.i64=VERTICAL}, 0, NB_ORIENTATIONS-1, V, "orientation" },
94 { "vertical", NULL, 0, AV_OPT_TYPE_CONST, {.i64=VERTICAL}, 0, 0, V, "orientation" },
95 { "horizontal", NULL, 0, AV_OPT_TYPE_CONST, {.i64=HORIZONTAL}, 0, 0, V, "orientation" },
99 AVFILTER_DEFINE_CLASS(spectrumsynth);
101 static int query_formats(AVFilterContext *ctx)
103 SpectrumSynthContext *s = ctx->priv;
104 AVFilterFormats *formats = NULL;
105 AVFilterChannelLayouts *layout = NULL;
106 AVFilterLink *magnitude = ctx->inputs[0];
107 AVFilterLink *phase = ctx->inputs[1];
108 AVFilterLink *outlink = ctx->outputs[0];
109 static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE };
110 static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_GRAY8, AV_PIX_FMT_GRAY16,
111 AV_PIX_FMT_YUV444P, AV_PIX_FMT_YUVJ444P,
112 AV_PIX_FMT_YUV444P16, AV_PIX_FMT_NONE };
113 int ret, sample_rates[] = { 48000, -1 };
115 formats = ff_make_format_list(sample_fmts);
116 if ((ret = ff_formats_ref (formats, &outlink->in_formats )) < 0 ||
117 (ret = ff_add_channel_layout (&layout, FF_COUNT2LAYOUT(s->channels))) < 0 ||
118 (ret = ff_channel_layouts_ref (layout , &outlink->in_channel_layouts)) < 0)
121 sample_rates[0] = s->sample_rate;
122 formats = ff_make_format_list(sample_rates);
124 return AVERROR(ENOMEM);
125 if ((ret = ff_formats_ref(formats, &outlink->in_samplerates)) < 0)
128 formats = ff_make_format_list(pix_fmts);
130 return AVERROR(ENOMEM);
131 if ((ret = ff_formats_ref(formats, &magnitude->out_formats)) < 0)
134 formats = ff_make_format_list(pix_fmts);
136 return AVERROR(ENOMEM);
137 if ((ret = ff_formats_ref(formats, &phase->out_formats)) < 0)
143 static int config_output(AVFilterLink *outlink)
145 AVFilterContext *ctx = outlink->src;
146 SpectrumSynthContext *s = ctx->priv;
147 int width = ctx->inputs[0]->w;
148 int height = ctx->inputs[0]->h;
149 AVRational time_base = ctx->inputs[0]->time_base;
150 AVRational frame_rate = ctx->inputs[0]->frame_rate;
152 float factor, overlap;
154 outlink->sample_rate = s->sample_rate;
155 outlink->time_base = (AVRational){1, s->sample_rate};
157 if (width != ctx->inputs[1]->w ||
158 height != ctx->inputs[1]->h) {
159 av_log(ctx, AV_LOG_ERROR,
160 "Magnitude and Phase sizes differ (%dx%d vs %dx%d).\n",
162 ctx->inputs[1]->w, ctx->inputs[1]->h);
163 return AVERROR_INVALIDDATA;
164 } else if (av_cmp_q(time_base, ctx->inputs[1]->time_base) != 0) {
165 av_log(ctx, AV_LOG_ERROR,
166 "Magnitude and Phase time bases differ (%d/%d vs %d/%d).\n",
167 time_base.num, time_base.den,
168 ctx->inputs[1]->time_base.num,
169 ctx->inputs[1]->time_base.den);
170 return AVERROR_INVALIDDATA;
171 } else if (av_cmp_q(frame_rate, ctx->inputs[1]->frame_rate) != 0) {
172 av_log(ctx, AV_LOG_ERROR,
173 "Magnitude and Phase framerates differ (%d/%d vs %d/%d).\n",
174 frame_rate.num, frame_rate.den,
175 ctx->inputs[1]->frame_rate.num,
176 ctx->inputs[1]->frame_rate.den);
177 return AVERROR_INVALIDDATA;
180 s->size = s->orientation == VERTICAL ? height / s->channels : width / s->channels;
181 s->xend = s->orientation == VERTICAL ? width : height;
183 for (fft_bits = 1; 1 << fft_bits < 2 * s->size; fft_bits++);
185 s->win_size = 1 << fft_bits;
186 s->nb_freq = 1 << (fft_bits - 1);
188 s->fft = av_fft_init(fft_bits, 1);
190 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT context. "
191 "The window size might be too high.\n");
192 return AVERROR(EINVAL);
194 s->fft_data = av_calloc(s->channels, sizeof(*s->fft_data));
196 return AVERROR(ENOMEM);
197 for (ch = 0; ch < s->channels; ch++) {
198 s->fft_data[ch] = av_calloc(s->win_size, sizeof(**s->fft_data));
199 if (!s->fft_data[ch])
200 return AVERROR(ENOMEM);
203 s->buffer = ff_get_audio_buffer(outlink, s->win_size * 2);
205 return AVERROR(ENOMEM);
207 /* pre-calc windowing function */
208 s->window_func_lut = av_realloc_f(s->window_func_lut, s->win_size,
209 sizeof(*s->window_func_lut));
210 if (!s->window_func_lut)
211 return AVERROR(ENOMEM);
212 ff_generate_window_func(s->window_func_lut, s->win_size, s->win_func, &overlap);
214 s->overlap = overlap;
215 s->hop_size = (1 - s->overlap) * s->win_size;
216 for (factor = 0, i = 0; i < s->win_size; i++) {
217 factor += s->window_func_lut[i] * s->window_func_lut[i];
219 s->factor = (factor / s->win_size) / FFMAX(1 / (1 - s->overlap) - 1, 1);
224 static int request_frame(AVFilterLink *outlink)
226 AVFilterContext *ctx = outlink->src;
227 SpectrumSynthContext *s = ctx->priv;
231 ret = ff_request_frame(ctx->inputs[0]);
236 ret = ff_request_frame(ctx->inputs[1]);
243 static void read16_fft_bin(SpectrumSynthContext *s,
244 int x, int y, int f, int ch)
246 const int m_linesize = s->magnitude->linesize[0];
247 const int p_linesize = s->phase->linesize[0];
248 const uint16_t *m = (uint16_t *)(s->magnitude->data[0] + y * m_linesize);
249 const uint16_t *p = (uint16_t *)(s->phase->data[0] + y * p_linesize);
250 float magnitude, phase;
254 magnitude = m[x] / (double)UINT16_MAX;
257 magnitude = ff_exp10(((m[x] / (double)UINT16_MAX) - 1.) * 6.);
262 phase = ((p[x] / (double)UINT16_MAX) * 2. - 1.) * M_PI;
264 s->fft_data[ch][f].re = magnitude * cos(phase);
265 s->fft_data[ch][f].im = magnitude * sin(phase);
268 static void read8_fft_bin(SpectrumSynthContext *s,
269 int x, int y, int f, int ch)
271 const int m_linesize = s->magnitude->linesize[0];
272 const int p_linesize = s->phase->linesize[0];
273 const uint8_t *m = (uint8_t *)(s->magnitude->data[0] + y * m_linesize);
274 const uint8_t *p = (uint8_t *)(s->phase->data[0] + y * p_linesize);
275 float magnitude, phase;
279 magnitude = m[x] / (double)UINT8_MAX;
282 magnitude = ff_exp10(((m[x] / (double)UINT8_MAX) - 1.) * 6.);
287 phase = ((p[x] / (double)UINT8_MAX) * 2. - 1.) * M_PI;
289 s->fft_data[ch][f].re = magnitude * cos(phase);
290 s->fft_data[ch][f].im = magnitude * sin(phase);
293 static void read_fft_data(AVFilterContext *ctx, int x, int h, int ch)
295 SpectrumSynthContext *s = ctx->priv;
296 AVFilterLink *inlink = ctx->inputs[0];
297 int start = h * (s->channels - ch) - 1;
298 int end = h * (s->channels - ch - 1);
301 switch (s->orientation) {
303 switch (inlink->format) {
304 case AV_PIX_FMT_YUV444P16:
305 case AV_PIX_FMT_GRAY16:
306 for (y = start, f = 0; y >= end; y--, f++) {
307 read16_fft_bin(s, x, y, f, ch);
310 case AV_PIX_FMT_YUVJ444P:
311 case AV_PIX_FMT_YUV444P:
312 case AV_PIX_FMT_GRAY8:
313 for (y = start, f = 0; y >= end; y--, f++) {
314 read8_fft_bin(s, x, y, f, ch);
320 switch (inlink->format) {
321 case AV_PIX_FMT_YUV444P16:
322 case AV_PIX_FMT_GRAY16:
323 for (y = end, f = 0; y <= start; y++, f++) {
324 read16_fft_bin(s, y, x, f, ch);
327 case AV_PIX_FMT_YUVJ444P:
328 case AV_PIX_FMT_YUV444P:
329 case AV_PIX_FMT_GRAY8:
330 for (y = end, f = 0; y <= start; y++, f++) {
331 read8_fft_bin(s, y, x, f, ch);
339 static void synth_window(AVFilterContext *ctx, int x)
341 SpectrumSynthContext *s = ctx->priv;
342 const int h = s->size;
343 int nb = s->win_size;
346 for (ch = 0; ch < s->channels; ch++) {
347 read_fft_data(ctx, x, h, ch);
349 for (y = h; y <= s->nb_freq; y++) {
350 s->fft_data[ch][y].re = 0;
351 s->fft_data[ch][y].im = 0;
354 for (y = s->nb_freq + 1, f = s->nb_freq - 1; y < nb; y++, f--) {
355 s->fft_data[ch][y].re = s->fft_data[ch][f].re;
356 s->fft_data[ch][y].im = -s->fft_data[ch][f].im;
359 av_fft_permute(s->fft, s->fft_data[ch]);
360 av_fft_calc(s->fft, s->fft_data[ch]);
364 static int try_push_frame(AVFilterContext *ctx, int x)
366 SpectrumSynthContext *s = ctx->priv;
367 AVFilterLink *outlink = ctx->outputs[0];
368 const float factor = s->factor;
373 synth_window(ctx, x);
375 for (ch = 0; ch < s->channels; ch++) {
376 float *buf = (float *)s->buffer->extended_data[ch];
382 for (i = 0, j = start; j < k && i < s->win_size; i++, j++) {
383 buf[j] += s->fft_data[ch][i].re;
386 for (; i < s->win_size; i++, j++) {
387 buf[j] = s->fft_data[ch][i].re;
390 start += s->hop_size;
393 if (start >= s->win_size) {
394 start -= s->win_size;
397 if (ch == s->channels - 1) {
401 out = ff_get_audio_buffer(outlink, s->win_size);
403 av_frame_free(&s->magnitude);
404 av_frame_free(&s->phase);
405 return AVERROR(ENOMEM);
409 s->pts += s->win_size;
410 for (c = 0; c < s->channels; c++) {
411 dst = (float *)out->extended_data[c];
412 buf = (float *)s->buffer->extended_data[c];
414 for (n = 0; n < s->win_size; n++) {
415 dst[n] = buf[n] * factor;
417 memmove(buf, buf + s->win_size, s->win_size * 4);
420 ret = ff_filter_frame(outlink, out);
431 static int try_push_frames(AVFilterContext *ctx)
433 SpectrumSynthContext *s = ctx->priv;
436 if (!(s->magnitude && s->phase))
439 switch (s->sliding) {
441 ret = try_push_frame(ctx, s->xpos);
443 if (s->xpos >= s->xend)
447 s->xpos = s->xend - 1;
448 ret = try_push_frame(ctx, s->xpos);
452 ret = try_push_frame(ctx, s->xpos);
455 for (x = 0; x < s->xend; x++) {
456 ret = try_push_frame(ctx, x);
465 av_frame_free(&s->magnitude);
466 av_frame_free(&s->phase);
470 static int filter_frame_magnitude(AVFilterLink *inlink, AVFrame *magnitude)
472 AVFilterContext *ctx = inlink->dst;
473 SpectrumSynthContext *s = ctx->priv;
475 s->magnitude = magnitude;
476 return try_push_frames(ctx);
479 static int filter_frame_phase(AVFilterLink *inlink, AVFrame *phase)
481 AVFilterContext *ctx = inlink->dst;
482 SpectrumSynthContext *s = ctx->priv;
485 return try_push_frames(ctx);
488 static av_cold void uninit(AVFilterContext *ctx)
490 SpectrumSynthContext *s = ctx->priv;
493 av_frame_free(&s->magnitude);
494 av_frame_free(&s->phase);
495 av_frame_free(&s->buffer);
498 for (i = 0; i < s->channels; i++)
499 av_freep(&s->fft_data[i]);
501 av_freep(&s->fft_data);
502 av_freep(&s->window_func_lut);
505 static const AVFilterPad spectrumsynth_inputs[] = {
508 .type = AVMEDIA_TYPE_VIDEO,
509 .filter_frame = filter_frame_magnitude,
514 .type = AVMEDIA_TYPE_VIDEO,
515 .filter_frame = filter_frame_phase,
521 static const AVFilterPad spectrumsynth_outputs[] = {
524 .type = AVMEDIA_TYPE_AUDIO,
525 .config_props = config_output,
526 .request_frame = request_frame,
531 AVFilter ff_vaf_spectrumsynth = {
532 .name = "spectrumsynth",
533 .description = NULL_IF_CONFIG_SMALL("Convert input spectrum videos to audio output."),
535 .query_formats = query_formats,
536 .priv_size = sizeof(SpectrumSynthContext),
537 .inputs = spectrumsynth_inputs,
538 .outputs = spectrumsynth_outputs,
539 .priv_class = &spectrumsynth_class,