2 * Copyright (c) 2015 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 #include "libavcodec/avfft.h"
24 #include "libavutil/audio_fifo.h"
25 #include "libavutil/avassert.h"
26 #include "libavutil/avstring.h"
27 #include "libavutil/channel_layout.h"
28 #include "libavutil/intreadwrite.h"
29 #include "libavutil/opt.h"
30 #include "libavutil/parseutils.h"
36 enum DisplayMode { LINE, BAR, DOT, NB_MODES };
37 enum FrequencyScale { FS_LINEAR, FS_LOG, FS_RLOG, NB_FSCALES };
38 enum AmplitudeScale { AS_LINEAR, AS_SQRT, AS_CBRT, AS_LOG, NB_ASCALES };
39 enum WindowFunc { WFUNC_RECT, WFUNC_HANNING, WFUNC_HAMMING, WFUNC_BLACKMAN,
40 WFUNC_BARTLETT, WFUNC_WELCH, WFUNC_FLATTOP,
41 WFUNC_BHARRIS, WFUNC_BNUTTALL, WFUNC_SINE, WFUNC_NUTTALL,
42 WFUNC_BHANN, WFUNC_LANCZOS, WFUNC_GAUSS, NB_WFUNC };
44 typedef struct ShowFreqsContext {
53 FFTComplex **fft_data;
55 float *window_func_lut;
67 #define OFFSET(x) offsetof(ShowFreqsContext, x)
68 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
70 static const AVOption showfreqs_options[] = {
71 { "size", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "1024x512"}, 0, 0, FLAGS },
72 { "s", "set video size", OFFSET(w), AV_OPT_TYPE_IMAGE_SIZE, {.str = "1024x512"}, 0, 0, FLAGS },
73 { "mode", "set display mode", OFFSET(mode), AV_OPT_TYPE_INT, {.i64=BAR}, 0, NB_MODES-1, FLAGS, "mode" },
74 { "line", "show lines", 0, AV_OPT_TYPE_CONST, {.i64=LINE}, 0, 0, FLAGS, "mode" },
75 { "bar", "show bars", 0, AV_OPT_TYPE_CONST, {.i64=BAR}, 0, 0, FLAGS, "mode" },
76 { "dot", "show dots", 0, AV_OPT_TYPE_CONST, {.i64=DOT}, 0, 0, FLAGS, "mode" },
77 { "ascale", "set amplitude scale", OFFSET(ascale), AV_OPT_TYPE_INT, {.i64=AS_LOG}, 0, NB_ASCALES-1, FLAGS, "ascale" },
78 { "lin", "linear", 0, AV_OPT_TYPE_CONST, {.i64=AS_LINEAR}, 0, 0, FLAGS, "ascale" },
79 { "sqrt", "square root", 0, AV_OPT_TYPE_CONST, {.i64=AS_SQRT}, 0, 0, FLAGS, "ascale" },
80 { "cbrt", "cubic root", 0, AV_OPT_TYPE_CONST, {.i64=AS_CBRT}, 0, 0, FLAGS, "ascale" },
81 { "log", "logarithmic", 0, AV_OPT_TYPE_CONST, {.i64=AS_LOG}, 0, 0, FLAGS, "ascale" },
82 { "fscale", "set frequency scale", OFFSET(fscale), AV_OPT_TYPE_INT, {.i64=FS_LINEAR}, 0, NB_FSCALES-1, FLAGS, "fscale" },
83 { "lin", "linear", 0, AV_OPT_TYPE_CONST, {.i64=FS_LINEAR}, 0, 0, FLAGS, "fscale" },
84 { "log", "logarithmic", 0, AV_OPT_TYPE_CONST, {.i64=FS_LOG}, 0, 0, FLAGS, "fscale" },
85 { "rlog", "reverse logarithmic", 0, AV_OPT_TYPE_CONST, {.i64=FS_RLOG}, 0, 0, FLAGS, "fscale" },
86 { "win_size", "set window size", OFFSET(fft_bits), AV_OPT_TYPE_INT, {.i64=11}, 4, 16, FLAGS, "fft" },
87 { "w16", 0, 0, AV_OPT_TYPE_CONST, {.i64=4}, 0, 0, FLAGS, "fft" },
88 { "w32", 0, 0, AV_OPT_TYPE_CONST, {.i64=5}, 0, 0, FLAGS, "fft" },
89 { "w64", 0, 0, AV_OPT_TYPE_CONST, {.i64=6}, 0, 0, FLAGS, "fft" },
90 { "w128", 0, 0, AV_OPT_TYPE_CONST, {.i64=7}, 0, 0, FLAGS, "fft" },
91 { "w256", 0, 0, AV_OPT_TYPE_CONST, {.i64=8}, 0, 0, FLAGS, "fft" },
92 { "w512", 0, 0, AV_OPT_TYPE_CONST, {.i64=9}, 0, 0, FLAGS, "fft" },
93 { "w1024", 0, 0, AV_OPT_TYPE_CONST, {.i64=10}, 0, 0, FLAGS, "fft" },
94 { "w2048", 0, 0, AV_OPT_TYPE_CONST, {.i64=11}, 0, 0, FLAGS, "fft" },
95 { "w4096", 0, 0, AV_OPT_TYPE_CONST, {.i64=12}, 0, 0, FLAGS, "fft" },
96 { "w8192", 0, 0, AV_OPT_TYPE_CONST, {.i64=13}, 0, 0, FLAGS, "fft" },
97 { "w16384", 0, 0, AV_OPT_TYPE_CONST, {.i64=14}, 0, 0, FLAGS, "fft" },
98 { "w32768", 0, 0, AV_OPT_TYPE_CONST, {.i64=15}, 0, 0, FLAGS, "fft" },
99 { "w65536", 0, 0, AV_OPT_TYPE_CONST, {.i64=16}, 0, 0, FLAGS, "fft" },
100 { "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64=WFUNC_HANNING}, 0, NB_WFUNC-1, FLAGS, "win_func" },
101 { "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, FLAGS, "win_func" },
102 { "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, FLAGS, "win_func" },
103 { "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, FLAGS, "win_func" },
104 { "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, FLAGS, "win_func" },
105 { "blackman", "Blackman", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BLACKMAN}, 0, 0, FLAGS, "win_func" },
106 { "welch", "Welch", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_WELCH}, 0, 0, FLAGS, "win_func" },
107 { "flattop", "Flat-top", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_FLATTOP}, 0, 0, FLAGS, "win_func" },
108 { "bharris", "Blackman-Harris", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BHARRIS}, 0, 0, FLAGS, "win_func" },
109 { "bnuttall", "Blackman-Nuttall", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BNUTTALL}, 0, 0, FLAGS, "win_func" },
110 { "bhann", "Bartlett-Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BHANN}, 0, 0, FLAGS, "win_func" },
111 { "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, FLAGS, "win_func" },
112 { "nuttall", "Nuttall", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_NUTTALL}, 0, 0, FLAGS, "win_func" },
113 { "lanczos", "Lanczos", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_LANCZOS}, 0, 0, FLAGS, "win_func" },
114 { "gauss", "Gauss", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_GAUSS}, 0, 0, FLAGS, "win_func" },
115 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=1.}, 0., 1., FLAGS },
116 { "averaging", "set time averaging", OFFSET(avg), AV_OPT_TYPE_INT, {.i64=1}, 0, INT32_MAX, FLAGS },
117 { "colors", "set channels colors", OFFSET(colors), AV_OPT_TYPE_STRING, {.str = "red|green|blue|yellow|orange|lime|pink|magenta|brown" }, 0, 0, FLAGS },
121 AVFILTER_DEFINE_CLASS(showfreqs);
123 static int query_formats(AVFilterContext *ctx)
125 AVFilterFormats *formats = NULL;
126 AVFilterChannelLayouts *layouts = NULL;
127 AVFilterLink *inlink = ctx->inputs[0];
128 AVFilterLink *outlink = ctx->outputs[0];
129 static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_FLTP, AV_SAMPLE_FMT_NONE };
130 static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGBA, AV_PIX_FMT_NONE };
132 /* set input audio formats */
133 formats = ff_make_format_list(sample_fmts);
135 return AVERROR(ENOMEM);
136 ff_formats_ref(formats, &inlink->out_formats);
138 layouts = ff_all_channel_layouts();
140 return AVERROR(ENOMEM);
141 ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts);
143 formats = ff_all_samplerates();
145 return AVERROR(ENOMEM);
146 ff_formats_ref(formats, &inlink->out_samplerates);
148 /* set output video format */
149 formats = ff_make_format_list(pix_fmts);
151 return AVERROR(ENOMEM);
152 ff_formats_ref(formats, &outlink->in_formats);
157 static void generate_window_func(float *lut, int N, int win_func, float *overlap)
163 for (n = 0; n < N; n++)
168 for (n = 0; n < N; n++)
169 lut[n] = 1.-FFABS((n-(N-1)/2.)/((N-1)/2.));
173 for (n = 0; n < N; n++)
174 lut[n] = .5*(1-cos(2*M_PI*n/(N-1)));
178 for (n = 0; n < N; n++)
179 lut[n] = .54-.46*cos(2*M_PI*n/(N-1));
183 for (n = 0; n < N; n++)
184 lut[n] = .42659-.49656*cos(2*M_PI*n/(N-1))+.076849*cos(4*M_PI*n/(N-1));
188 for (n = 0; n < N; n++)
189 lut[n] = 1.-(n-(N-1)/2.)/((N-1)/2.)*(n-(N-1)/2.)/((N-1)/2.);
193 for (n = 0; n < N; n++)
194 lut[n] = 1.-1.985844164102*cos( 2*M_PI*n/(N-1))+1.791176438506*cos( 4*M_PI*n/(N-1))-
195 1.282075284005*cos( 6*M_PI*n/(N-1))+0.667777530266*cos( 8*M_PI*n/(N-1))-
196 0.240160796576*cos(10*M_PI*n/(N-1))+0.056656381764*cos(12*M_PI*n/(N-1))-
197 0.008134974479*cos(14*M_PI*n/(N-1))+0.000624544650*cos(16*M_PI*n/(N-1))-
198 0.000019808998*cos(18*M_PI*n/(N-1))+0.000000132974*cos(20*M_PI*n/(N-1));
202 for (n = 0; n < N; n++)
203 lut[n] = 0.35875-0.48829*cos(2*M_PI*n/(N-1))+0.14128*cos(4*M_PI*n/(N-1))-0.01168*cos(6*M_PI*n/(N-1));
207 for (n = 0; n < N; n++)
208 lut[n] = 0.3635819-0.4891775*cos(2*M_PI*n/(N-1))+0.1365995*cos(4*M_PI*n/(N-1))-0.0106411*cos(6*M_PI*n/(N-1));
212 for (n = 0; n < N; n++)
213 lut[n] = 0.62-0.48*FFABS(n/(double)(N-1)-.5)-0.38*cos(2*M_PI*n/(N-1));
217 for (n = 0; n < N; n++)
218 lut[n] = sin(M_PI*n/(N-1));
222 for (n = 0; n < N; n++)
223 lut[n] = 0.355768-0.487396*cos(2*M_PI*n/(N-1))+0.144232*cos(4*M_PI*n/(N-1))-0.012604*cos(6*M_PI*n/(N-1));
227 #define SINC(x) (!(x)) ? 1 : sin(M_PI * (x))/(M_PI * (x));
228 for (n = 0; n < N; n++)
229 lut[n] = SINC((2.*n)/(N-1)-1);
233 for (n = 0; n < N; n++)
234 lut[n] = pow(M_E,-0.5*pow((n-(N-1)/2)/(0.4*(N-1)/2.f),2));
242 static int config_output(AVFilterLink *outlink)
244 AVFilterContext *ctx = outlink->src;
245 AVFilterLink *inlink = ctx->inputs[0];
246 ShowFreqsContext *s = ctx->priv;
250 s->nb_freq = 1 << (s->fft_bits - 1);
251 s->win_size = s->nb_freq << 1;
252 av_audio_fifo_free(s->fifo);
254 s->fft = av_fft_init(s->fft_bits, 0);
256 av_log(ctx, AV_LOG_ERROR, "Unable to create FFT context. "
257 "The window size might be too high.\n");
258 return AVERROR(ENOMEM);
261 /* FFT buffers: x2 for each (display) channel buffer.
262 * Note: we use free and malloc instead of a realloc-like function to
263 * make sure the buffer is aligned in memory for the FFT functions. */
264 for (i = 0; i < s->nb_channels; i++) {
265 av_freep(&s->fft_data[i]);
266 av_freep(&s->avg_data[i]);
268 av_freep(&s->fft_data);
269 av_freep(&s->avg_data);
270 s->nb_channels = inlink->channels;
272 s->fft_data = av_calloc(s->nb_channels, sizeof(*s->fft_data));
274 return AVERROR(ENOMEM);
275 s->avg_data = av_calloc(s->nb_channels, sizeof(*s->avg_data));
277 return AVERROR(ENOMEM);
278 for (i = 0; i < s->nb_channels; i++) {
279 s->fft_data[i] = av_calloc(s->win_size, sizeof(**s->fft_data));
280 s->avg_data[i] = av_calloc(s->nb_freq, sizeof(**s->avg_data));
281 if (!s->fft_data[i] || !s->avg_data[i])
282 return AVERROR(ENOMEM);
285 /* pre-calc windowing function */
286 s->window_func_lut = av_realloc_f(s->window_func_lut, s->win_size,
287 sizeof(*s->window_func_lut));
288 if (!s->window_func_lut)
289 return AVERROR(ENOMEM);
290 generate_window_func(s->window_func_lut, s->win_size, s->win_func, &overlap);
291 if (s->overlap == 1.)
292 s->overlap = overlap;
293 s->skip_samples = (1. - s->overlap) * s->win_size;
294 if (s->skip_samples < 1) {
295 av_log(ctx, AV_LOG_ERROR, "overlap %f too big\n", s->overlap);
296 return AVERROR(EINVAL);
299 for (s->scale = 0, i = 0; i < s->win_size; i++) {
300 s->scale += s->window_func_lut[i] * s->window_func_lut[i];
303 outlink->flags |= FF_LINK_FLAG_REQUEST_LOOP;
304 outlink->frame_rate = av_make_q(inlink->sample_rate, s->win_size * (1.-s->overlap));
305 outlink->sample_aspect_ratio = (AVRational){1,1};
309 s->fifo = av_audio_fifo_alloc(inlink->format, inlink->channels, s->win_size);
311 return AVERROR(ENOMEM);
315 static inline void draw_dot(AVFrame *out, int x, int y, uint8_t fg[4])
318 uint32_t color = AV_RL32(out->data[0] + y * out->linesize[0] + x * 4);
320 if ((color & 0xffffff) != 0)
321 AV_WL32(out->data[0] + y * out->linesize[0] + x * 4, AV_RL32(fg) | color);
323 AV_WL32(out->data[0] + y * out->linesize[0] + x * 4, AV_RL32(fg));
326 static int get_sx(ShowFreqsContext *s, int f)
330 return (s->w/(float)s->nb_freq)*f;
332 return s->w-pow(s->w, (s->nb_freq-f-1)/(s->nb_freq-1.));
334 return pow(s->w, f/(s->nb_freq-1.));
340 static float get_bsize(ShowFreqsContext *s, int f)
344 return s->w/(float)s->nb_freq;
346 return pow(s->w, (s->nb_freq-f-1)/(s->nb_freq-1.))-
347 pow(s->w, (s->nb_freq-f-2)/(s->nb_freq-1.));
349 return pow(s->w, (f+1)/(s->nb_freq-1.))-
350 pow(s->w, f /(s->nb_freq-1.));
356 static inline void plot_freq(ShowFreqsContext *s, int ch,
357 double a, int f, uint8_t fg[4], int *prev_y,
358 AVFrame *out, AVFilterLink *outlink)
361 const float avg = s->avg_data[ch][f];
362 const float bsize = get_bsize(s, f);
363 const int sx = get_sx(s, f);
374 a = log(av_clipd(a, 1e-6, 1)) / log(1e-6);
380 y = a * outlink->h - 1;
386 y = s->avg_data[ch][f] = !outlink->frame_count ? y : FFMIN(avg, y);
391 s->avg_data[ch][f] = avg + y * (y - avg) / (FFMIN(outlink->frame_count + 1, s->avg) * y);
392 y = s->avg_data[ch][f];
402 for (x = sx + 1; x < sx + bsize && x < w; x++)
403 draw_dot(out, x, y, fg);
404 for (i = y; i <= *prev_y; i++)
405 draw_dot(out, sx, i, fg);
407 for (i = *prev_y; i <= y; i++)
408 draw_dot(out, sx, i, fg);
409 for (x = sx + 1; x < sx + bsize && x < w; x++)
410 draw_dot(out, x, i - 1, fg);
415 for (x = sx; x < sx + bsize && x < w; x++)
416 for (i = y; i < outlink->h; i++)
417 draw_dot(out, x, i, fg);
420 for (x = sx; x < sx + bsize && x < w; x++)
421 draw_dot(out, x, y, fg);
426 static int plot_freqs(AVFilterLink *inlink, AVFrame *in)
428 AVFilterContext *ctx = inlink->dst;
429 AVFilterLink *outlink = ctx->outputs[0];
430 ShowFreqsContext *s = ctx->priv;
431 const int win_size = s->win_size;
432 char *colors, *color, *saveptr = NULL;
436 out = ff_get_video_buffer(outlink, outlink->w, outlink->h);
438 return AVERROR(ENOMEM);
440 for (n = 0; n < outlink->h; n++)
441 memset(out->data[0] + out->linesize[0] * n, 0, outlink->w * 4);
443 /* fill FFT input with the number of samples available */
444 for (ch = 0; ch < s->nb_channels; ch++) {
445 const float *p = (float *)in->extended_data[ch];
447 for (n = 0; n < in->nb_samples; n++) {
448 s->fft_data[ch][n].re = p[n] * s->window_func_lut[n];
449 s->fft_data[ch][n].im = 0;
451 for (; n < win_size; n++) {
452 s->fft_data[ch][n].re = 0;
453 s->fft_data[ch][n].im = 0;
457 /* run FFT on each samples set */
458 for (ch = 0; ch < s->nb_channels; ch++) {
459 av_fft_permute(s->fft, s->fft_data[ch]);
460 av_fft_calc(s->fft, s->fft_data[ch]);
463 #define RE(x, ch) s->fft_data[ch][x].re
464 #define IM(x, ch) s->fft_data[ch][x].im
465 #define M(a, b) (sqrt((a) * (a) + (b) * (b)))
467 colors = av_strdup(s->colors);
470 return AVERROR(ENOMEM);
473 for (ch = 0; ch < s->nb_channels; ch++) {
474 uint8_t fg[4] = { 0xff, 0xff, 0xff, 0xff };
478 color = av_strtok(ch == 0 ? colors : NULL, " |", &saveptr);
480 av_parse_color(fg, color, -1, ctx);
482 a = av_clipd(M(RE(0, ch), 0) / s->scale, 0, 1);
483 plot_freq(s, ch, a, 0, fg, &prev_y, out, outlink);
485 for (f = 1; f < s->nb_freq; f++) {
486 a = av_clipd(M(RE(f, ch), IM(f, ch)) / s->scale, 0, 1);
488 plot_freq(s, ch, a, f, fg, &prev_y, out, outlink);
494 return ff_filter_frame(outlink, out);
497 static int filter_frame(AVFilterLink *inlink, AVFrame *in)
499 AVFilterContext *ctx = inlink->dst;
500 ShowFreqsContext *s = ctx->priv;
504 av_audio_fifo_write(s->fifo, (void **)in->extended_data, in->nb_samples);
505 while (av_audio_fifo_size(s->fifo) >= s->win_size) {
506 fin = ff_get_audio_buffer(inlink, s->win_size);
508 ret = AVERROR(ENOMEM);
513 s->pts += s->skip_samples;
514 ret = av_audio_fifo_peek(s->fifo, (void **)fin->extended_data, s->win_size);
518 ret = plot_freqs(inlink, fin);
520 av_audio_fifo_drain(s->fifo, s->skip_samples);
531 static av_cold void uninit(AVFilterContext *ctx)
533 ShowFreqsContext *s = ctx->priv;
537 for (i = 0; i < s->nb_channels; i++) {
538 av_freep(&s->fft_data[i]);
539 av_freep(&s->avg_data[i]);
541 av_freep(&s->fft_data);
542 av_freep(&s->avg_data);
543 av_freep(&s->window_func_lut);
544 av_audio_fifo_free(s->fifo);
547 static const AVFilterPad showfreqs_inputs[] = {
550 .type = AVMEDIA_TYPE_AUDIO,
551 .filter_frame = filter_frame,
556 static const AVFilterPad showfreqs_outputs[] = {
559 .type = AVMEDIA_TYPE_VIDEO,
560 .config_props = config_output,
565 AVFilter ff_avf_showfreqs = {
567 .description = NULL_IF_CONFIG_SMALL("Convert input audio to a frequencies video output."),
569 .query_formats = query_formats,
570 .priv_size = sizeof(ShowFreqsContext),
571 .inputs = showfreqs_inputs,
572 .outputs = showfreqs_outputs,
573 .priv_class = &showfreqs_class,