FFmpeg
 All Data Structures Namespaces Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
avf_showcqt.c
Go to the documentation of this file.
1 /*
2  * Copyright (c) 2014 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 "config.h"
22 #include "libavcodec/avfft.h"
23 #include "libavutil/avassert.h"
25 #include "libavutil/opt.h"
27 #include "libavutil/qsort.h"
28 #include "libavutil/time.h"
29 #include "libavutil/eval.h"
30 #include "avfilter.h"
31 #include "internal.h"
32 
33 #include <math.h>
34 #include <stdlib.h>
35 
36 #if CONFIG_LIBFREETYPE
37 #include <ft2build.h>
38 #include FT_FREETYPE_H
39 #endif
40 
41 /* this filter is designed to do 16 bins/semitones constant Q transform with Brown-Puckette algorithm
42  * start from E0 to D#10 (10 octaves)
43  * so there are 16 bins/semitones * 12 semitones/octaves * 10 octaves = 1920 bins
44  * match with full HD resolution */
45 
46 #define VIDEO_WIDTH 1920
47 #define VIDEO_HEIGHT 1080
48 #define FONT_HEIGHT 32
49 #define SPECTOGRAM_HEIGHT ((VIDEO_HEIGHT-FONT_HEIGHT)/2)
50 #define SPECTOGRAM_START (VIDEO_HEIGHT-SPECTOGRAM_HEIGHT)
51 #define BASE_FREQ 20.051392800492
52 #define COEFF_CLAMP 1.0e-4
53 #define TLENGTH_MIN 0.001
54 #define TLENGTH_DEFAULT "384/f*tc/(384/f+tc)"
55 #define VOLUME_MIN 1e-10
56 #define VOLUME_MAX 100.0
57 #define FONTCOLOR_DEFAULT "st(0, (midi(f)-59.5)/12);" \
58  "st(1, if(between(ld(0),0,1), 0.5-0.5*cos(2*PI*ld(0)), 0));" \
59  "r(1-ld(1)) + b(ld(1))"
60 
61 typedef struct {
63  int index;
64 } SparseCoeff;
65 
66 typedef struct {
67  const AVClass *class;
77  char *fontfile; /* using freetype */
78  int coeffs_len[VIDEO_WIDTH];
79  uint8_t fontcolor_value[VIDEO_WIDTH*3]; /* result of fontcolor option */
80  int64_t frame_count;
83  int fft_bits;
86  char *tlength;
87  char *volume;
88  char *fontcolor;
89  double timeclamp; /* lower timeclamp, time-accurate, higher timeclamp, freq-accurate (at low freq)*/
90  float coeffclamp; /* lower coeffclamp, more precise, higher coeffclamp, faster */
91  int fullhd; /* if true, output video is at full HD resolution, otherwise it will be halved */
92  float gamma; /* lower gamma, more contrast, higher gamma, more range */
93  float gamma2; /* gamma of bargraph */
94  int fps; /* the required fps is so strict, so it's enough to be int, but 24000/1001 etc cannot be encoded */
95  int count; /* fps * count = transform rate */
97 
98 #define OFFSET(x) offsetof(ShowCQTContext, x)
99 #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM
100 
101 static const AVOption showcqt_options[] = {
102  { "volume", "set volume", OFFSET(volume), AV_OPT_TYPE_STRING, { .str = "16" }, CHAR_MIN, CHAR_MAX, FLAGS },
103  { "tlength", "set transform length", OFFSET(tlength), AV_OPT_TYPE_STRING, { .str = TLENGTH_DEFAULT }, CHAR_MIN, CHAR_MAX, FLAGS },
104  { "timeclamp", "set timeclamp", OFFSET(timeclamp), AV_OPT_TYPE_DOUBLE, { .dbl = 0.17 }, 0.1, 1.0, FLAGS },
105  { "coeffclamp", "set coeffclamp", OFFSET(coeffclamp), AV_OPT_TYPE_FLOAT, { .dbl = 1 }, 0.1, 10, FLAGS },
106  { "gamma", "set gamma", OFFSET(gamma), AV_OPT_TYPE_FLOAT, { .dbl = 3 }, 1, 7, FLAGS },
107  { "gamma2", "set gamma of bargraph", OFFSET(gamma2), AV_OPT_TYPE_FLOAT, { .dbl = 1 }, 1, 7, FLAGS },
108  { "fullhd", "set full HD resolution", OFFSET(fullhd), AV_OPT_TYPE_INT, { .i64 = 1 }, 0, 1, FLAGS },
109  { "fps", "set video fps", OFFSET(fps), AV_OPT_TYPE_INT, { .i64 = 25 }, 10, 100, FLAGS },
110  { "count", "set number of transform per frame", OFFSET(count), AV_OPT_TYPE_INT, { .i64 = 6 }, 1, 30, FLAGS },
111  { "fontfile", "set font file", OFFSET(fontfile), AV_OPT_TYPE_STRING, { .str = NULL }, CHAR_MIN, CHAR_MAX, FLAGS },
112  { "fontcolor", "set font color", OFFSET(fontcolor), AV_OPT_TYPE_STRING, { .str = FONTCOLOR_DEFAULT }, CHAR_MIN, CHAR_MAX, FLAGS },
113  { NULL }
114 };
115 
116 AVFILTER_DEFINE_CLASS(showcqt);
117 
118 static av_cold void uninit(AVFilterContext *ctx)
119 {
120  int k;
121 
122  ShowCQTContext *s = ctx->priv;
124  s->fft_context = NULL;
125  for (k = 0; k < VIDEO_WIDTH; k++)
126  av_freep(&s->coeffs[k]);
127  av_freep(&s->fft_data);
130  av_freep(&s->coeff_sort);
131  av_freep(&s->spectogram);
132  av_freep(&s->font_alpha);
134 }
135 
137 {
140  AVFilterLink *inlink = ctx->inputs[0];
141  AVFilterLink *outlink = ctx->outputs[0];
143  static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE };
144  static const int64_t channel_layouts[] = { AV_CH_LAYOUT_STEREO, AV_CH_LAYOUT_STEREO_DOWNMIX, -1 };
145  static const int samplerates[] = { 44100, 48000, -1 };
146 
147  /* set input audio formats */
148  formats = ff_make_format_list(sample_fmts);
149  if (!formats)
150  return AVERROR(ENOMEM);
151  ff_formats_ref(formats, &inlink->out_formats);
152 
153  layouts = avfilter_make_format64_list(channel_layouts);
154  if (!layouts)
155  return AVERROR(ENOMEM);
156  ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts);
157 
158  formats = ff_make_format_list(samplerates);
159  if (!formats)
160  return AVERROR(ENOMEM);
161  ff_formats_ref(formats, &inlink->out_samplerates);
162 
163  /* set output video format */
164  formats = ff_make_format_list(pix_fmts);
165  if (!formats)
166  return AVERROR(ENOMEM);
167  ff_formats_ref(formats, &outlink->in_formats);
168 
169  return 0;
170 }
171 
172 #if CONFIG_LIBFREETYPE
173 static void load_freetype_font(AVFilterContext *ctx)
174 {
175  static const char str[] = "EF G A BC D ";
176  ShowCQTContext *s = ctx->priv;
177  FT_Library lib = NULL;
178  FT_Face face = NULL;
179  int video_scale = s->fullhd ? 2 : 1;
180  int video_width = (VIDEO_WIDTH/2) * video_scale;
181  int font_height = (FONT_HEIGHT/2) * video_scale;
182  int font_width = 8 * video_scale;
183  int font_repeat = font_width * 12;
184  int linear_hori_advance = font_width * 65536;
185  int non_monospace_warning = 0;
186  int x;
187 
188  s->font_alpha = NULL;
189 
190  if (!s->fontfile)
191  return;
192 
193  if (FT_Init_FreeType(&lib))
194  goto fail;
195 
196  if (FT_New_Face(lib, s->fontfile, 0, &face))
197  goto fail;
198 
199  if (FT_Set_Char_Size(face, 16*64, 0, 0, 0))
200  goto fail;
201 
202  if (FT_Load_Char(face, 'A', FT_LOAD_RENDER))
203  goto fail;
204 
205  if (FT_Set_Char_Size(face, 16*64 * linear_hori_advance / face->glyph->linearHoriAdvance, 0, 0, 0))
206  goto fail;
207 
208  s->font_alpha = av_malloc_array(font_height, video_width);
209  if (!s->font_alpha)
210  goto fail;
211 
212  memset(s->font_alpha, 0, font_height * video_width);
213 
214  for (x = 0; x < 12; x++) {
215  int sx, sy, rx, bx, by, dx, dy;
216 
217  if (str[x] == ' ')
218  continue;
219 
220  if (FT_Load_Char(face, str[x], FT_LOAD_RENDER))
221  goto fail;
222 
223  if (face->glyph->advance.x != font_width*64 && !non_monospace_warning) {
224  av_log(ctx, AV_LOG_WARNING, "Font is not monospace\n");
225  non_monospace_warning = 1;
226  }
227 
228  sy = font_height - 4*video_scale - face->glyph->bitmap_top;
229  for (rx = 0; rx < 10; rx++) {
230  sx = rx * font_repeat + x * font_width + face->glyph->bitmap_left;
231  for (by = 0; by < face->glyph->bitmap.rows; by++) {
232  dy = by + sy;
233  if (dy < 0)
234  continue;
235  if (dy >= font_height)
236  break;
237 
238  for (bx = 0; bx < face->glyph->bitmap.width; bx++) {
239  dx = bx + sx;
240  if (dx < 0)
241  continue;
242  if (dx >= video_width)
243  break;
244  s->font_alpha[dy*video_width+dx] = face->glyph->bitmap.buffer[by*face->glyph->bitmap.width+bx];
245  }
246  }
247  }
248  }
249 
250  FT_Done_Face(face);
251  FT_Done_FreeType(lib);
252  return;
253 
254  fail:
255  av_log(ctx, AV_LOG_WARNING, "Error while loading freetype font, using default font instead\n");
256  FT_Done_Face(face);
257  FT_Done_FreeType(lib);
258  av_freep(&s->font_alpha);
259  return;
260 }
261 #endif
262 
263 static double a_weighting(void *p, double f)
264 {
265  double ret = 12200.0*12200.0 * (f*f*f*f);
266  ret /= (f*f + 20.6*20.6) * (f*f + 12200.0*12200.0) *
267  sqrt((f*f + 107.7*107.7) * (f*f + 737.9*737.9));
268  return ret;
269 }
270 
271 static double b_weighting(void *p, double f)
272 {
273  double ret = 12200.0*12200.0 * (f*f*f);
274  ret /= (f*f + 20.6*20.6) * (f*f + 12200.0*12200.0) * sqrt(f*f + 158.5*158.5);
275  return ret;
276 }
277 
278 static double c_weighting(void *p, double f)
279 {
280  double ret = 12200.0*12200.0 * (f*f);
281  ret /= (f*f + 20.6*20.6) * (f*f + 12200.0*12200.0);
282  return ret;
283 }
284 
285 static double midi(void *p, double f)
286 {
287  return log2(f/440.0) * 12.0 + 69.0;
288 }
289 
290 static double r_func(void *p, double x)
291 {
292  x = av_clipd(x, 0.0, 1.0);
293  return (int)(x*255.0+0.5) << 16;
294 }
295 
296 static double g_func(void *p, double x)
297 {
298  x = av_clipd(x, 0.0, 1.0);
299  return (int)(x*255.0+0.5) << 8;
300 }
301 
302 static double b_func(void *p, double x)
303 {
304  x = av_clipd(x, 0.0, 1.0);
305  return (int)(x*255.0+0.5);
306 }
307 
308 static inline int qsort_sparsecoeff(const SparseCoeff *a, const SparseCoeff *b)
309 {
310  if (fabsf(a->value) >= fabsf(b->value))
311  return 1;
312  else
313  return -1;
314 }
315 
316 static int config_output(AVFilterLink *outlink)
317 {
318  AVFilterContext *ctx = outlink->src;
319  AVFilterLink *inlink = ctx->inputs[0];
320  ShowCQTContext *s = ctx->priv;
321  AVExpr *tlength_expr = NULL, *volume_expr = NULL, *fontcolor_expr = NULL;
322  uint8_t *fontcolor_value = s->fontcolor_value;
323  static const char * const expr_vars[] = { "timeclamp", "tc", "frequency", "freq", "f", NULL };
324  static const char * const expr_func_names[] = { "a_weighting", "b_weighting", "c_weighting", NULL };
325  static const char * const expr_fontcolor_func_names[] = { "midi", "r", "g", "b", NULL };
326  static double (* const expr_funcs[])(void *, double) = { a_weighting, b_weighting, c_weighting, NULL };
327  static double (* const expr_fontcolor_funcs[])(void *, double) = { midi, r_func, g_func, b_func, NULL };
328  int fft_len, k, x, y, ret;
329  int num_coeffs = 0;
330  int rate = inlink->sample_rate;
331  double max_len = rate * (double) s->timeclamp;
332  int64_t start_time, end_time;
333  int video_scale = s->fullhd ? 2 : 1;
334  int video_width = (VIDEO_WIDTH/2) * video_scale;
335  int video_height = (VIDEO_HEIGHT/2) * video_scale;
336  int spectogram_height = (SPECTOGRAM_HEIGHT/2) * video_scale;
337 
338  s->fft_bits = ceil(log2(max_len));
339  fft_len = 1 << s->fft_bits;
340 
341  if (rate % (s->fps * s->count)) {
342  av_log(ctx, AV_LOG_ERROR, "Rate (%u) is not divisible by fps*count (%u*%u)\n", rate, s->fps, s->count);
343  return AVERROR(EINVAL);
344  }
345 
346  s->fft_data = av_malloc_array(fft_len, sizeof(*s->fft_data));
347  s->coeff_sort = av_malloc_array(fft_len, sizeof(*s->coeff_sort));
348  s->fft_result_left = av_malloc_array(fft_len, sizeof(*s->fft_result_left));
349  s->fft_result_right = av_malloc_array(fft_len, sizeof(*s->fft_result_right));
350  s->fft_context = av_fft_init(s->fft_bits, 0);
351 
352  if (!s->fft_data || !s->coeff_sort || !s->fft_result_left || !s->fft_result_right || !s->fft_context)
353  return AVERROR(ENOMEM);
354 
355 #if CONFIG_LIBFREETYPE
356  load_freetype_font(ctx);
357 #else
358  if (s->fontfile)
359  av_log(ctx, AV_LOG_WARNING, "Freetype is not available, ignoring fontfile option\n");
360  s->font_alpha = NULL;
361 #endif
362 
363  av_log(ctx, AV_LOG_INFO, "Calculating spectral kernel, please wait\n");
364  start_time = av_gettime_relative();
365  ret = av_expr_parse(&tlength_expr, s->tlength, expr_vars, NULL, NULL, NULL, NULL, 0, ctx);
366  if (ret < 0)
367  goto eval_error;
368 
369  ret = av_expr_parse(&volume_expr, s->volume, expr_vars, expr_func_names,
370  expr_funcs, NULL, NULL, 0, ctx);
371  if (ret < 0)
372  goto eval_error;
373 
374  ret = av_expr_parse(&fontcolor_expr, s->fontcolor, expr_vars, expr_fontcolor_func_names,
375  expr_fontcolor_funcs, NULL, NULL, 0, ctx);
376  if (ret < 0)
377  goto eval_error;
378 
379  for (k = 0; k < VIDEO_WIDTH; k++) {
380  int hlen = fft_len >> 1;
381  float total = 0;
382  float partial = 0;
383  double freq = BASE_FREQ * exp2(k * (1.0/192.0));
384  double tlen, tlength, volume;
385  double expr_vars_val[] = { s->timeclamp, s->timeclamp, freq, freq, freq, 0 };
386  /* a window function from Albert H. Nuttall,
387  * "Some Windows with Very Good Sidelobe Behavior"
388  * -93.32 dB peak sidelobe and 18 dB/octave asymptotic decay
389  * coefficient normalized to a0 = 1 */
390  double a0 = 0.355768;
391  double a1 = 0.487396/a0;
392  double a2 = 0.144232/a0;
393  double a3 = 0.012604/a0;
394  double sv_step, cv_step, sv, cv;
395  double sw_step, cw_step, sw, cw, w;
396 
397  tlength = av_expr_eval(tlength_expr, expr_vars_val, NULL);
398  if (isnan(tlength)) {
399  av_log(ctx, AV_LOG_WARNING, "at freq %g: tlength is nan, setting it to %g\n", freq, s->timeclamp);
400  tlength = s->timeclamp;
401  } else if (tlength < TLENGTH_MIN) {
402  av_log(ctx, AV_LOG_WARNING, "at freq %g: tlength is %g, setting it to %g\n", freq, tlength, TLENGTH_MIN);
403  tlength = TLENGTH_MIN;
404  } else if (tlength > s->timeclamp) {
405  av_log(ctx, AV_LOG_WARNING, "at freq %g: tlength is %g, setting it to %g\n", freq, tlength, s->timeclamp);
406  tlength = s->timeclamp;
407  }
408 
409  volume = FFABS(av_expr_eval(volume_expr, expr_vars_val, NULL));
410  if (isnan(volume)) {
411  av_log(ctx, AV_LOG_WARNING, "at freq %g: volume is nan, setting it to 0\n", freq);
412  volume = VOLUME_MIN;
413  } else if (volume < VOLUME_MIN) {
414  volume = VOLUME_MIN;
415  } else if (volume > VOLUME_MAX) {
416  av_log(ctx, AV_LOG_WARNING, "at freq %g: volume is %g, setting it to %g\n", freq, volume, VOLUME_MAX);
417  volume = VOLUME_MAX;
418  }
419 
420  if (s->fullhd || !(k & 1)) {
421  int fontcolor = av_expr_eval(fontcolor_expr, expr_vars_val, NULL);
422  fontcolor_value[0] = (fontcolor >> 16) & 0xFF;
423  fontcolor_value[1] = (fontcolor >> 8) & 0xFF;
424  fontcolor_value[2] = fontcolor & 0xFF;
425  fontcolor_value += 3;
426  }
427 
428  tlen = tlength * rate;
429  s->fft_data[0].re = 0;
430  s->fft_data[0].im = 0;
431  s->fft_data[hlen].re = (1.0 + a1 + a2 + a3) * (1.0/tlen) * volume * (1.0/fft_len);
432  s->fft_data[hlen].im = 0;
433  sv_step = sv = sin(2.0*M_PI*freq*(1.0/rate));
434  cv_step = cv = cos(2.0*M_PI*freq*(1.0/rate));
435  /* also optimizing window func */
436  sw_step = sw = sin(2.0*M_PI*(1.0/tlen));
437  cw_step = cw = cos(2.0*M_PI*(1.0/tlen));
438  for (x = 1; x < 0.5 * tlen; x++) {
439  double cv_tmp, cw_tmp;
440  double cw2, cw3, sw2;
441 
442  cw2 = cw * cw - sw * sw;
443  sw2 = cw * sw + sw * cw;
444  cw3 = cw * cw2 - sw * sw2;
445  w = (1.0 + a1 * cw + a2 * cw2 + a3 * cw3) * (1.0/tlen) * volume * (1.0/fft_len);
446  s->fft_data[hlen + x].re = w * cv;
447  s->fft_data[hlen + x].im = w * sv;
448  s->fft_data[hlen - x].re = s->fft_data[hlen + x].re;
449  s->fft_data[hlen - x].im = -s->fft_data[hlen + x].im;
450 
451  cv_tmp = cv * cv_step - sv * sv_step;
452  sv = sv * cv_step + cv * sv_step;
453  cv = cv_tmp;
454  cw_tmp = cw * cw_step - sw * sw_step;
455  sw = sw * cw_step + cw * sw_step;
456  cw = cw_tmp;
457  }
458  for (; x < hlen; x++) {
459  s->fft_data[hlen + x].re = 0;
460  s->fft_data[hlen + x].im = 0;
461  s->fft_data[hlen - x].re = 0;
462  s->fft_data[hlen - x].im = 0;
463  }
466 
467  for (x = 0; x < fft_len; x++) {
468  s->coeff_sort[x].index = x;
469  s->coeff_sort[x].value = s->fft_data[x].re;
470  }
471 
473  for (x = 0; x < fft_len; x++)
474  total += fabsf(s->coeff_sort[x].value);
475 
476  for (x = 0; x < fft_len; x++) {
477  partial += fabsf(s->coeff_sort[x].value);
478  if (partial > total * s->coeffclamp * COEFF_CLAMP) {
479  s->coeffs_len[k] = fft_len - x;
480  num_coeffs += s->coeffs_len[k];
481  s->coeffs[k] = av_malloc_array(s->coeffs_len[k], sizeof(*s->coeffs[k]));
482  if (!s->coeffs[k]) {
483  ret = AVERROR(ENOMEM);
484  goto eval_error;
485  }
486  for (y = 0; y < s->coeffs_len[k]; y++)
487  s->coeffs[k][y] = s->coeff_sort[x+y];
488  break;
489  }
490  }
491  }
492  av_expr_free(fontcolor_expr);
493  av_expr_free(volume_expr);
494  av_expr_free(tlength_expr);
495  end_time = av_gettime_relative();
496  av_log(ctx, AV_LOG_INFO, "Elapsed time %.6f s (fft_len=%u, num_coeffs=%u)\n", 1e-6 * (end_time-start_time), fft_len, num_coeffs);
497 
498  outlink->w = video_width;
499  outlink->h = video_height;
500 
501  s->req_fullfilled = 0;
502  s->spectogram_index = 0;
503  s->frame_count = 0;
504  s->spectogram_count = 0;
505  s->remaining_fill = fft_len >> 1;
506  memset(s->fft_data, 0, fft_len * sizeof(*s->fft_data));
507 
508  s->outpicref = ff_get_video_buffer(outlink, outlink->w, outlink->h);
509  if (!s->outpicref)
510  return AVERROR(ENOMEM);
511 
512  s->spectogram = av_calloc(spectogram_height, s->outpicref->linesize[0]);
513  if (!s->spectogram)
514  return AVERROR(ENOMEM);
515 
516  outlink->sample_aspect_ratio = av_make_q(1, 1);
517  outlink->time_base = av_make_q(1, s->fps);
518  outlink->frame_rate = av_make_q(s->fps, 1);
519  return 0;
520 
521 eval_error:
522  av_expr_free(fontcolor_expr);
523  av_expr_free(volume_expr);
524  av_expr_free(tlength_expr);
525  return ret;
526 }
527 
528 static int plot_cqt(AVFilterLink *inlink)
529 {
530  AVFilterContext *ctx = inlink->dst;
531  ShowCQTContext *s = ctx->priv;
532  AVFilterLink *outlink = ctx->outputs[0];
533  int fft_len = 1 << s->fft_bits;
534  FFTSample result[VIDEO_WIDTH][4];
535  int x, y, ret = 0;
536  int linesize = s->outpicref->linesize[0];
537  int video_scale = s->fullhd ? 2 : 1;
538  int video_width = (VIDEO_WIDTH/2) * video_scale;
539  int spectogram_height = (SPECTOGRAM_HEIGHT/2) * video_scale;
540  int spectogram_start = (SPECTOGRAM_START/2) * video_scale;
541  int font_height = (FONT_HEIGHT/2) * video_scale;
542 
543  /* real part contains left samples, imaginary part contains right samples */
544  memcpy(s->fft_result_left, s->fft_data, fft_len * sizeof(*s->fft_data));
547 
548  /* separate left and right, (and multiply by 2.0) */
549  s->fft_result_right[0].re = 2.0f * s->fft_result_left[0].im;
550  s->fft_result_right[0].im = 0;
551  s->fft_result_left[0].re = 2.0f * s->fft_result_left[0].re;
552  s->fft_result_left[0].im = 0;
553  for (x = 1; x <= fft_len >> 1; x++) {
554  FFTSample tmpy = s->fft_result_left[fft_len-x].im - s->fft_result_left[x].im;
555 
556  s->fft_result_right[x].re = s->fft_result_left[x].im + s->fft_result_left[fft_len-x].im;
557  s->fft_result_right[x].im = s->fft_result_left[x].re - s->fft_result_left[fft_len-x].re;
558  s->fft_result_right[fft_len-x].re = s->fft_result_right[x].re;
559  s->fft_result_right[fft_len-x].im = -s->fft_result_right[x].im;
560 
561  s->fft_result_left[x].re = s->fft_result_left[x].re + s->fft_result_left[fft_len-x].re;
562  s->fft_result_left[x].im = tmpy;
563  s->fft_result_left[fft_len-x].re = s->fft_result_left[x].re;
564  s->fft_result_left[fft_len-x].im = -s->fft_result_left[x].im;
565  }
566 
567  /* calculating cqt */
568  for (x = 0; x < VIDEO_WIDTH; x++) {
569  int u;
570  float g = 1.0f / s->gamma;
571  float g2 = 1.0f / s->gamma2;
572  FFTComplex l = {0,0};
573  FFTComplex r = {0,0};
574 
575  for (u = 0; u < s->coeffs_len[x]; u++) {
576  FFTSample value = s->coeffs[x][u].value;
577  int index = s->coeffs[x][u].index;
578  l.re += value * s->fft_result_left[index].re;
579  l.im += value * s->fft_result_left[index].im;
580  r.re += value * s->fft_result_right[index].re;
581  r.im += value * s->fft_result_right[index].im;
582  }
583  /* result is power, not amplitude */
584  result[x][0] = l.re * l.re + l.im * l.im;
585  result[x][2] = r.re * r.re + r.im * r.im;
586  result[x][1] = 0.5f * (result[x][0] + result[x][2]);
587  result[x][3] = (g2 == 1.0f) ? result[x][1] : powf(result[x][1], g2);
588  result[x][0] = 255.0f * powf(FFMIN(1.0f,result[x][0]), g);
589  result[x][1] = 255.0f * powf(FFMIN(1.0f,result[x][1]), g);
590  result[x][2] = 255.0f * powf(FFMIN(1.0f,result[x][2]), g);
591  }
592 
593  if (!s->fullhd) {
594  for (x = 0; x < video_width; x++) {
595  result[x][0] = 0.5f * (result[2*x][0] + result[2*x+1][0]);
596  result[x][1] = 0.5f * (result[2*x][1] + result[2*x+1][1]);
597  result[x][2] = 0.5f * (result[2*x][2] + result[2*x+1][2]);
598  result[x][3] = 0.5f * (result[2*x][3] + result[2*x+1][3]);
599  }
600  }
601 
602  for (x = 0; x < video_width; x++) {
603  s->spectogram[s->spectogram_index*linesize + 3*x] = result[x][0] + 0.5f;
604  s->spectogram[s->spectogram_index*linesize + 3*x + 1] = result[x][1] + 0.5f;
605  s->spectogram[s->spectogram_index*linesize + 3*x + 2] = result[x][2] + 0.5f;
606  }
607 
608  /* drawing */
609  if (!s->spectogram_count) {
610  uint8_t *data = (uint8_t*) s->outpicref->data[0];
611  float rcp_result[VIDEO_WIDTH];
612  int total_length = linesize * spectogram_height;
613  int back_length = linesize * s->spectogram_index;
614 
615  for (x = 0; x < video_width; x++)
616  rcp_result[x] = 1.0f / (result[x][3]+0.0001f);
617 
618  /* drawing bar */
619  for (y = 0; y < spectogram_height; y++) {
620  float height = (spectogram_height - y) * (1.0f/spectogram_height);
621  uint8_t *lineptr = data + y * linesize;
622  for (x = 0; x < video_width; x++) {
623  float mul;
624  if (result[x][3] <= height) {
625  *lineptr++ = 0;
626  *lineptr++ = 0;
627  *lineptr++ = 0;
628  } else {
629  mul = (result[x][3] - height) * rcp_result[x];
630  *lineptr++ = mul * result[x][0] + 0.5f;
631  *lineptr++ = mul * result[x][1] + 0.5f;
632  *lineptr++ = mul * result[x][2] + 0.5f;
633  }
634  }
635  }
636 
637  /* drawing font */
638  if (s->font_alpha) {
639  for (y = 0; y < font_height; y++) {
640  uint8_t *lineptr = data + (spectogram_height + y) * linesize;
641  uint8_t *spectogram_src = s->spectogram + s->spectogram_index * linesize;
642  uint8_t *fontcolor_value = s->fontcolor_value;
643  for (x = 0; x < video_width; x++) {
644  uint8_t alpha = s->font_alpha[y*video_width+x];
645  lineptr[3*x] = (spectogram_src[3*x] * (255-alpha) + fontcolor_value[0] * alpha + 255) >> 8;
646  lineptr[3*x+1] = (spectogram_src[3*x+1] * (255-alpha) + fontcolor_value[1] * alpha + 255) >> 8;
647  lineptr[3*x+2] = (spectogram_src[3*x+2] * (255-alpha) + fontcolor_value[2] * alpha + 255) >> 8;
648  fontcolor_value += 3;
649  }
650  }
651  } else {
652  for (y = 0; y < font_height; y++) {
653  uint8_t *lineptr = data + (spectogram_height + y) * linesize;
654  memcpy(lineptr, s->spectogram + s->spectogram_index * linesize, video_width*3);
655  }
656  for (x = 0; x < video_width; x += video_width/10) {
657  int u;
658  static const char str[] = "EF G A BC D ";
659  uint8_t *startptr = data + spectogram_height * linesize + x * 3;
660  for (u = 0; str[u]; u++) {
661  int v;
662  for (v = 0; v < 16; v++) {
663  uint8_t *p = startptr + v * linesize * video_scale + 8 * 3 * u * video_scale;
664  int ux = x + 8 * u * video_scale;
665  int mask;
666  for (mask = 0x80; mask; mask >>= 1) {
667  if (mask & avpriv_vga16_font[str[u] * 16 + v]) {
668  p[0] = s->fontcolor_value[3*ux];
669  p[1] = s->fontcolor_value[3*ux+1];
670  p[2] = s->fontcolor_value[3*ux+2];
671  if (video_scale == 2) {
672  p[linesize] = p[0];
673  p[linesize+1] = p[1];
674  p[linesize+2] = p[2];
675  p[3] = p[linesize+3] = s->fontcolor_value[3*ux+3];
676  p[4] = p[linesize+4] = s->fontcolor_value[3*ux+4];
677  p[5] = p[linesize+5] = s->fontcolor_value[3*ux+5];
678  }
679  }
680  p += 3 * video_scale;
681  ux += video_scale;
682  }
683  }
684  }
685  }
686  }
687 
688  /* drawing spectogram/sonogram */
689  data += spectogram_start * linesize;
690  memcpy(data, s->spectogram + s->spectogram_index*linesize, total_length - back_length);
691 
692  data += total_length - back_length;
693  if (back_length)
694  memcpy(data, s->spectogram, back_length);
695 
696  s->outpicref->pts = s->frame_count;
697  ret = ff_filter_frame(outlink, av_frame_clone(s->outpicref));
698  s->req_fullfilled = 1;
699  s->frame_count++;
700  }
701  s->spectogram_count = (s->spectogram_count + 1) % s->count;
702  s->spectogram_index = (s->spectogram_index + spectogram_height - 1) % spectogram_height;
703  return ret;
704 }
705 
706 static int filter_frame(AVFilterLink *inlink, AVFrame *insamples)
707 {
708  AVFilterContext *ctx = inlink->dst;
709  ShowCQTContext *s = ctx->priv;
710  int step = inlink->sample_rate / (s->fps * s->count);
711  int fft_len = 1 << s->fft_bits;
712  int remaining;
713  float *audio_data;
714 
715  if (!insamples) {
716  while (s->remaining_fill < (fft_len >> 1)) {
717  int ret, x;
718  memset(&s->fft_data[fft_len - s->remaining_fill], 0, sizeof(*s->fft_data) * s->remaining_fill);
719  ret = plot_cqt(inlink);
720  if (ret < 0)
721  return ret;
722  for (x = 0; x < (fft_len-step); x++)
723  s->fft_data[x] = s->fft_data[x+step];
724  s->remaining_fill += step;
725  }
726  return AVERROR(EOF);
727  }
728 
729  remaining = insamples->nb_samples;
730  audio_data = (float*) insamples->data[0];
731 
732  while (remaining) {
733  if (remaining >= s->remaining_fill) {
734  int i = insamples->nb_samples - remaining;
735  int j = fft_len - s->remaining_fill;
736  int m, ret;
737  for (m = 0; m < s->remaining_fill; m++) {
738  s->fft_data[j+m].re = audio_data[2*(i+m)];
739  s->fft_data[j+m].im = audio_data[2*(i+m)+1];
740  }
741  ret = plot_cqt(inlink);
742  if (ret < 0) {
743  av_frame_free(&insamples);
744  return ret;
745  }
746  remaining -= s->remaining_fill;
747  for (m = 0; m < fft_len-step; m++)
748  s->fft_data[m] = s->fft_data[m+step];
749  s->remaining_fill = step;
750  } else {
751  int i = insamples->nb_samples - remaining;
752  int j = fft_len - s->remaining_fill;
753  int m;
754  for (m = 0; m < remaining; m++) {
755  s->fft_data[m+j].re = audio_data[2*(i+m)];
756  s->fft_data[m+j].im = audio_data[2*(i+m)+1];
757  }
758  s->remaining_fill -= remaining;
759  remaining = 0;
760  }
761  }
762  av_frame_free(&insamples);
763  return 0;
764 }
765 
766 static int request_frame(AVFilterLink *outlink)
767 {
768  ShowCQTContext *s = outlink->src->priv;
769  AVFilterLink *inlink = outlink->src->inputs[0];
770  int ret;
771 
772  s->req_fullfilled = 0;
773  do {
774  ret = ff_request_frame(inlink);
775  } while (!s->req_fullfilled && ret >= 0);
776 
777  if (ret == AVERROR_EOF && s->outpicref)
778  filter_frame(inlink, NULL);
779  return ret;
780 }
781 
782 static const AVFilterPad showcqt_inputs[] = {
783  {
784  .name = "default",
785  .type = AVMEDIA_TYPE_AUDIO,
786  .filter_frame = filter_frame,
787  },
788  { NULL }
789 };
790 
791 static const AVFilterPad showcqt_outputs[] = {
792  {
793  .name = "default",
794  .type = AVMEDIA_TYPE_VIDEO,
795  .config_props = config_output,
796  .request_frame = request_frame,
797  },
798  { NULL }
799 };
800 
802  .name = "showcqt",
803  .description = NULL_IF_CONFIG_SMALL("Convert input audio to a CQT (Constant Q Transform) spectrum video output."),
804  .uninit = uninit,
805  .query_formats = query_formats,
806  .priv_size = sizeof(ShowCQTContext),
807  .inputs = showcqt_inputs,
808  .outputs = showcqt_outputs,
809  .priv_class = &showcqt_class,
810 };