| 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" |
| 24 | #include "libavutil/channel_layout.h" |
| 25 | #include "libavutil/opt.h" |
| 26 | #include "libavutil/xga_font_data.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 { |
| 62 | FFTSample value; |
| 63 | int index; |
| 64 | } SparseCoeff; |
| 65 | |
| 66 | typedef struct { |
| 67 | const AVClass *class; |
| 68 | AVFrame *outpicref; |
| 69 | FFTContext *fft_context; |
| 70 | FFTComplex *fft_data; |
| 71 | FFTComplex *fft_result_left; |
| 72 | FFTComplex *fft_result_right; |
| 73 | uint8_t *spectogram; |
| 74 | SparseCoeff *coeff_sort; |
| 75 | SparseCoeff *coeffs[VIDEO_WIDTH]; |
| 76 | uint8_t *font_alpha; |
| 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; |
| 81 | int spectogram_count; |
| 82 | int spectogram_index; |
| 83 | int fft_bits; |
| 84 | int req_fullfilled; |
| 85 | int remaining_fill; |
| 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 | int fps; /* the required fps is so strict, so it's enough to be int, but 24000/1001 etc cannot be encoded */ |
| 94 | int count; /* fps * count = transform rate */ |
| 95 | } ShowCQTContext; |
| 96 | |
| 97 | #define OFFSET(x) offsetof(ShowCQTContext, x) |
| 98 | #define FLAGS AV_OPT_FLAG_FILTERING_PARAM|AV_OPT_FLAG_VIDEO_PARAM |
| 99 | |
| 100 | static const AVOption showcqt_options[] = { |
| 101 | { "volume", "set volume", OFFSET(volume), AV_OPT_TYPE_STRING, { .str = "16" }, CHAR_MIN, CHAR_MAX, FLAGS }, |
| 102 | { "tlength", "set transform length", OFFSET(tlength), AV_OPT_TYPE_STRING, { .str = TLENGTH_DEFAULT }, CHAR_MIN, CHAR_MAX, FLAGS }, |
| 103 | { "timeclamp", "set timeclamp", OFFSET(timeclamp), AV_OPT_TYPE_DOUBLE, { .dbl = 0.17 }, 0.1, 1.0, FLAGS }, |
| 104 | { "coeffclamp", "set coeffclamp", OFFSET(coeffclamp), AV_OPT_TYPE_FLOAT, { .dbl = 1 }, 0.1, 10, FLAGS }, |
| 105 | { "gamma", "set gamma", OFFSET(gamma), AV_OPT_TYPE_FLOAT, { .dbl = 3 }, 1, 7, FLAGS }, |
| 106 | { "fullhd", "set full HD resolution", OFFSET(fullhd), AV_OPT_TYPE_INT, { .i64 = 1 }, 0, 1, FLAGS }, |
| 107 | { "fps", "set video fps", OFFSET(fps), AV_OPT_TYPE_INT, { .i64 = 25 }, 10, 100, FLAGS }, |
| 108 | { "count", "set number of transform per frame", OFFSET(count), AV_OPT_TYPE_INT, { .i64 = 6 }, 1, 30, FLAGS }, |
| 109 | { "fontfile", "set font file", OFFSET(fontfile), AV_OPT_TYPE_STRING, { .str = NULL }, CHAR_MIN, CHAR_MAX, FLAGS }, |
| 110 | { "fontcolor", "set font color", OFFSET(fontcolor), AV_OPT_TYPE_STRING, { .str = FONTCOLOR_DEFAULT }, CHAR_MIN, CHAR_MAX, FLAGS }, |
| 111 | { NULL } |
| 112 | }; |
| 113 | |
| 114 | AVFILTER_DEFINE_CLASS(showcqt); |
| 115 | |
| 116 | static av_cold void uninit(AVFilterContext *ctx) |
| 117 | { |
| 118 | int k; |
| 119 | |
| 120 | ShowCQTContext *s = ctx->priv; |
| 121 | av_fft_end(s->fft_context); |
| 122 | s->fft_context = NULL; |
| 123 | for (k = 0; k < VIDEO_WIDTH; k++) |
| 124 | av_freep(&s->coeffs[k]); |
| 125 | av_freep(&s->fft_data); |
| 126 | av_freep(&s->fft_result_left); |
| 127 | av_freep(&s->fft_result_right); |
| 128 | av_freep(&s->coeff_sort); |
| 129 | av_freep(&s->spectogram); |
| 130 | av_freep(&s->font_alpha); |
| 131 | av_frame_free(&s->outpicref); |
| 132 | } |
| 133 | |
| 134 | static int query_formats(AVFilterContext *ctx) |
| 135 | { |
| 136 | AVFilterFormats *formats = NULL; |
| 137 | AVFilterChannelLayouts *layouts = NULL; |
| 138 | AVFilterLink *inlink = ctx->inputs[0]; |
| 139 | AVFilterLink *outlink = ctx->outputs[0]; |
| 140 | static const enum AVSampleFormat sample_fmts[] = { AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_NONE }; |
| 141 | static const enum AVPixelFormat pix_fmts[] = { AV_PIX_FMT_RGB24, AV_PIX_FMT_NONE }; |
| 142 | static const int64_t channel_layouts[] = { AV_CH_LAYOUT_STEREO, AV_CH_LAYOUT_STEREO_DOWNMIX, -1 }; |
| 143 | static const int samplerates[] = { 44100, 48000, -1 }; |
| 144 | |
| 145 | /* set input audio formats */ |
| 146 | formats = ff_make_format_list(sample_fmts); |
| 147 | if (!formats) |
| 148 | return AVERROR(ENOMEM); |
| 149 | ff_formats_ref(formats, &inlink->out_formats); |
| 150 | |
| 151 | layouts = avfilter_make_format64_list(channel_layouts); |
| 152 | if (!layouts) |
| 153 | return AVERROR(ENOMEM); |
| 154 | ff_channel_layouts_ref(layouts, &inlink->out_channel_layouts); |
| 155 | |
| 156 | formats = ff_make_format_list(samplerates); |
| 157 | if (!formats) |
| 158 | return AVERROR(ENOMEM); |
| 159 | ff_formats_ref(formats, &inlink->out_samplerates); |
| 160 | |
| 161 | /* set output video format */ |
| 162 | formats = ff_make_format_list(pix_fmts); |
| 163 | if (!formats) |
| 164 | return AVERROR(ENOMEM); |
| 165 | ff_formats_ref(formats, &outlink->in_formats); |
| 166 | |
| 167 | return 0; |
| 168 | } |
| 169 | |
| 170 | #if CONFIG_LIBFREETYPE |
| 171 | static void load_freetype_font(AVFilterContext *ctx) |
| 172 | { |
| 173 | static const char str[] = "EF G A BC D "; |
| 174 | ShowCQTContext *s = ctx->priv; |
| 175 | FT_Library lib = NULL; |
| 176 | FT_Face face = NULL; |
| 177 | int video_scale = s->fullhd ? 2 : 1; |
| 178 | int video_width = (VIDEO_WIDTH/2) * video_scale; |
| 179 | int font_height = (FONT_HEIGHT/2) * video_scale; |
| 180 | int font_width = 8 * video_scale; |
| 181 | int font_repeat = font_width * 12; |
| 182 | int linear_hori_advance = font_width * 65536; |
| 183 | int non_monospace_warning = 0; |
| 184 | int x; |
| 185 | |
| 186 | s->font_alpha = NULL; |
| 187 | |
| 188 | if (!s->fontfile) |
| 189 | return; |
| 190 | |
| 191 | if (FT_Init_FreeType(&lib)) |
| 192 | goto fail; |
| 193 | |
| 194 | if (FT_New_Face(lib, s->fontfile, 0, &face)) |
| 195 | goto fail; |
| 196 | |
| 197 | if (FT_Set_Char_Size(face, 16*64, 0, 0, 0)) |
| 198 | goto fail; |
| 199 | |
| 200 | if (FT_Load_Char(face, 'A', FT_LOAD_RENDER)) |
| 201 | goto fail; |
| 202 | |
| 203 | if (FT_Set_Char_Size(face, 16*64 * linear_hori_advance / face->glyph->linearHoriAdvance, 0, 0, 0)) |
| 204 | goto fail; |
| 205 | |
| 206 | s->font_alpha = av_malloc(font_height * video_width); |
| 207 | if (!s->font_alpha) |
| 208 | goto fail; |
| 209 | |
| 210 | memset(s->font_alpha, 0, font_height * video_width); |
| 211 | |
| 212 | for (x = 0; x < 12; x++) { |
| 213 | int sx, sy, rx, bx, by, dx, dy; |
| 214 | |
| 215 | if (str[x] == ' ') |
| 216 | continue; |
| 217 | |
| 218 | if (FT_Load_Char(face, str[x], FT_LOAD_RENDER)) |
| 219 | goto fail; |
| 220 | |
| 221 | if (face->glyph->advance.x != font_width*64 && !non_monospace_warning) { |
| 222 | av_log(ctx, AV_LOG_WARNING, "Font is not monospace\n"); |
| 223 | non_monospace_warning = 1; |
| 224 | } |
| 225 | |
| 226 | sy = font_height - 4*video_scale - face->glyph->bitmap_top; |
| 227 | for (rx = 0; rx < 10; rx++) { |
| 228 | sx = rx * font_repeat + x * font_width + face->glyph->bitmap_left; |
| 229 | for (by = 0; by < face->glyph->bitmap.rows; by++) { |
| 230 | dy = by + sy; |
| 231 | if (dy < 0) |
| 232 | continue; |
| 233 | if (dy >= font_height) |
| 234 | break; |
| 235 | |
| 236 | for (bx = 0; bx < face->glyph->bitmap.width; bx++) { |
| 237 | dx = bx + sx; |
| 238 | if (dx < 0) |
| 239 | continue; |
| 240 | if (dx >= video_width) |
| 241 | break; |
| 242 | s->font_alpha[dy*video_width+dx] = face->glyph->bitmap.buffer[by*face->glyph->bitmap.width+bx]; |
| 243 | } |
| 244 | } |
| 245 | } |
| 246 | } |
| 247 | |
| 248 | FT_Done_Face(face); |
| 249 | FT_Done_FreeType(lib); |
| 250 | return; |
| 251 | |
| 252 | fail: |
| 253 | av_log(ctx, AV_LOG_WARNING, "Error while loading freetype font, using default font instead\n"); |
| 254 | FT_Done_Face(face); |
| 255 | FT_Done_FreeType(lib); |
| 256 | av_freep(&s->font_alpha); |
| 257 | return; |
| 258 | } |
| 259 | #endif |
| 260 | |
| 261 | static double a_weighting(void *p, double f) |
| 262 | { |
| 263 | double ret = 12200.0*12200.0 * (f*f*f*f); |
| 264 | ret /= (f*f + 20.6*20.6) * (f*f + 12200.0*12200.0) * |
| 265 | sqrt((f*f + 107.7*107.7) * (f*f + 737.9*737.9)); |
| 266 | return ret; |
| 267 | } |
| 268 | |
| 269 | static double b_weighting(void *p, double f) |
| 270 | { |
| 271 | double ret = 12200.0*12200.0 * (f*f*f); |
| 272 | ret /= (f*f + 20.6*20.6) * (f*f + 12200.0*12200.0) * sqrt(f*f + 158.5*158.5); |
| 273 | return ret; |
| 274 | } |
| 275 | |
| 276 | static double c_weighting(void *p, double f) |
| 277 | { |
| 278 | double ret = 12200.0*12200.0 * (f*f); |
| 279 | ret /= (f*f + 20.6*20.6) * (f*f + 12200.0*12200.0); |
| 280 | return ret; |
| 281 | } |
| 282 | |
| 283 | static double midi(void *p, double f) |
| 284 | { |
| 285 | return log2(f/440.0) * 12.0 + 69.0; |
| 286 | } |
| 287 | |
| 288 | static double r_func(void *p, double x) |
| 289 | { |
| 290 | x = av_clipd(x, 0.0, 1.0); |
| 291 | return (int)(x*255.0+0.5) << 16; |
| 292 | } |
| 293 | |
| 294 | static double g_func(void *p, double x) |
| 295 | { |
| 296 | x = av_clipd(x, 0.0, 1.0); |
| 297 | return (int)(x*255.0+0.5) << 8; |
| 298 | } |
| 299 | |
| 300 | static double b_func(void *p, double x) |
| 301 | { |
| 302 | x = av_clipd(x, 0.0, 1.0); |
| 303 | return (int)(x*255.0+0.5); |
| 304 | } |
| 305 | |
| 306 | static inline int qsort_sparsecoeff(const SparseCoeff *a, const SparseCoeff *b) |
| 307 | { |
| 308 | if (fabsf(a->value) >= fabsf(b->value)) |
| 309 | return 1; |
| 310 | else |
| 311 | return -1; |
| 312 | } |
| 313 | |
| 314 | static int config_output(AVFilterLink *outlink) |
| 315 | { |
| 316 | AVFilterContext *ctx = outlink->src; |
| 317 | AVFilterLink *inlink = ctx->inputs[0]; |
| 318 | ShowCQTContext *s = ctx->priv; |
| 319 | AVExpr *tlength_expr = NULL, *volume_expr = NULL, *fontcolor_expr = NULL; |
| 320 | uint8_t *fontcolor_value = s->fontcolor_value; |
| 321 | static const char * const expr_vars[] = { "timeclamp", "tc", "frequency", "freq", "f", NULL }; |
| 322 | static const char * const expr_func_names[] = { "a_weighting", "b_weighting", "c_weighting", NULL }; |
| 323 | static const char * const expr_fontcolor_func_names[] = { "midi", "r", "g", "b", NULL }; |
| 324 | static double (* const expr_funcs[])(void *, double) = { a_weighting, b_weighting, c_weighting, NULL }; |
| 325 | static double (* const expr_fontcolor_funcs[])(void *, double) = { midi, r_func, g_func, b_func, NULL }; |
| 326 | int fft_len, k, x, y, ret; |
| 327 | int num_coeffs = 0; |
| 328 | int rate = inlink->sample_rate; |
| 329 | double max_len = rate * (double) s->timeclamp; |
| 330 | int64_t start_time, end_time; |
| 331 | int video_scale = s->fullhd ? 2 : 1; |
| 332 | int video_width = (VIDEO_WIDTH/2) * video_scale; |
| 333 | int video_height = (VIDEO_HEIGHT/2) * video_scale; |
| 334 | int spectogram_height = (SPECTOGRAM_HEIGHT/2) * video_scale; |
| 335 | |
| 336 | s->fft_bits = ceil(log2(max_len)); |
| 337 | fft_len = 1 << s->fft_bits; |
| 338 | |
| 339 | if (rate % (s->fps * s->count)) { |
| 340 | av_log(ctx, AV_LOG_ERROR, "Rate (%u) is not divisible by fps*count (%u*%u)\n", rate, s->fps, s->count); |
| 341 | return AVERROR(EINVAL); |
| 342 | } |
| 343 | |
| 344 | s->fft_data = av_malloc_array(fft_len, sizeof(*s->fft_data)); |
| 345 | s->coeff_sort = av_malloc_array(fft_len, sizeof(*s->coeff_sort)); |
| 346 | s->fft_result_left = av_malloc_array(fft_len, sizeof(*s->fft_result_left)); |
| 347 | s->fft_result_right = av_malloc_array(fft_len, sizeof(*s->fft_result_right)); |
| 348 | s->fft_context = av_fft_init(s->fft_bits, 0); |
| 349 | |
| 350 | if (!s->fft_data || !s->coeff_sort || !s->fft_result_left || !s->fft_result_right || !s->fft_context) |
| 351 | return AVERROR(ENOMEM); |
| 352 | |
| 353 | #if CONFIG_LIBFREETYPE |
| 354 | load_freetype_font(ctx); |
| 355 | #else |
| 356 | if (s->fontfile) |
| 357 | av_log(ctx, AV_LOG_WARNING, "Freetype is not available, ignoring fontfile option\n"); |
| 358 | s->font_alpha = NULL; |
| 359 | #endif |
| 360 | |
| 361 | av_log(ctx, AV_LOG_INFO, "Calculating spectral kernel, please wait\n"); |
| 362 | start_time = av_gettime_relative(); |
| 363 | ret = av_expr_parse(&tlength_expr, s->tlength, expr_vars, NULL, NULL, NULL, NULL, 0, ctx); |
| 364 | if (ret < 0) |
| 365 | goto eval_error; |
| 366 | |
| 367 | ret = av_expr_parse(&volume_expr, s->volume, expr_vars, expr_func_names, |
| 368 | expr_funcs, NULL, NULL, 0, ctx); |
| 369 | if (ret < 0) |
| 370 | goto eval_error; |
| 371 | |
| 372 | ret = av_expr_parse(&fontcolor_expr, s->fontcolor, expr_vars, expr_fontcolor_func_names, |
| 373 | expr_fontcolor_funcs, NULL, NULL, 0, ctx); |
| 374 | if (ret < 0) |
| 375 | goto eval_error; |
| 376 | |
| 377 | for (k = 0; k < VIDEO_WIDTH; k++) { |
| 378 | int hlen = fft_len >> 1; |
| 379 | float total = 0; |
| 380 | float partial = 0; |
| 381 | double freq = BASE_FREQ * exp2(k * (1.0/192.0)); |
| 382 | double tlen, tlength, volume; |
| 383 | double expr_vars_val[] = { s->timeclamp, s->timeclamp, freq, freq, freq, 0 }; |
| 384 | /* a window function from Albert H. Nuttall, |
| 385 | * "Some Windows with Very Good Sidelobe Behavior" |
| 386 | * -93.32 dB peak sidelobe and 18 dB/octave asymptotic decay |
| 387 | * coefficient normalized to a0 = 1 */ |
| 388 | double a0 = 0.355768; |
| 389 | double a1 = 0.487396/a0; |
| 390 | double a2 = 0.144232/a0; |
| 391 | double a3 = 0.012604/a0; |
| 392 | double sv_step, cv_step, sv, cv; |
| 393 | double sw_step, cw_step, sw, cw, w; |
| 394 | |
| 395 | tlength = av_expr_eval(tlength_expr, expr_vars_val, NULL); |
| 396 | if (isnan(tlength)) { |
| 397 | av_log(ctx, AV_LOG_WARNING, "at freq %g: tlength is nan, setting it to %g\n", freq, s->timeclamp); |
| 398 | tlength = s->timeclamp; |
| 399 | } else if (tlength < TLENGTH_MIN) { |
| 400 | av_log(ctx, AV_LOG_WARNING, "at freq %g: tlength is %g, setting it to %g\n", freq, tlength, TLENGTH_MIN); |
| 401 | tlength = TLENGTH_MIN; |
| 402 | } else if (tlength > s->timeclamp) { |
| 403 | av_log(ctx, AV_LOG_WARNING, "at freq %g: tlength is %g, setting it to %g\n", freq, tlength, s->timeclamp); |
| 404 | tlength = s->timeclamp; |
| 405 | } |
| 406 | |
| 407 | volume = FFABS(av_expr_eval(volume_expr, expr_vars_val, NULL)); |
| 408 | if (isnan(volume)) { |
| 409 | av_log(ctx, AV_LOG_WARNING, "at freq %g: volume is nan, setting it to 0\n", freq); |
| 410 | volume = VOLUME_MIN; |
| 411 | } else if (volume < VOLUME_MIN) { |
| 412 | volume = VOLUME_MIN; |
| 413 | } else if (volume > VOLUME_MAX) { |
| 414 | av_log(ctx, AV_LOG_WARNING, "at freq %g: volume is %g, setting it to %g\n", freq, volume, VOLUME_MAX); |
| 415 | volume = VOLUME_MAX; |
| 416 | } |
| 417 | |
| 418 | if (s->fullhd || !(k & 1)) { |
| 419 | int fontcolor = av_expr_eval(fontcolor_expr, expr_vars_val, NULL); |
| 420 | fontcolor_value[0] = (fontcolor >> 16) & 0xFF; |
| 421 | fontcolor_value[1] = (fontcolor >> 8) & 0xFF; |
| 422 | fontcolor_value[2] = fontcolor & 0xFF; |
| 423 | fontcolor_value += 3; |
| 424 | } |
| 425 | |
| 426 | tlen = tlength * rate; |
| 427 | s->fft_data[0].re = 0; |
| 428 | s->fft_data[0].im = 0; |
| 429 | s->fft_data[hlen].re = (1.0 + a1 + a2 + a3) * (1.0/tlen) * volume * (1.0/fft_len); |
| 430 | s->fft_data[hlen].im = 0; |
| 431 | sv_step = sv = sin(2.0*M_PI*freq*(1.0/rate)); |
| 432 | cv_step = cv = cos(2.0*M_PI*freq*(1.0/rate)); |
| 433 | /* also optimizing window func */ |
| 434 | sw_step = sw = sin(2.0*M_PI*(1.0/tlen)); |
| 435 | cw_step = cw = cos(2.0*M_PI*(1.0/tlen)); |
| 436 | for (x = 1; x < 0.5 * tlen; x++) { |
| 437 | double cv_tmp, cw_tmp; |
| 438 | double cw2, cw3, sw2; |
| 439 | |
| 440 | cw2 = cw * cw - sw * sw; |
| 441 | sw2 = cw * sw + sw * cw; |
| 442 | cw3 = cw * cw2 - sw * sw2; |
| 443 | w = (1.0 + a1 * cw + a2 * cw2 + a3 * cw3) * (1.0/tlen) * volume * (1.0/fft_len); |
| 444 | s->fft_data[hlen + x].re = w * cv; |
| 445 | s->fft_data[hlen + x].im = w * sv; |
| 446 | s->fft_data[hlen - x].re = s->fft_data[hlen + x].re; |
| 447 | s->fft_data[hlen - x].im = -s->fft_data[hlen + x].im; |
| 448 | |
| 449 | cv_tmp = cv * cv_step - sv * sv_step; |
| 450 | sv = sv * cv_step + cv * sv_step; |
| 451 | cv = cv_tmp; |
| 452 | cw_tmp = cw * cw_step - sw * sw_step; |
| 453 | sw = sw * cw_step + cw * sw_step; |
| 454 | cw = cw_tmp; |
| 455 | } |
| 456 | for (; x < hlen; x++) { |
| 457 | s->fft_data[hlen + x].re = 0; |
| 458 | s->fft_data[hlen + x].im = 0; |
| 459 | s->fft_data[hlen - x].re = 0; |
| 460 | s->fft_data[hlen - x].im = 0; |
| 461 | } |
| 462 | av_fft_permute(s->fft_context, s->fft_data); |
| 463 | av_fft_calc(s->fft_context, s->fft_data); |
| 464 | |
| 465 | for (x = 0; x < fft_len; x++) { |
| 466 | s->coeff_sort[x].index = x; |
| 467 | s->coeff_sort[x].value = s->fft_data[x].re; |
| 468 | } |
| 469 | |
| 470 | AV_QSORT(s->coeff_sort, fft_len, SparseCoeff, qsort_sparsecoeff); |
| 471 | for (x = 0; x < fft_len; x++) |
| 472 | total += fabsf(s->coeff_sort[x].value); |
| 473 | |
| 474 | for (x = 0; x < fft_len; x++) { |
| 475 | partial += fabsf(s->coeff_sort[x].value); |
| 476 | if (partial > total * s->coeffclamp * COEFF_CLAMP) { |
| 477 | s->coeffs_len[k] = fft_len - x; |
| 478 | num_coeffs += s->coeffs_len[k]; |
| 479 | s->coeffs[k] = av_malloc_array(s->coeffs_len[k], sizeof(*s->coeffs[k])); |
| 480 | if (!s->coeffs[k]) { |
| 481 | ret = AVERROR(ENOMEM); |
| 482 | goto eval_error; |
| 483 | } |
| 484 | for (y = 0; y < s->coeffs_len[k]; y++) |
| 485 | s->coeffs[k][y] = s->coeff_sort[x+y]; |
| 486 | break; |
| 487 | } |
| 488 | } |
| 489 | } |
| 490 | av_expr_free(fontcolor_expr); |
| 491 | av_expr_free(volume_expr); |
| 492 | av_expr_free(tlength_expr); |
| 493 | end_time = av_gettime_relative(); |
| 494 | 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); |
| 495 | |
| 496 | outlink->w = video_width; |
| 497 | outlink->h = video_height; |
| 498 | |
| 499 | s->req_fullfilled = 0; |
| 500 | s->spectogram_index = 0; |
| 501 | s->frame_count = 0; |
| 502 | s->spectogram_count = 0; |
| 503 | s->remaining_fill = fft_len >> 1; |
| 504 | memset(s->fft_data, 0, fft_len * sizeof(*s->fft_data)); |
| 505 | |
| 506 | s->outpicref = ff_get_video_buffer(outlink, outlink->w, outlink->h); |
| 507 | if (!s->outpicref) |
| 508 | return AVERROR(ENOMEM); |
| 509 | |
| 510 | s->spectogram = av_calloc(spectogram_height, s->outpicref->linesize[0]); |
| 511 | if (!s->spectogram) |
| 512 | return AVERROR(ENOMEM); |
| 513 | |
| 514 | outlink->sample_aspect_ratio = av_make_q(1, 1); |
| 515 | outlink->time_base = av_make_q(1, s->fps); |
| 516 | outlink->frame_rate = av_make_q(s->fps, 1); |
| 517 | return 0; |
| 518 | |
| 519 | eval_error: |
| 520 | av_expr_free(fontcolor_expr); |
| 521 | av_expr_free(volume_expr); |
| 522 | av_expr_free(tlength_expr); |
| 523 | return ret; |
| 524 | } |
| 525 | |
| 526 | static int plot_cqt(AVFilterLink *inlink) |
| 527 | { |
| 528 | AVFilterContext *ctx = inlink->dst; |
| 529 | ShowCQTContext *s = ctx->priv; |
| 530 | AVFilterLink *outlink = ctx->outputs[0]; |
| 531 | int fft_len = 1 << s->fft_bits; |
| 532 | FFTSample result[VIDEO_WIDTH][4]; |
| 533 | int x, y, ret = 0; |
| 534 | int linesize = s->outpicref->linesize[0]; |
| 535 | int video_scale = s->fullhd ? 2 : 1; |
| 536 | int video_width = (VIDEO_WIDTH/2) * video_scale; |
| 537 | int spectogram_height = (SPECTOGRAM_HEIGHT/2) * video_scale; |
| 538 | int spectogram_start = (SPECTOGRAM_START/2) * video_scale; |
| 539 | int font_height = (FONT_HEIGHT/2) * video_scale; |
| 540 | |
| 541 | /* real part contains left samples, imaginary part contains right samples */ |
| 542 | memcpy(s->fft_result_left, s->fft_data, fft_len * sizeof(*s->fft_data)); |
| 543 | av_fft_permute(s->fft_context, s->fft_result_left); |
| 544 | av_fft_calc(s->fft_context, s->fft_result_left); |
| 545 | |
| 546 | /* separate left and right, (and multiply by 2.0) */ |
| 547 | s->fft_result_right[0].re = 2.0f * s->fft_result_left[0].im; |
| 548 | s->fft_result_right[0].im = 0; |
| 549 | s->fft_result_left[0].re = 2.0f * s->fft_result_left[0].re; |
| 550 | s->fft_result_left[0].im = 0; |
| 551 | for (x = 1; x <= fft_len >> 1; x++) { |
| 552 | FFTSample tmpy = s->fft_result_left[fft_len-x].im - s->fft_result_left[x].im; |
| 553 | |
| 554 | s->fft_result_right[x].re = s->fft_result_left[x].im + s->fft_result_left[fft_len-x].im; |
| 555 | s->fft_result_right[x].im = s->fft_result_left[x].re - s->fft_result_left[fft_len-x].re; |
| 556 | s->fft_result_right[fft_len-x].re = s->fft_result_right[x].re; |
| 557 | s->fft_result_right[fft_len-x].im = -s->fft_result_right[x].im; |
| 558 | |
| 559 | s->fft_result_left[x].re = s->fft_result_left[x].re + s->fft_result_left[fft_len-x].re; |
| 560 | s->fft_result_left[x].im = tmpy; |
| 561 | s->fft_result_left[fft_len-x].re = s->fft_result_left[x].re; |
| 562 | s->fft_result_left[fft_len-x].im = -s->fft_result_left[x].im; |
| 563 | } |
| 564 | |
| 565 | /* calculating cqt */ |
| 566 | for (x = 0; x < VIDEO_WIDTH; x++) { |
| 567 | int u; |
| 568 | float g = 1.0f / s->gamma; |
| 569 | FFTComplex l = {0,0}; |
| 570 | FFTComplex r = {0,0}; |
| 571 | |
| 572 | for (u = 0; u < s->coeffs_len[x]; u++) { |
| 573 | FFTSample value = s->coeffs[x][u].value; |
| 574 | int index = s->coeffs[x][u].index; |
| 575 | l.re += value * s->fft_result_left[index].re; |
| 576 | l.im += value * s->fft_result_left[index].im; |
| 577 | r.re += value * s->fft_result_right[index].re; |
| 578 | r.im += value * s->fft_result_right[index].im; |
| 579 | } |
| 580 | /* result is power, not amplitude */ |
| 581 | result[x][0] = l.re * l.re + l.im * l.im; |
| 582 | result[x][2] = r.re * r.re + r.im * r.im; |
| 583 | result[x][1] = 0.5f * (result[x][0] + result[x][2]); |
| 584 | result[x][3] = result[x][1]; |
| 585 | result[x][0] = 255.0f * powf(FFMIN(1.0f,result[x][0]), g); |
| 586 | result[x][1] = 255.0f * powf(FFMIN(1.0f,result[x][1]), g); |
| 587 | result[x][2] = 255.0f * powf(FFMIN(1.0f,result[x][2]), g); |
| 588 | } |
| 589 | |
| 590 | if (!s->fullhd) { |
| 591 | for (x = 0; x < video_width; x++) { |
| 592 | result[x][0] = 0.5f * (result[2*x][0] + result[2*x+1][0]); |
| 593 | result[x][1] = 0.5f * (result[2*x][1] + result[2*x+1][1]); |
| 594 | result[x][2] = 0.5f * (result[2*x][2] + result[2*x+1][2]); |
| 595 | result[x][3] = 0.5f * (result[2*x][3] + result[2*x+1][3]); |
| 596 | } |
| 597 | } |
| 598 | |
| 599 | for (x = 0; x < video_width; x++) { |
| 600 | s->spectogram[s->spectogram_index*linesize + 3*x] = result[x][0] + 0.5f; |
| 601 | s->spectogram[s->spectogram_index*linesize + 3*x + 1] = result[x][1] + 0.5f; |
| 602 | s->spectogram[s->spectogram_index*linesize + 3*x + 2] = result[x][2] + 0.5f; |
| 603 | } |
| 604 | |
| 605 | /* drawing */ |
| 606 | if (!s->spectogram_count) { |
| 607 | uint8_t *data = (uint8_t*) s->outpicref->data[0]; |
| 608 | float rcp_result[VIDEO_WIDTH]; |
| 609 | int total_length = linesize * spectogram_height; |
| 610 | int back_length = linesize * s->spectogram_index; |
| 611 | |
| 612 | for (x = 0; x < video_width; x++) |
| 613 | rcp_result[x] = 1.0f / (result[x][3]+0.0001f); |
| 614 | |
| 615 | /* drawing bar */ |
| 616 | for (y = 0; y < spectogram_height; y++) { |
| 617 | float height = (spectogram_height - y) * (1.0f/spectogram_height); |
| 618 | uint8_t *lineptr = data + y * linesize; |
| 619 | for (x = 0; x < video_width; x++) { |
| 620 | float mul; |
| 621 | if (result[x][3] <= height) { |
| 622 | *lineptr++ = 0; |
| 623 | *lineptr++ = 0; |
| 624 | *lineptr++ = 0; |
| 625 | } else { |
| 626 | mul = (result[x][3] - height) * rcp_result[x]; |
| 627 | *lineptr++ = mul * result[x][0] + 0.5f; |
| 628 | *lineptr++ = mul * result[x][1] + 0.5f; |
| 629 | *lineptr++ = mul * result[x][2] + 0.5f; |
| 630 | } |
| 631 | } |
| 632 | } |
| 633 | |
| 634 | /* drawing font */ |
| 635 | if (s->font_alpha) { |
| 636 | for (y = 0; y < font_height; y++) { |
| 637 | uint8_t *lineptr = data + (spectogram_height + y) * linesize; |
| 638 | uint8_t *spectogram_src = s->spectogram + s->spectogram_index * linesize; |
| 639 | uint8_t *fontcolor_value = s->fontcolor_value; |
| 640 | for (x = 0; x < video_width; x++) { |
| 641 | uint8_t alpha = s->font_alpha[y*video_width+x]; |
| 642 | lineptr[3*x] = (spectogram_src[3*x] * (255-alpha) + fontcolor_value[0] * alpha + 255) >> 8; |
| 643 | lineptr[3*x+1] = (spectogram_src[3*x+1] * (255-alpha) + fontcolor_value[1] * alpha + 255) >> 8; |
| 644 | lineptr[3*x+2] = (spectogram_src[3*x+2] * (255-alpha) + fontcolor_value[2] * alpha + 255) >> 8; |
| 645 | fontcolor_value += 3; |
| 646 | } |
| 647 | } |
| 648 | } else { |
| 649 | for (y = 0; y < font_height; y++) { |
| 650 | uint8_t *lineptr = data + (spectogram_height + y) * linesize; |
| 651 | memcpy(lineptr, s->spectogram + s->spectogram_index * linesize, video_width*3); |
| 652 | } |
| 653 | for (x = 0; x < video_width; x += video_width/10) { |
| 654 | int u; |
| 655 | static const char str[] = "EF G A BC D "; |
| 656 | uint8_t *startptr = data + spectogram_height * linesize + x * 3; |
| 657 | for (u = 0; str[u]; u++) { |
| 658 | int v; |
| 659 | for (v = 0; v < 16; v++) { |
| 660 | uint8_t *p = startptr + v * linesize * video_scale + 8 * 3 * u * video_scale; |
| 661 | int ux = x + 8 * u * video_scale; |
| 662 | int mask; |
| 663 | for (mask = 0x80; mask; mask >>= 1) { |
| 664 | if (mask & avpriv_vga16_font[str[u] * 16 + v]) { |
| 665 | p[0] = s->fontcolor_value[3*ux]; |
| 666 | p[1] = s->fontcolor_value[3*ux+1]; |
| 667 | p[2] = s->fontcolor_value[3*ux+2]; |
| 668 | if (video_scale == 2) { |
| 669 | p[linesize] = p[0]; |
| 670 | p[linesize+1] = p[1]; |
| 671 | p[linesize+2] = p[2]; |
| 672 | p[3] = p[linesize+3] = s->fontcolor_value[3*ux+3]; |
| 673 | p[4] = p[linesize+4] = s->fontcolor_value[3*ux+4]; |
| 674 | p[5] = p[linesize+5] = s->fontcolor_value[3*ux+5]; |
| 675 | } |
| 676 | } |
| 677 | p += 3 * video_scale; |
| 678 | ux += video_scale; |
| 679 | } |
| 680 | } |
| 681 | } |
| 682 | } |
| 683 | } |
| 684 | |
| 685 | /* drawing spectogram/sonogram */ |
| 686 | data += spectogram_start * linesize; |
| 687 | memcpy(data, s->spectogram + s->spectogram_index*linesize, total_length - back_length); |
| 688 | |
| 689 | data += total_length - back_length; |
| 690 | if (back_length) |
| 691 | memcpy(data, s->spectogram, back_length); |
| 692 | |
| 693 | s->outpicref->pts = s->frame_count; |
| 694 | ret = ff_filter_frame(outlink, av_frame_clone(s->outpicref)); |
| 695 | s->req_fullfilled = 1; |
| 696 | s->frame_count++; |
| 697 | } |
| 698 | s->spectogram_count = (s->spectogram_count + 1) % s->count; |
| 699 | s->spectogram_index = (s->spectogram_index + spectogram_height - 1) % spectogram_height; |
| 700 | return ret; |
| 701 | } |
| 702 | |
| 703 | static int filter_frame(AVFilterLink *inlink, AVFrame *insamples) |
| 704 | { |
| 705 | AVFilterContext *ctx = inlink->dst; |
| 706 | ShowCQTContext *s = ctx->priv; |
| 707 | int step = inlink->sample_rate / (s->fps * s->count); |
| 708 | int fft_len = 1 << s->fft_bits; |
| 709 | int remaining; |
| 710 | float *audio_data; |
| 711 | |
| 712 | if (!insamples) { |
| 713 | while (s->remaining_fill < (fft_len >> 1)) { |
| 714 | int ret, x; |
| 715 | memset(&s->fft_data[fft_len - s->remaining_fill], 0, sizeof(*s->fft_data) * s->remaining_fill); |
| 716 | ret = plot_cqt(inlink); |
| 717 | if (ret < 0) |
| 718 | return ret; |
| 719 | for (x = 0; x < (fft_len-step); x++) |
| 720 | s->fft_data[x] = s->fft_data[x+step]; |
| 721 | s->remaining_fill += step; |
| 722 | } |
| 723 | return AVERROR(EOF); |
| 724 | } |
| 725 | |
| 726 | remaining = insamples->nb_samples; |
| 727 | audio_data = (float*) insamples->data[0]; |
| 728 | |
| 729 | while (remaining) { |
| 730 | if (remaining >= s->remaining_fill) { |
| 731 | int i = insamples->nb_samples - remaining; |
| 732 | int j = fft_len - s->remaining_fill; |
| 733 | int m, ret; |
| 734 | for (m = 0; m < s->remaining_fill; m++) { |
| 735 | s->fft_data[j+m].re = audio_data[2*(i+m)]; |
| 736 | s->fft_data[j+m].im = audio_data[2*(i+m)+1]; |
| 737 | } |
| 738 | ret = plot_cqt(inlink); |
| 739 | if (ret < 0) { |
| 740 | av_frame_free(&insamples); |
| 741 | return ret; |
| 742 | } |
| 743 | remaining -= s->remaining_fill; |
| 744 | for (m = 0; m < fft_len-step; m++) |
| 745 | s->fft_data[m] = s->fft_data[m+step]; |
| 746 | s->remaining_fill = step; |
| 747 | } else { |
| 748 | int i = insamples->nb_samples - remaining; |
| 749 | int j = fft_len - s->remaining_fill; |
| 750 | int m; |
| 751 | for (m = 0; m < remaining; m++) { |
| 752 | s->fft_data[m+j].re = audio_data[2*(i+m)]; |
| 753 | s->fft_data[m+j].im = audio_data[2*(i+m)+1]; |
| 754 | } |
| 755 | s->remaining_fill -= remaining; |
| 756 | remaining = 0; |
| 757 | } |
| 758 | } |
| 759 | av_frame_free(&insamples); |
| 760 | return 0; |
| 761 | } |
| 762 | |
| 763 | static int request_frame(AVFilterLink *outlink) |
| 764 | { |
| 765 | ShowCQTContext *s = outlink->src->priv; |
| 766 | AVFilterLink *inlink = outlink->src->inputs[0]; |
| 767 | int ret; |
| 768 | |
| 769 | s->req_fullfilled = 0; |
| 770 | do { |
| 771 | ret = ff_request_frame(inlink); |
| 772 | } while (!s->req_fullfilled && ret >= 0); |
| 773 | |
| 774 | if (ret == AVERROR_EOF && s->outpicref) |
| 775 | filter_frame(inlink, NULL); |
| 776 | return ret; |
| 777 | } |
| 778 | |
| 779 | static const AVFilterPad showcqt_inputs[] = { |
| 780 | { |
| 781 | .name = "default", |
| 782 | .type = AVMEDIA_TYPE_AUDIO, |
| 783 | .filter_frame = filter_frame, |
| 784 | }, |
| 785 | { NULL } |
| 786 | }; |
| 787 | |
| 788 | static const AVFilterPad showcqt_outputs[] = { |
| 789 | { |
| 790 | .name = "default", |
| 791 | .type = AVMEDIA_TYPE_VIDEO, |
| 792 | .config_props = config_output, |
| 793 | .request_frame = request_frame, |
| 794 | }, |
| 795 | { NULL } |
| 796 | }; |
| 797 | |
| 798 | AVFilter ff_avf_showcqt = { |
| 799 | .name = "showcqt", |
| 800 | .description = NULL_IF_CONFIG_SMALL("Convert input audio to a CQT (Constant Q Transform) spectrum video output."), |
| 801 | .uninit = uninit, |
| 802 | .query_formats = query_formats, |
| 803 | .priv_size = sizeof(ShowCQTContext), |
| 804 | .inputs = showcqt_inputs, |
| 805 | .outputs = showcqt_outputs, |
| 806 | .priv_class = &showcqt_class, |
| 807 | }; |