| Line | Branch | Exec | Source |
|---|---|---|---|
| 1 | /* | ||
| 2 | * Copyright (c) 2018 Paul B Mahol | ||
| 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 "libavutil/audio_fifo.h" | ||
| 22 | #include "libavutil/mem.h" | ||
| 23 | #include "libavutil/opt.h" | ||
| 24 | #include "libavutil/tx.h" | ||
| 25 | #include "avfilter.h" | ||
| 26 | #include "audio.h" | ||
| 27 | #include "filters.h" | ||
| 28 | |||
| 29 | typedef struct DeclickChannel { | ||
| 30 | double *auxiliary; | ||
| 31 | double *detection; | ||
| 32 | double *acoefficients; | ||
| 33 | double *acorrelation; | ||
| 34 | double *tmp; | ||
| 35 | double *interpolated; | ||
| 36 | double *matrix; | ||
| 37 | int matrix_size; | ||
| 38 | double *vector; | ||
| 39 | int vector_size; | ||
| 40 | double *y; | ||
| 41 | int y_size; | ||
| 42 | uint8_t *click; | ||
| 43 | int *index; | ||
| 44 | unsigned *histogram; | ||
| 45 | int histogram_size; | ||
| 46 | } DeclickChannel; | ||
| 47 | |||
| 48 | typedef struct AudioDeclickContext { | ||
| 49 | const AVClass *class; | ||
| 50 | |||
| 51 | double w; | ||
| 52 | double overlap; | ||
| 53 | double threshold; | ||
| 54 | double ar; | ||
| 55 | double burst; | ||
| 56 | int method; | ||
| 57 | int nb_hbins; | ||
| 58 | |||
| 59 | int is_declip; | ||
| 60 | int ar_order; | ||
| 61 | int nb_burst_samples; | ||
| 62 | int window_size; | ||
| 63 | int hop_size; | ||
| 64 | int overlap_skip; | ||
| 65 | |||
| 66 | AVFrame *enabled; | ||
| 67 | AVFrame *in; | ||
| 68 | AVFrame *out; | ||
| 69 | AVFrame *buffer; | ||
| 70 | AVFrame *is; | ||
| 71 | |||
| 72 | DeclickChannel *chan; | ||
| 73 | |||
| 74 | int64_t pts; | ||
| 75 | int nb_channels; | ||
| 76 | uint64_t nb_samples; | ||
| 77 | uint64_t detected_errors; | ||
| 78 | int samples_left; | ||
| 79 | int eof; | ||
| 80 | |||
| 81 | AVAudioFifo *efifo; | ||
| 82 | AVAudioFifo *fifo; | ||
| 83 | double *window_func_lut; | ||
| 84 | |||
| 85 | int (*detector)(struct AudioDeclickContext *s, DeclickChannel *c, | ||
| 86 | double sigmae, double *detection, | ||
| 87 | double *acoefficients, uint8_t *click, int *index, | ||
| 88 | const double *src, double *dst); | ||
| 89 | } AudioDeclickContext; | ||
| 90 | |||
| 91 | #define OFFSET(x) offsetof(AudioDeclickContext, x) | ||
| 92 | #define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM | ||
| 93 | |||
| 94 | static const AVOption adeclick_options[] = { | ||
| 95 | { "window", "set window size", OFFSET(w), AV_OPT_TYPE_DOUBLE, {.dbl=55}, 10, 100, AF }, | ||
| 96 | { "w", "set window size", OFFSET(w), AV_OPT_TYPE_DOUBLE, {.dbl=55}, 10, 100, AF }, | ||
| 97 | { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_DOUBLE, {.dbl=75}, 50, 95, AF }, | ||
| 98 | { "o", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_DOUBLE, {.dbl=75}, 50, 95, AF }, | ||
| 99 | { "arorder", "set autoregression order", OFFSET(ar), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 0, 25, AF }, | ||
| 100 | { "a", "set autoregression order", OFFSET(ar), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 0, 25, AF }, | ||
| 101 | { "threshold", "set threshold", OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 1, 100, AF }, | ||
| 102 | { "t", "set threshold", OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 1, 100, AF }, | ||
| 103 | { "burst", "set burst fusion", OFFSET(burst), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 0, 10, AF }, | ||
| 104 | { "b", "set burst fusion", OFFSET(burst), AV_OPT_TYPE_DOUBLE, {.dbl=2}, 0, 10, AF }, | ||
| 105 | { "method", "set overlap method", OFFSET(method), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, .unit = "m" }, | ||
| 106 | { "m", "set overlap method", OFFSET(method), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, .unit = "m" }, | ||
| 107 | { "add", "overlap-add", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, .unit = "m" }, | ||
| 108 | { "a", "overlap-add", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, .unit = "m" }, | ||
| 109 | { "save", "overlap-save", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, .unit = "m" }, | ||
| 110 | { "s", "overlap-save", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, .unit = "m" }, | ||
| 111 | { NULL } | ||
| 112 | }; | ||
| 113 | |||
| 114 | AVFILTER_DEFINE_CLASS(adeclick); | ||
| 115 | |||
| 116 | ✗ | static int config_input(AVFilterLink *inlink) | |
| 117 | { | ||
| 118 | ✗ | AVFilterContext *ctx = inlink->dst; | |
| 119 | ✗ | AudioDeclickContext *s = ctx->priv; | |
| 120 | int i; | ||
| 121 | |||
| 122 | ✗ | s->pts = AV_NOPTS_VALUE; | |
| 123 | ✗ | s->window_size = FFMAX(100, inlink->sample_rate * s->w / 1000.); | |
| 124 | ✗ | s->ar_order = FFMAX(s->window_size * s->ar / 100., 1); | |
| 125 | ✗ | s->nb_burst_samples = s->window_size * s->burst / 1000.; | |
| 126 | ✗ | s->hop_size = FFMAX(1, s->window_size * (1. - (s->overlap / 100.))); | |
| 127 | |||
| 128 | ✗ | s->window_func_lut = av_calloc(s->window_size, sizeof(*s->window_func_lut)); | |
| 129 | ✗ | if (!s->window_func_lut) | |
| 130 | ✗ | return AVERROR(ENOMEM); | |
| 131 | |||
| 132 | { | ||
| 133 | double *tx_in[2], *tx_out[2]; | ||
| 134 | AVTXContext *tx, *itx; | ||
| 135 | av_tx_fn tx_fn, itx_fn; | ||
| 136 | int ret, tx_size; | ||
| 137 | double scale; | ||
| 138 | |||
| 139 | ✗ | tx_size = 1 << (32 - ff_clz(s->window_size)); | |
| 140 | |||
| 141 | ✗ | scale = 1.0; | |
| 142 | ✗ | ret = av_tx_init(&tx, &tx_fn, AV_TX_DOUBLE_RDFT, 0, tx_size, &scale, 0); | |
| 143 | ✗ | if (ret < 0) | |
| 144 | ✗ | return ret; | |
| 145 | |||
| 146 | ✗ | scale = 1.0 / tx_size; | |
| 147 | ✗ | ret = av_tx_init(&itx, &itx_fn, AV_TX_DOUBLE_RDFT, 1, tx_size, &scale, 0); | |
| 148 | ✗ | if (ret < 0) | |
| 149 | ✗ | return ret; | |
| 150 | |||
| 151 | ✗ | tx_in[0] = av_calloc(tx_size + 2, sizeof(*tx_in[0])); | |
| 152 | ✗ | tx_in[1] = av_calloc(tx_size + 2, sizeof(*tx_in[1])); | |
| 153 | ✗ | tx_out[0] = av_calloc(tx_size + 2, sizeof(*tx_out[0])); | |
| 154 | ✗ | tx_out[1] = av_calloc(tx_size + 2, sizeof(*tx_out[1])); | |
| 155 | ✗ | if (!tx_in[0] || !tx_in[1] || !tx_out[0] || !tx_out[1]) | |
| 156 | ✗ | return AVERROR(ENOMEM); | |
| 157 | |||
| 158 | ✗ | for (int n = 0; n < s->window_size - s->hop_size; n++) | |
| 159 | ✗ | tx_in[0][n] = 1.0; | |
| 160 | |||
| 161 | ✗ | for (int n = 0; n < s->hop_size; n++) | |
| 162 | ✗ | tx_in[1][n] = 1.0; | |
| 163 | |||
| 164 | ✗ | tx_fn(tx, tx_out[0], tx_in[0], sizeof(double)); | |
| 165 | ✗ | tx_fn(tx, tx_out[1], tx_in[1], sizeof(double)); | |
| 166 | |||
| 167 | ✗ | for (int n = 0; n <= tx_size/2; n++) { | |
| 168 | ✗ | double re0 = tx_out[0][2*n]; | |
| 169 | ✗ | double im0 = tx_out[0][2*n+1]; | |
| 170 | ✗ | double re1 = tx_out[1][2*n]; | |
| 171 | ✗ | double im1 = tx_out[1][2*n+1]; | |
| 172 | |||
| 173 | ✗ | tx_in[0][2*n] = re0 * re1 - im0 * im1; | |
| 174 | ✗ | tx_in[0][2*n+1] = re0 * im1 + re1 * im0; | |
| 175 | } | ||
| 176 | |||
| 177 | ✗ | itx_fn(itx, tx_out[0], tx_in[0], sizeof(AVComplexDouble)); | |
| 178 | |||
| 179 | ✗ | scale = 1.0 / (s->window_size - s->hop_size); | |
| 180 | ✗ | for (int n = 0; n < s->window_size; n++) | |
| 181 | ✗ | s->window_func_lut[n] = tx_out[0][n] * scale; | |
| 182 | |||
| 183 | ✗ | av_tx_uninit(&tx); | |
| 184 | ✗ | av_tx_uninit(&itx); | |
| 185 | |||
| 186 | ✗ | av_freep(&tx_in[0]); | |
| 187 | ✗ | av_freep(&tx_in[1]); | |
| 188 | ✗ | av_freep(&tx_out[0]); | |
| 189 | ✗ | av_freep(&tx_out[1]); | |
| 190 | } | ||
| 191 | |||
| 192 | ✗ | av_frame_free(&s->in); | |
| 193 | ✗ | av_frame_free(&s->out); | |
| 194 | ✗ | av_frame_free(&s->buffer); | |
| 195 | ✗ | av_frame_free(&s->is); | |
| 196 | ✗ | s->enabled = ff_get_audio_buffer(inlink, s->window_size); | |
| 197 | ✗ | s->in = ff_get_audio_buffer(inlink, s->window_size); | |
| 198 | ✗ | s->out = ff_get_audio_buffer(inlink, s->window_size); | |
| 199 | ✗ | s->buffer = ff_get_audio_buffer(inlink, s->window_size * 2); | |
| 200 | ✗ | s->is = ff_get_audio_buffer(inlink, s->window_size); | |
| 201 | ✗ | if (!s->in || !s->out || !s->buffer || !s->is || !s->enabled) | |
| 202 | ✗ | return AVERROR(ENOMEM); | |
| 203 | |||
| 204 | ✗ | s->efifo = av_audio_fifo_alloc(inlink->format, 1, s->window_size); | |
| 205 | ✗ | if (!s->efifo) | |
| 206 | ✗ | return AVERROR(ENOMEM); | |
| 207 | ✗ | s->fifo = av_audio_fifo_alloc(inlink->format, inlink->ch_layout.nb_channels, s->window_size); | |
| 208 | ✗ | if (!s->fifo) | |
| 209 | ✗ | return AVERROR(ENOMEM); | |
| 210 | ✗ | s->overlap_skip = s->method ? (s->window_size - s->hop_size) / 2 : 0; | |
| 211 | ✗ | if (s->overlap_skip > 0) { | |
| 212 | ✗ | av_audio_fifo_write(s->fifo, (void **)s->in->extended_data, | |
| 213 | s->overlap_skip); | ||
| 214 | } | ||
| 215 | |||
| 216 | ✗ | s->nb_channels = inlink->ch_layout.nb_channels; | |
| 217 | ✗ | s->chan = av_calloc(inlink->ch_layout.nb_channels, sizeof(*s->chan)); | |
| 218 | ✗ | if (!s->chan) | |
| 219 | ✗ | return AVERROR(ENOMEM); | |
| 220 | |||
| 221 | ✗ | for (i = 0; i < inlink->ch_layout.nb_channels; i++) { | |
| 222 | ✗ | DeclickChannel *c = &s->chan[i]; | |
| 223 | |||
| 224 | ✗ | c->detection = av_calloc(s->window_size, sizeof(*c->detection)); | |
| 225 | ✗ | c->auxiliary = av_calloc(s->ar_order + 1, sizeof(*c->auxiliary)); | |
| 226 | ✗ | c->acoefficients = av_calloc(s->ar_order + 1, sizeof(*c->acoefficients)); | |
| 227 | ✗ | c->acorrelation = av_calloc(s->ar_order + 1, sizeof(*c->acorrelation)); | |
| 228 | ✗ | c->tmp = av_calloc(s->ar_order, sizeof(*c->tmp)); | |
| 229 | ✗ | c->click = av_calloc(s->window_size, sizeof(*c->click)); | |
| 230 | ✗ | c->index = av_calloc(s->window_size, sizeof(*c->index)); | |
| 231 | ✗ | c->interpolated = av_calloc(s->window_size, sizeof(*c->interpolated)); | |
| 232 | ✗ | if (!c->auxiliary || !c->acoefficients || !c->detection || !c->click || | |
| 233 | ✗ | !c->index || !c->interpolated || !c->acorrelation || !c->tmp) | |
| 234 | ✗ | return AVERROR(ENOMEM); | |
| 235 | } | ||
| 236 | |||
| 237 | ✗ | return 0; | |
| 238 | } | ||
| 239 | |||
| 240 | ✗ | static void autocorrelation(const double *input, int order, int size, | |
| 241 | double *output, double scale) | ||
| 242 | { | ||
| 243 | int i, j; | ||
| 244 | |||
| 245 | ✗ | for (i = 0; i <= order; i++) { | |
| 246 | ✗ | double value = 0.; | |
| 247 | |||
| 248 | ✗ | for (j = i; j < size; j++) | |
| 249 | ✗ | value += input[j] * input[j - i]; | |
| 250 | |||
| 251 | ✗ | output[i] = value * scale; | |
| 252 | } | ||
| 253 | ✗ | } | |
| 254 | |||
| 255 | ✗ | static double autoregression(const double *samples, int ar_order, | |
| 256 | int nb_samples, double *k, double *r, double *a) | ||
| 257 | { | ||
| 258 | double alpha; | ||
| 259 | int i, j; | ||
| 260 | |||
| 261 | ✗ | memset(a, 0, ar_order * sizeof(*a)); | |
| 262 | |||
| 263 | ✗ | autocorrelation(samples, ar_order, nb_samples, r, 1. / nb_samples); | |
| 264 | |||
| 265 | /* Levinson-Durbin algorithm */ | ||
| 266 | ✗ | k[0] = a[0] = -r[1] / r[0]; | |
| 267 | ✗ | alpha = r[0] * (1. - k[0] * k[0]); | |
| 268 | ✗ | for (i = 1; i < ar_order; i++) { | |
| 269 | ✗ | double epsilon = 0.; | |
| 270 | |||
| 271 | ✗ | for (j = 0; j < i; j++) | |
| 272 | ✗ | epsilon += a[j] * r[i - j]; | |
| 273 | ✗ | epsilon += r[i + 1]; | |
| 274 | |||
| 275 | ✗ | k[i] = -epsilon / alpha; | |
| 276 | ✗ | alpha *= (1. - k[i] * k[i]); | |
| 277 | ✗ | for (j = i - 1; j >= 0; j--) | |
| 278 | ✗ | k[j] = a[j] + k[i] * a[i - j - 1]; | |
| 279 | ✗ | for (j = 0; j <= i; j++) | |
| 280 | ✗ | a[j] = k[j]; | |
| 281 | } | ||
| 282 | |||
| 283 | ✗ | k[0] = 1.; | |
| 284 | ✗ | for (i = 1; i <= ar_order; i++) | |
| 285 | ✗ | k[i] = a[i - 1]; | |
| 286 | |||
| 287 | ✗ | return sqrt(alpha); | |
| 288 | } | ||
| 289 | |||
| 290 | ✗ | static int isfinite_array(double *samples, int nb_samples) | |
| 291 | { | ||
| 292 | int i; | ||
| 293 | |||
| 294 | ✗ | for (i = 0; i < nb_samples; i++) | |
| 295 | ✗ | if (!isfinite(samples[i])) | |
| 296 | ✗ | return 0; | |
| 297 | |||
| 298 | ✗ | return 1; | |
| 299 | } | ||
| 300 | |||
| 301 | ✗ | static int find_index(int *index, int value, int size) | |
| 302 | { | ||
| 303 | int i, start, end; | ||
| 304 | |||
| 305 | ✗ | if ((value < index[0]) || (value > index[size - 1])) | |
| 306 | ✗ | return 1; | |
| 307 | |||
| 308 | ✗ | i = start = 0; | |
| 309 | ✗ | end = size - 1; | |
| 310 | |||
| 311 | ✗ | while (start <= end) { | |
| 312 | ✗ | i = (end + start) / 2; | |
| 313 | ✗ | if (index[i] == value) | |
| 314 | ✗ | return 0; | |
| 315 | ✗ | if (value < index[i]) | |
| 316 | ✗ | end = i - 1; | |
| 317 | ✗ | if (value > index[i]) | |
| 318 | ✗ | start = i + 1; | |
| 319 | } | ||
| 320 | |||
| 321 | ✗ | return 1; | |
| 322 | } | ||
| 323 | |||
| 324 | ✗ | static int factorization(double *matrix, int n) | |
| 325 | { | ||
| 326 | int i, j, k; | ||
| 327 | |||
| 328 | ✗ | for (i = 0; i < n; i++) { | |
| 329 | ✗ | const int in = i * n; | |
| 330 | double value; | ||
| 331 | |||
| 332 | ✗ | value = matrix[in + i]; | |
| 333 | ✗ | for (j = 0; j < i; j++) | |
| 334 | ✗ | value -= matrix[j * n + j] * matrix[in + j] * matrix[in + j]; | |
| 335 | |||
| 336 | ✗ | if (value == 0.) { | |
| 337 | ✗ | return -1; | |
| 338 | } | ||
| 339 | |||
| 340 | ✗ | matrix[in + i] = value; | |
| 341 | ✗ | for (j = i + 1; j < n; j++) { | |
| 342 | ✗ | const int jn = j * n; | |
| 343 | double x; | ||
| 344 | |||
| 345 | ✗ | x = matrix[jn + i]; | |
| 346 | ✗ | for (k = 0; k < i; k++) | |
| 347 | ✗ | x -= matrix[k * n + k] * matrix[in + k] * matrix[jn + k]; | |
| 348 | ✗ | matrix[jn + i] = x / matrix[in + i]; | |
| 349 | } | ||
| 350 | } | ||
| 351 | |||
| 352 | ✗ | return 0; | |
| 353 | } | ||
| 354 | |||
| 355 | ✗ | static int do_interpolation(DeclickChannel *c, double *matrix, | |
| 356 | double *vector, int n, double *out) | ||
| 357 | { | ||
| 358 | int i, j, ret; | ||
| 359 | double *y; | ||
| 360 | |||
| 361 | ✗ | ret = factorization(matrix, n); | |
| 362 | ✗ | if (ret < 0) | |
| 363 | ✗ | return ret; | |
| 364 | |||
| 365 | ✗ | av_fast_malloc(&c->y, &c->y_size, n * sizeof(*c->y)); | |
| 366 | ✗ | y = c->y; | |
| 367 | ✗ | if (!y) | |
| 368 | ✗ | return AVERROR(ENOMEM); | |
| 369 | |||
| 370 | ✗ | for (i = 0; i < n; i++) { | |
| 371 | ✗ | const int in = i * n; | |
| 372 | double value; | ||
| 373 | |||
| 374 | ✗ | value = vector[i]; | |
| 375 | ✗ | for (j = 0; j < i; j++) | |
| 376 | ✗ | value -= matrix[in + j] * y[j]; | |
| 377 | ✗ | y[i] = value; | |
| 378 | } | ||
| 379 | |||
| 380 | ✗ | for (i = n - 1; i >= 0; i--) { | |
| 381 | ✗ | out[i] = y[i] / matrix[i * n + i]; | |
| 382 | ✗ | for (j = i + 1; j < n; j++) | |
| 383 | ✗ | out[i] -= matrix[j * n + i] * out[j]; | |
| 384 | } | ||
| 385 | |||
| 386 | ✗ | return 0; | |
| 387 | } | ||
| 388 | |||
| 389 | ✗ | static int interpolation(DeclickChannel *c, const double *src, int ar_order, | |
| 390 | double *acoefficients, int *index, int nb_errors, | ||
| 391 | double *auxiliary, double *interpolated) | ||
| 392 | { | ||
| 393 | double *vector, *matrix; | ||
| 394 | int i, j; | ||
| 395 | |||
| 396 | ✗ | av_fast_malloc(&c->matrix, &c->matrix_size, nb_errors * nb_errors * sizeof(*c->matrix)); | |
| 397 | ✗ | matrix = c->matrix; | |
| 398 | ✗ | if (!matrix) | |
| 399 | ✗ | return AVERROR(ENOMEM); | |
| 400 | |||
| 401 | ✗ | av_fast_malloc(&c->vector, &c->vector_size, nb_errors * sizeof(*c->vector)); | |
| 402 | ✗ | vector = c->vector; | |
| 403 | ✗ | if (!vector) | |
| 404 | ✗ | return AVERROR(ENOMEM); | |
| 405 | |||
| 406 | ✗ | autocorrelation(acoefficients, ar_order, ar_order + 1, auxiliary, 1.); | |
| 407 | |||
| 408 | ✗ | for (i = 0; i < nb_errors; i++) { | |
| 409 | ✗ | const int im = i * nb_errors; | |
| 410 | |||
| 411 | ✗ | for (j = i; j < nb_errors; j++) { | |
| 412 | ✗ | if (abs(index[j] - index[i]) <= ar_order) { | |
| 413 | ✗ | matrix[j * nb_errors + i] = matrix[im + j] = auxiliary[abs(index[j] - index[i])]; | |
| 414 | } else { | ||
| 415 | ✗ | matrix[j * nb_errors + i] = matrix[im + j] = 0; | |
| 416 | } | ||
| 417 | } | ||
| 418 | } | ||
| 419 | |||
| 420 | ✗ | for (i = 0; i < nb_errors; i++) { | |
| 421 | ✗ | double value = 0.; | |
| 422 | |||
| 423 | ✗ | for (j = -ar_order; j <= ar_order; j++) | |
| 424 | ✗ | if (find_index(index, index[i] - j, nb_errors)) | |
| 425 | ✗ | value -= src[index[i] - j] * auxiliary[abs(j)]; | |
| 426 | |||
| 427 | ✗ | vector[i] = value; | |
| 428 | } | ||
| 429 | |||
| 430 | ✗ | return do_interpolation(c, matrix, vector, nb_errors, interpolated); | |
| 431 | } | ||
| 432 | |||
| 433 | ✗ | static int detect_clips(AudioDeclickContext *s, DeclickChannel *c, | |
| 434 | double unused0, | ||
| 435 | double *unused1, double *unused2, | ||
| 436 | uint8_t *clip, int *index, | ||
| 437 | const double *src, double *dst) | ||
| 438 | { | ||
| 439 | ✗ | const double threshold = s->threshold; | |
| 440 | ✗ | double max_amplitude = 0; | |
| 441 | unsigned *histogram; | ||
| 442 | ✗ | int i, nb_clips = 0; | |
| 443 | |||
| 444 | ✗ | av_fast_malloc(&c->histogram, &c->histogram_size, s->nb_hbins * sizeof(*c->histogram)); | |
| 445 | ✗ | if (!c->histogram) | |
| 446 | ✗ | return AVERROR(ENOMEM); | |
| 447 | ✗ | histogram = c->histogram; | |
| 448 | ✗ | memset(histogram, 0, sizeof(*histogram) * s->nb_hbins); | |
| 449 | |||
| 450 | ✗ | for (i = 0; i < s->window_size; i++) { | |
| 451 | ✗ | const unsigned index = fmin(fabs(src[i]), 1) * (s->nb_hbins - 1); | |
| 452 | |||
| 453 | ✗ | histogram[index]++; | |
| 454 | ✗ | dst[i] = src[i]; | |
| 455 | ✗ | clip[i] = 0; | |
| 456 | } | ||
| 457 | |||
| 458 | ✗ | for (i = s->nb_hbins - 1; i > 1; i--) { | |
| 459 | ✗ | if (histogram[i]) { | |
| 460 | ✗ | if (histogram[i] / (double)FFMAX(histogram[i - 1], 1) > threshold) { | |
| 461 | ✗ | max_amplitude = i / (double)s->nb_hbins; | |
| 462 | } | ||
| 463 | ✗ | break; | |
| 464 | } | ||
| 465 | } | ||
| 466 | |||
| 467 | ✗ | if (max_amplitude > 0.) { | |
| 468 | ✗ | for (i = 0; i < s->window_size; i++) { | |
| 469 | ✗ | clip[i] = fabs(src[i]) >= max_amplitude; | |
| 470 | } | ||
| 471 | } | ||
| 472 | |||
| 473 | ✗ | memset(clip, 0, s->ar_order * sizeof(*clip)); | |
| 474 | ✗ | memset(clip + (s->window_size - s->ar_order), 0, s->ar_order * sizeof(*clip)); | |
| 475 | |||
| 476 | ✗ | for (i = s->ar_order; i < s->window_size - s->ar_order; i++) | |
| 477 | ✗ | if (clip[i]) | |
| 478 | ✗ | index[nb_clips++] = i; | |
| 479 | |||
| 480 | ✗ | return nb_clips; | |
| 481 | } | ||
| 482 | |||
| 483 | ✗ | static int detect_clicks(AudioDeclickContext *s, DeclickChannel *c, | |
| 484 | double sigmae, | ||
| 485 | double *detection, double *acoefficients, | ||
| 486 | uint8_t *click, int *index, | ||
| 487 | const double *src, double *dst) | ||
| 488 | { | ||
| 489 | ✗ | const double threshold = s->threshold; | |
| 490 | ✗ | int i, j, nb_clicks = 0, prev = -1; | |
| 491 | |||
| 492 | ✗ | memset(detection, 0, s->window_size * sizeof(*detection)); | |
| 493 | |||
| 494 | ✗ | for (i = s->ar_order; i < s->window_size; i++) { | |
| 495 | ✗ | for (j = 0; j <= s->ar_order; j++) { | |
| 496 | ✗ | detection[i] += acoefficients[j] * src[i - j]; | |
| 497 | } | ||
| 498 | } | ||
| 499 | |||
| 500 | ✗ | for (i = 0; i < s->window_size; i++) { | |
| 501 | ✗ | click[i] = fabs(detection[i]) > sigmae * threshold; | |
| 502 | ✗ | dst[i] = src[i]; | |
| 503 | } | ||
| 504 | |||
| 505 | ✗ | for (i = 0; i < s->window_size; i++) { | |
| 506 | ✗ | if (!click[i]) | |
| 507 | ✗ | continue; | |
| 508 | |||
| 509 | ✗ | if (prev >= 0 && (i > prev + 1) && (i <= s->nb_burst_samples + prev)) | |
| 510 | ✗ | for (j = prev + 1; j < i; j++) | |
| 511 | ✗ | click[j] = 1; | |
| 512 | ✗ | prev = i; | |
| 513 | } | ||
| 514 | |||
| 515 | ✗ | memset(click, 0, s->ar_order * sizeof(*click)); | |
| 516 | ✗ | memset(click + (s->window_size - s->ar_order), 0, s->ar_order * sizeof(*click)); | |
| 517 | |||
| 518 | ✗ | for (i = s->ar_order; i < s->window_size - s->ar_order; i++) | |
| 519 | ✗ | if (click[i]) | |
| 520 | ✗ | index[nb_clicks++] = i; | |
| 521 | |||
| 522 | ✗ | return nb_clicks; | |
| 523 | } | ||
| 524 | |||
| 525 | typedef struct ThreadData { | ||
| 526 | AVFrame *out; | ||
| 527 | } ThreadData; | ||
| 528 | |||
| 529 | ✗ | static int filter_channel(AVFilterContext *ctx, void *arg, int ch, int nb_jobs) | |
| 530 | { | ||
| 531 | ✗ | AudioDeclickContext *s = ctx->priv; | |
| 532 | ✗ | ThreadData *td = arg; | |
| 533 | ✗ | AVFrame *out = td->out; | |
| 534 | ✗ | const double *src = (const double *)s->in->extended_data[ch]; | |
| 535 | ✗ | double *is = (double *)s->is->extended_data[ch]; | |
| 536 | ✗ | double *dst = (double *)s->out->extended_data[ch]; | |
| 537 | ✗ | double *ptr = (double *)out->extended_data[ch]; | |
| 538 | ✗ | double *buf = (double *)s->buffer->extended_data[ch]; | |
| 539 | ✗ | const double *w = s->window_func_lut; | |
| 540 | ✗ | DeclickChannel *c = &s->chan[ch]; | |
| 541 | double sigmae; | ||
| 542 | int j, ret; | ||
| 543 | |||
| 544 | ✗ | sigmae = autoregression(src, s->ar_order, s->window_size, c->acoefficients, c->acorrelation, c->tmp); | |
| 545 | |||
| 546 | ✗ | if (isfinite_array(c->acoefficients, s->ar_order + 1)) { | |
| 547 | ✗ | double *interpolated = c->interpolated; | |
| 548 | ✗ | int *index = c->index; | |
| 549 | int nb_errors; | ||
| 550 | |||
| 551 | ✗ | nb_errors = s->detector(s, c, sigmae, c->detection, c->acoefficients, | |
| 552 | c->click, index, src, dst); | ||
| 553 | ✗ | if (nb_errors > 0) { | |
| 554 | ✗ | double *enabled = (double *)s->enabled->extended_data[0]; | |
| 555 | |||
| 556 | ✗ | ret = interpolation(c, src, s->ar_order, c->acoefficients, index, | |
| 557 | nb_errors, c->auxiliary, interpolated); | ||
| 558 | ✗ | if (ret < 0) | |
| 559 | ✗ | return ret; | |
| 560 | |||
| 561 | ✗ | av_audio_fifo_peek(s->efifo, (void**)s->enabled->extended_data, s->window_size); | |
| 562 | |||
| 563 | ✗ | for (j = 0; j < nb_errors; j++) { | |
| 564 | ✗ | if (enabled[index[j]]) { | |
| 565 | ✗ | dst[index[j]] = interpolated[j]; | |
| 566 | ✗ | is[index[j]] = 1; | |
| 567 | } | ||
| 568 | } | ||
| 569 | } | ||
| 570 | } else { | ||
| 571 | ✗ | memcpy(dst, src, s->window_size * sizeof(*dst)); | |
| 572 | } | ||
| 573 | |||
| 574 | ✗ | if (s->method == 0) { | |
| 575 | ✗ | for (j = 0; j < s->window_size; j++) | |
| 576 | ✗ | buf[j] += dst[j] * w[j]; | |
| 577 | } else { | ||
| 578 | ✗ | const int skip = s->overlap_skip; | |
| 579 | |||
| 580 | ✗ | for (j = 0; j < s->hop_size; j++) | |
| 581 | ✗ | buf[j] = dst[skip + j]; | |
| 582 | } | ||
| 583 | ✗ | for (j = 0; j < s->hop_size; j++) | |
| 584 | ✗ | ptr[j] = buf[j]; | |
| 585 | |||
| 586 | ✗ | memmove(buf, buf + s->hop_size, (s->window_size * 2 - s->hop_size) * sizeof(*buf)); | |
| 587 | ✗ | memmove(is, is + s->hop_size, (s->window_size - s->hop_size) * sizeof(*is)); | |
| 588 | ✗ | memset(buf + s->window_size * 2 - s->hop_size, 0, s->hop_size * sizeof(*buf)); | |
| 589 | ✗ | memset(is + s->window_size - s->hop_size, 0, s->hop_size * sizeof(*is)); | |
| 590 | |||
| 591 | ✗ | return 0; | |
| 592 | } | ||
| 593 | |||
| 594 | ✗ | static int filter_frame(AVFilterLink *inlink) | |
| 595 | { | ||
| 596 | ✗ | AVFilterContext *ctx = inlink->dst; | |
| 597 | ✗ | AVFilterLink *outlink = ctx->outputs[0]; | |
| 598 | ✗ | AudioDeclickContext *s = ctx->priv; | |
| 599 | ✗ | AVFrame *out = NULL; | |
| 600 | ✗ | int ret = 0, j, ch, detected_errors = 0; | |
| 601 | ThreadData td; | ||
| 602 | |||
| 603 | ✗ | out = ff_get_audio_buffer(outlink, s->hop_size); | |
| 604 | ✗ | if (!out) | |
| 605 | ✗ | return AVERROR(ENOMEM); | |
| 606 | |||
| 607 | ✗ | ret = av_audio_fifo_peek(s->fifo, (void **)s->in->extended_data, | |
| 608 | s->window_size); | ||
| 609 | ✗ | if (ret < 0) | |
| 610 | ✗ | goto fail; | |
| 611 | |||
| 612 | ✗ | td.out = out; | |
| 613 | ✗ | ret = ff_filter_execute(ctx, filter_channel, &td, NULL, inlink->ch_layout.nb_channels); | |
| 614 | ✗ | if (ret < 0) | |
| 615 | ✗ | goto fail; | |
| 616 | |||
| 617 | ✗ | for (ch = 0; ch < s->in->ch_layout.nb_channels; ch++) { | |
| 618 | ✗ | double *is = (double *)s->is->extended_data[ch]; | |
| 619 | |||
| 620 | ✗ | for (j = 0; j < s->hop_size; j++) { | |
| 621 | ✗ | if (is[j]) | |
| 622 | ✗ | detected_errors++; | |
| 623 | } | ||
| 624 | } | ||
| 625 | |||
| 626 | ✗ | av_audio_fifo_drain(s->fifo, s->hop_size); | |
| 627 | ✗ | av_audio_fifo_drain(s->efifo, s->hop_size); | |
| 628 | |||
| 629 | ✗ | if (s->samples_left > 0) | |
| 630 | ✗ | out->nb_samples = FFMIN(s->hop_size, s->samples_left); | |
| 631 | |||
| 632 | ✗ | out->pts = s->pts; | |
| 633 | ✗ | s->pts += av_rescale_q(s->hop_size, (AVRational){1, outlink->sample_rate}, outlink->time_base); | |
| 634 | |||
| 635 | ✗ | s->detected_errors += detected_errors; | |
| 636 | ✗ | s->nb_samples += out->nb_samples * inlink->ch_layout.nb_channels; | |
| 637 | |||
| 638 | ✗ | ret = ff_filter_frame(outlink, out); | |
| 639 | ✗ | if (ret < 0) | |
| 640 | ✗ | return ret; | |
| 641 | |||
| 642 | ✗ | if (s->samples_left > 0) { | |
| 643 | ✗ | s->samples_left -= s->hop_size; | |
| 644 | ✗ | if (s->samples_left <= 0) | |
| 645 | ✗ | av_audio_fifo_drain(s->fifo, av_audio_fifo_size(s->fifo)); | |
| 646 | } | ||
| 647 | |||
| 648 | ✗ | fail: | |
| 649 | ✗ | if (ret < 0) | |
| 650 | ✗ | av_frame_free(&out); | |
| 651 | ✗ | return ret; | |
| 652 | } | ||
| 653 | |||
| 654 | ✗ | static int activate(AVFilterContext *ctx) | |
| 655 | { | ||
| 656 | ✗ | AVFilterLink *inlink = ctx->inputs[0]; | |
| 657 | ✗ | AVFilterLink *outlink = ctx->outputs[0]; | |
| 658 | ✗ | AudioDeclickContext *s = ctx->priv; | |
| 659 | AVFrame *in; | ||
| 660 | int ret, status; | ||
| 661 | int64_t pts; | ||
| 662 | |||
| 663 | ✗ | FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink); | |
| 664 | |||
| 665 | ✗ | ret = ff_inlink_consume_samples(inlink, s->window_size, s->window_size, &in); | |
| 666 | ✗ | if (ret < 0) | |
| 667 | ✗ | return ret; | |
| 668 | ✗ | if (ret > 0) { | |
| 669 | ✗ | double *e = (double *)s->enabled->extended_data[0]; | |
| 670 | |||
| 671 | ✗ | if (s->pts == AV_NOPTS_VALUE) | |
| 672 | ✗ | s->pts = in->pts; | |
| 673 | |||
| 674 | ✗ | ret = av_audio_fifo_write(s->fifo, (void **)in->extended_data, | |
| 675 | ✗ | in->nb_samples); | |
| 676 | ✗ | for (int i = 0; i < in->nb_samples; i++) | |
| 677 | ✗ | e[i] = !ctx->is_disabled; | |
| 678 | |||
| 679 | ✗ | av_audio_fifo_write(s->efifo, (void**)s->enabled->extended_data, in->nb_samples); | |
| 680 | ✗ | av_frame_free(&in); | |
| 681 | ✗ | if (ret < 0) | |
| 682 | ✗ | return ret; | |
| 683 | } | ||
| 684 | |||
| 685 | ✗ | if (av_audio_fifo_size(s->fifo) >= s->window_size || | |
| 686 | ✗ | s->samples_left > 0) | |
| 687 | ✗ | return filter_frame(inlink); | |
| 688 | |||
| 689 | ✗ | if (av_audio_fifo_size(s->fifo) >= s->window_size) { | |
| 690 | ✗ | ff_filter_set_ready(ctx, 100); | |
| 691 | ✗ | return 0; | |
| 692 | } | ||
| 693 | |||
| 694 | ✗ | if (!s->eof && ff_inlink_acknowledge_status(inlink, &status, &pts)) { | |
| 695 | ✗ | if (status == AVERROR_EOF) { | |
| 696 | ✗ | s->eof = 1; | |
| 697 | ✗ | s->samples_left = av_audio_fifo_size(s->fifo) - s->overlap_skip; | |
| 698 | ✗ | ff_filter_set_ready(ctx, 100); | |
| 699 | ✗ | return 0; | |
| 700 | } | ||
| 701 | } | ||
| 702 | |||
| 703 | ✗ | if (s->eof && s->samples_left <= 0) { | |
| 704 | ✗ | ff_outlink_set_status(outlink, AVERROR_EOF, s->pts); | |
| 705 | ✗ | return 0; | |
| 706 | } | ||
| 707 | |||
| 708 | ✗ | if (!s->eof) | |
| 709 | ✗ | FF_FILTER_FORWARD_WANTED(outlink, inlink); | |
| 710 | |||
| 711 | ✗ | return FFERROR_NOT_READY; | |
| 712 | } | ||
| 713 | |||
| 714 | ✗ | static av_cold int init(AVFilterContext *ctx) | |
| 715 | { | ||
| 716 | ✗ | AudioDeclickContext *s = ctx->priv; | |
| 717 | |||
| 718 | ✗ | s->is_declip = !strcmp(ctx->filter->name, "adeclip"); | |
| 719 | ✗ | if (s->is_declip) { | |
| 720 | ✗ | s->detector = detect_clips; | |
| 721 | } else { | ||
| 722 | ✗ | s->detector = detect_clicks; | |
| 723 | } | ||
| 724 | |||
| 725 | ✗ | return 0; | |
| 726 | } | ||
| 727 | |||
| 728 | ✗ | static av_cold void uninit(AVFilterContext *ctx) | |
| 729 | { | ||
| 730 | ✗ | AudioDeclickContext *s = ctx->priv; | |
| 731 | int i; | ||
| 732 | |||
| 733 | ✗ | if (s->nb_samples > 0) | |
| 734 | ✗ | av_log(ctx, AV_LOG_INFO, "Detected %s in %"PRId64" of %"PRId64" samples (%g%%).\n", | |
| 735 | ✗ | s->is_declip ? "clips" : "clicks", s->detected_errors, | |
| 736 | ✗ | s->nb_samples, 100. * s->detected_errors / s->nb_samples); | |
| 737 | |||
| 738 | ✗ | av_audio_fifo_free(s->fifo); | |
| 739 | ✗ | av_audio_fifo_free(s->efifo); | |
| 740 | ✗ | av_freep(&s->window_func_lut); | |
| 741 | ✗ | av_frame_free(&s->enabled); | |
| 742 | ✗ | av_frame_free(&s->in); | |
| 743 | ✗ | av_frame_free(&s->out); | |
| 744 | ✗ | av_frame_free(&s->buffer); | |
| 745 | ✗ | av_frame_free(&s->is); | |
| 746 | |||
| 747 | ✗ | if (s->chan) { | |
| 748 | ✗ | for (i = 0; i < s->nb_channels; i++) { | |
| 749 | ✗ | DeclickChannel *c = &s->chan[i]; | |
| 750 | |||
| 751 | ✗ | av_freep(&c->detection); | |
| 752 | ✗ | av_freep(&c->auxiliary); | |
| 753 | ✗ | av_freep(&c->acoefficients); | |
| 754 | ✗ | av_freep(&c->acorrelation); | |
| 755 | ✗ | av_freep(&c->tmp); | |
| 756 | ✗ | av_freep(&c->click); | |
| 757 | ✗ | av_freep(&c->index); | |
| 758 | ✗ | av_freep(&c->interpolated); | |
| 759 | ✗ | av_freep(&c->matrix); | |
| 760 | ✗ | c->matrix_size = 0; | |
| 761 | ✗ | av_freep(&c->histogram); | |
| 762 | ✗ | c->histogram_size = 0; | |
| 763 | ✗ | av_freep(&c->vector); | |
| 764 | ✗ | c->vector_size = 0; | |
| 765 | ✗ | av_freep(&c->y); | |
| 766 | ✗ | c->y_size = 0; | |
| 767 | } | ||
| 768 | } | ||
| 769 | ✗ | av_freep(&s->chan); | |
| 770 | ✗ | s->nb_channels = 0; | |
| 771 | ✗ | } | |
| 772 | |||
| 773 | static const AVFilterPad inputs[] = { | ||
| 774 | { | ||
| 775 | .name = "default", | ||
| 776 | .type = AVMEDIA_TYPE_AUDIO, | ||
| 777 | .config_props = config_input, | ||
| 778 | }, | ||
| 779 | }; | ||
| 780 | |||
| 781 | const FFFilter ff_af_adeclick = { | ||
| 782 | .p.name = "adeclick", | ||
| 783 | .p.description = NULL_IF_CONFIG_SMALL("Remove impulsive noise from input audio."), | ||
| 784 | .p.priv_class = &adeclick_class, | ||
| 785 | .p.flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL, | ||
| 786 | .priv_size = sizeof(AudioDeclickContext), | ||
| 787 | .init = init, | ||
| 788 | .activate = activate, | ||
| 789 | .uninit = uninit, | ||
| 790 | FILTER_INPUTS(inputs), | ||
| 791 | FILTER_OUTPUTS(ff_audio_default_filterpad), | ||
| 792 | FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_DBLP), | ||
| 793 | }; | ||
| 794 | |||
| 795 | static const AVOption adeclip_options[] = { | ||
| 796 | { "window", "set window size", OFFSET(w), AV_OPT_TYPE_DOUBLE, {.dbl=55}, 10, 100, AF }, | ||
| 797 | { "w", "set window size", OFFSET(w), AV_OPT_TYPE_DOUBLE, {.dbl=55}, 10, 100, AF }, | ||
| 798 | { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_DOUBLE, {.dbl=75}, 50, 95, AF }, | ||
| 799 | { "o", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_DOUBLE, {.dbl=75}, 50, 95, AF }, | ||
| 800 | { "arorder", "set autoregression order", OFFSET(ar), AV_OPT_TYPE_DOUBLE, {.dbl=8}, 0, 25, AF }, | ||
| 801 | { "a", "set autoregression order", OFFSET(ar), AV_OPT_TYPE_DOUBLE, {.dbl=8}, 0, 25, AF }, | ||
| 802 | { "threshold", "set threshold", OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=10}, 1, 100, AF }, | ||
| 803 | { "t", "set threshold", OFFSET(threshold), AV_OPT_TYPE_DOUBLE, {.dbl=10}, 1, 100, AF }, | ||
| 804 | { "hsize", "set histogram size", OFFSET(nb_hbins), AV_OPT_TYPE_INT, {.i64=1000}, 100, 9999, AF }, | ||
| 805 | { "n", "set histogram size", OFFSET(nb_hbins), AV_OPT_TYPE_INT, {.i64=1000}, 100, 9999, AF }, | ||
| 806 | { "method", "set overlap method", OFFSET(method), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, .unit = "m" }, | ||
| 807 | { "m", "set overlap method", OFFSET(method), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, .unit = "m" }, | ||
| 808 | { "add", "overlap-add", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, .unit = "m" }, | ||
| 809 | { "a", "overlap-add", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, .unit = "m" }, | ||
| 810 | { "save", "overlap-save", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, .unit = "m" }, | ||
| 811 | { "s", "overlap-save", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, .unit = "m" }, | ||
| 812 | { NULL } | ||
| 813 | }; | ||
| 814 | |||
| 815 | AVFILTER_DEFINE_CLASS(adeclip); | ||
| 816 | |||
| 817 | const FFFilter ff_af_adeclip = { | ||
| 818 | .p.name = "adeclip", | ||
| 819 | .p.description = NULL_IF_CONFIG_SMALL("Remove clipping from input audio."), | ||
| 820 | .p.priv_class = &adeclip_class, | ||
| 821 | .p.flags = AVFILTER_FLAG_SLICE_THREADS | AVFILTER_FLAG_SUPPORT_TIMELINE_INTERNAL, | ||
| 822 | .priv_size = sizeof(AudioDeclickContext), | ||
| 823 | .init = init, | ||
| 824 | .activate = activate, | ||
| 825 | .uninit = uninit, | ||
| 826 | FILTER_INPUTS(inputs), | ||
| 827 | FILTER_OUTPUTS(ff_audio_default_filterpad), | ||
| 828 | FILTER_SINGLE_SAMPLEFMT(AV_SAMPLE_FMT_DBLP), | ||
| 829 | }; | ||
| 830 |