2 * Copyright (c) 2016 Paul B Mahol
4 * This file is part of FFmpeg.
6 * FFmpeg is free software; you can redistribute it and/or modify it
7 * under the terms of the GNU Lesser General Public License as published
8 * by the Free Software Foundation; either version 2.1 of the License,
9 * or (at your option) any later version.
11 * FFmpeg is distributed in the hope that it will be useful,
12 * but WITHOUT ANY WARRANTY; without even the implied warranty of
13 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
14 * Lesser General Public License for more details.
16 * You should have received a copy of the GNU Lesser General Public
17 * License along with FFmpeg; if not, write to the Free Software
18 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "libavutil/audio_fifo.h"
22 #include "libavutil/avstring.h"
23 #include "libavfilter/internal.h"
24 #include "libavutil/common.h"
25 #include "libavutil/opt.h"
26 #include "libavcodec/avfft.h"
27 #include "libavutil/eval.h"
30 #include "window_func.h"
32 typedef struct AFFTFiltContext {
39 FFTContext *fft, *ifft;
40 FFTComplex **fft_data;
41 FFTComplex **fft_temp;
53 float *window_func_lut;
56 static const char *const var_names[] = { "sr", "b", "nb", "ch", "chs", "pts", "re", "im", NULL };
57 enum { VAR_SAMPLE_RATE, VAR_BIN, VAR_NBBINS, VAR_CHANNEL, VAR_CHANNELS, VAR_PTS, VAR_REAL, VAR_IMAG, VAR_VARS_NB };
59 #define OFFSET(x) offsetof(AFFTFiltContext, x)
60 #define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
62 static const AVOption afftfilt_options[] = {
63 { "real", "set channels real expressions", OFFSET(real_str), AV_OPT_TYPE_STRING, {.str = "re" }, 0, 0, A },
64 { "imag", "set channels imaginary expressions", OFFSET(img_str), AV_OPT_TYPE_STRING, {.str = "im" }, 0, 0, A },
65 { "win_size", "set window size", OFFSET(fft_size), AV_OPT_TYPE_INT, {.i64=4096}, 16, 131072, A },
66 { "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64 = WFUNC_HANNING}, 0, NB_WFUNC-1, A, "win_func" },
67 { "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, A, "win_func" },
68 { "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, A, "win_func" },
69 { "hann", "Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
70 { "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
71 { "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, A, "win_func" },
72 { "blackman", "Blackman", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BLACKMAN}, 0, 0, A, "win_func" },
73 { "welch", "Welch", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_WELCH}, 0, 0, A, "win_func" },
74 { "flattop", "Flat-top", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_FLATTOP}, 0, 0, A, "win_func" },
75 { "bharris", "Blackman-Harris", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BHARRIS}, 0, 0, A, "win_func" },
76 { "bnuttall", "Blackman-Nuttall", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BNUTTALL}, 0, 0, A, "win_func" },
77 { "bhann", "Bartlett-Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BHANN}, 0, 0, A, "win_func" },
78 { "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, A, "win_func" },
79 { "nuttall", "Nuttall", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_NUTTALL}, 0, 0, A, "win_func" },
80 { "lanczos", "Lanczos", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_LANCZOS}, 0, 0, A, "win_func" },
81 { "gauss", "Gauss", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_GAUSS}, 0, 0, A, "win_func" },
82 { "tukey", "Tukey", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_TUKEY}, 0, 0, A, "win_func" },
83 { "dolph", "Dolph-Chebyshev", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_DOLPH}, 0, 0, A, "win_func" },
84 { "cauchy", "Cauchy", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_CAUCHY}, 0, 0, A, "win_func" },
85 { "parzen", "Parzen", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_PARZEN}, 0, 0, A, "win_func" },
86 { "poisson", "Poisson", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_POISSON}, 0, 0, A, "win_func" },
87 { "bohman", "Bohman", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BOHMAN}, 0, 0, A, "win_func" },
88 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=0.75}, 0, 1, A },
92 AVFILTER_DEFINE_CLASS(afftfilt);
94 static inline double getreal(void *priv, double x, double ch)
96 AFFTFiltContext *s = priv;
99 ich = av_clip(ch, 0, s->nb_exprs - 1);
100 ix = av_clip(x, 0, s->window_size / 2);
102 return s->fft_data[ich][ix].re;
105 static inline double getimag(void *priv, double x, double ch)
107 AFFTFiltContext *s = priv;
110 ich = av_clip(ch, 0, s->nb_exprs - 1);
111 ix = av_clip(x, 0, s->window_size / 2);
113 return s->fft_data[ich][ix].im;
116 static double realf(void *priv, double x, double ch) { return getreal(priv, x, ch); }
117 static double imagf(void *priv, double x, double ch) { return getimag(priv, x, ch); }
119 static const char *const func2_names[] = { "real", "imag", NULL };
120 double (*func2[])(void *, double, double) = { realf, imagf, NULL };
122 static int config_input(AVFilterLink *inlink)
124 AVFilterContext *ctx = inlink->dst;
125 AFFTFiltContext *s = ctx->priv;
126 char *saveptr = NULL;
130 const char *last_expr = "1";
132 s->pts = AV_NOPTS_VALUE;
133 s->fft_bits = av_log2(s->fft_size);
134 s->fft = av_fft_init(s->fft_bits, 0);
135 s->ifft = av_fft_init(s->fft_bits, 1);
136 if (!s->fft || !s->ifft)
137 return AVERROR(ENOMEM);
139 s->window_size = 1 << s->fft_bits;
141 s->fft_data = av_calloc(inlink->channels, sizeof(*s->fft_data));
143 return AVERROR(ENOMEM);
145 s->fft_temp = av_calloc(inlink->channels, sizeof(*s->fft_temp));
147 return AVERROR(ENOMEM);
149 for (ch = 0; ch < inlink->channels; ch++) {
150 s->fft_data[ch] = av_calloc(s->window_size, sizeof(**s->fft_data));
151 if (!s->fft_data[ch])
152 return AVERROR(ENOMEM);
155 for (ch = 0; ch < inlink->channels; ch++) {
156 s->fft_temp[ch] = av_calloc(s->window_size, sizeof(**s->fft_temp));
157 if (!s->fft_temp[ch])
158 return AVERROR(ENOMEM);
161 s->real = av_calloc(inlink->channels, sizeof(*s->real));
163 return AVERROR(ENOMEM);
165 s->imag = av_calloc(inlink->channels, sizeof(*s->imag));
167 return AVERROR(ENOMEM);
169 args = av_strdup(s->real_str);
171 return AVERROR(ENOMEM);
173 for (ch = 0; ch < inlink->channels; ch++) {
174 char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
176 ret = av_expr_parse(&s->real[ch], arg ? arg : last_expr, var_names,
177 NULL, NULL, func2_names, func2, 0, ctx);
187 args = av_strdup(s->img_str ? s->img_str : s->real_str);
189 return AVERROR(ENOMEM);
193 for (ch = 0; ch < inlink->channels; ch++) {
194 char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
196 ret = av_expr_parse(&s->imag[ch], arg ? arg : last_expr, var_names,
197 NULL, NULL, func2_names, func2, 0, ctx);
206 s->fifo = av_audio_fifo_alloc(inlink->format, inlink->channels, s->window_size);
208 return AVERROR(ENOMEM);
210 s->window_func_lut = av_realloc_f(s->window_func_lut, s->window_size,
211 sizeof(*s->window_func_lut));
212 if (!s->window_func_lut)
213 return AVERROR(ENOMEM);
214 generate_window_func(s->window_func_lut, s->window_size, s->win_func, &overlap);
216 s->overlap = overlap;
218 s->hop_size = s->window_size * (1 - s->overlap);
219 if (s->hop_size <= 0)
220 return AVERROR(EINVAL);
222 s->buffer = ff_get_audio_buffer(inlink, s->window_size * 2);
224 return AVERROR(ENOMEM);
229 static int filter_frame(AVFilterLink *inlink)
231 AVFilterContext *ctx = inlink->dst;
232 AVFilterLink *outlink = ctx->outputs[0];
233 AFFTFiltContext *s = ctx->priv;
234 const int window_size = s->window_size;
235 const float f = 1. / (s->window_size / 2);
236 double values[VAR_VARS_NB];
237 AVFrame *out, *in = NULL;
241 in = ff_get_audio_buffer(outlink, window_size);
243 return AVERROR(ENOMEM);
246 ret = av_audio_fifo_peek(s->fifo, (void **)in->extended_data, window_size);
250 for (ch = 0; ch < inlink->channels; ch++) {
251 const float *src = (float *)in->extended_data[ch];
252 FFTComplex *fft_data = s->fft_data[ch];
254 for (n = 0; n < in->nb_samples; n++) {
255 fft_data[n].re = src[n] * s->window_func_lut[n];
259 for (; n < window_size; n++) {
265 values[VAR_PTS] = s->pts;
266 values[VAR_SAMPLE_RATE] = inlink->sample_rate;
267 values[VAR_NBBINS] = window_size / 2;
268 values[VAR_CHANNELS] = inlink->channels;
270 for (ch = 0; ch < inlink->channels; ch++) {
271 FFTComplex *fft_data = s->fft_data[ch];
273 av_fft_permute(s->fft, fft_data);
274 av_fft_calc(s->fft, fft_data);
277 for (ch = 0; ch < inlink->channels; ch++) {
278 FFTComplex *fft_data = s->fft_data[ch];
279 FFTComplex *fft_temp = s->fft_temp[ch];
280 float *buf = (float *)s->buffer->extended_data[ch];
282 values[VAR_CHANNEL] = ch;
284 for (n = 0; n <= window_size / 2; n++) {
288 values[VAR_REAL] = fft_data[n].re;
289 values[VAR_IMAG] = fft_data[n].im;
291 fr = av_expr_eval(s->real[ch], values, s);
292 fi = av_expr_eval(s->imag[ch], values, s);
298 for (n = window_size / 2 + 1, x = window_size / 2 - 1; n < window_size; n++, x--) {
299 fft_temp[n].re = fft_temp[x].re;
300 fft_temp[n].im = -fft_temp[x].im;
303 av_fft_permute(s->ifft, fft_temp);
304 av_fft_calc(s->ifft, fft_temp);
306 for (i = 0; i < window_size; i++) {
307 buf[i] += s->fft_temp[ch][i].re * f;
311 out = ff_get_audio_buffer(outlink, s->hop_size);
313 ret = AVERROR(ENOMEM);
318 s->pts += av_rescale_q(s->hop_size, (AVRational){1, outlink->sample_rate}, outlink->time_base);
320 for (ch = 0; ch < inlink->channels; ch++) {
321 float *dst = (float *)out->extended_data[ch];
322 float *buf = (float *)s->buffer->extended_data[ch];
324 for (n = 0; n < s->hop_size; n++)
325 dst[n] = buf[n] * (1.f - s->overlap);
326 memmove(buf, buf + s->hop_size, window_size * 4);
329 ret = ff_filter_frame(outlink, out);
333 av_audio_fifo_drain(s->fifo, s->hop_size);
337 return ret < 0 ? ret : 0;
340 static int activate(AVFilterContext *ctx)
342 AVFilterLink *inlink = ctx->inputs[0];
343 AVFilterLink *outlink = ctx->outputs[0];
344 AFFTFiltContext *s = ctx->priv;
349 FF_FILTER_FORWARD_STATUS_BACK(outlink, inlink);
351 if (!s->eof && av_audio_fifo_size(s->fifo) < s->window_size) {
352 ret = ff_inlink_consume_frame(inlink, &in);
357 ret = av_audio_fifo_write(s->fifo, (void **)in->extended_data,
359 if (ret >= 0 && s->pts == AV_NOPTS_VALUE)
368 if ((av_audio_fifo_size(s->fifo) >= s->window_size) ||
369 (av_audio_fifo_size(s->fifo) > 0 && s->eof)) {
370 ret = filter_frame(inlink);
371 if (av_audio_fifo_size(s->fifo) >= s->window_size)
372 ff_filter_set_ready(ctx, 100);
376 if (!s->eof && ff_inlink_acknowledge_status(inlink, &status, &pts)) {
377 if (status == AVERROR_EOF) {
379 if (av_audio_fifo_size(s->fifo) >= 0) {
380 ff_filter_set_ready(ctx, 100);
386 if (s->eof && av_audio_fifo_size(s->fifo) <= 0) {
387 ff_outlink_set_status(outlink, AVERROR_EOF, s->pts);
392 FF_FILTER_FORWARD_WANTED(outlink, inlink);
394 return FFERROR_NOT_READY;
397 static int query_formats(AVFilterContext *ctx)
399 AVFilterFormats *formats;
400 AVFilterChannelLayouts *layouts;
401 static const enum AVSampleFormat sample_fmts[] = {
407 layouts = ff_all_channel_counts();
409 return AVERROR(ENOMEM);
410 ret = ff_set_common_channel_layouts(ctx, layouts);
414 formats = ff_make_format_list(sample_fmts);
416 return AVERROR(ENOMEM);
417 ret = ff_set_common_formats(ctx, formats);
421 formats = ff_all_samplerates();
423 return AVERROR(ENOMEM);
424 return ff_set_common_samplerates(ctx, formats);
427 static av_cold void uninit(AVFilterContext *ctx)
429 AFFTFiltContext *s = ctx->priv;
435 for (i = 0; i < s->nb_exprs; i++) {
437 av_freep(&s->fft_data[i]);
439 av_freep(&s->fft_temp[i]);
441 av_freep(&s->fft_data);
442 av_freep(&s->fft_temp);
444 for (i = 0; i < s->nb_exprs; i++) {
445 av_expr_free(s->real[i]);
446 av_expr_free(s->imag[i]);
451 av_frame_free(&s->buffer);
452 av_freep(&s->window_func_lut);
454 av_audio_fifo_free(s->fifo);
457 static const AVFilterPad inputs[] = {
460 .type = AVMEDIA_TYPE_AUDIO,
461 .config_props = config_input,
466 static const AVFilterPad outputs[] = {
469 .type = AVMEDIA_TYPE_AUDIO,
474 AVFilter ff_af_afftfilt = {
476 .description = NULL_IF_CONFIG_SMALL("Apply arbitrary expressions to samples in frequency domain."),
477 .priv_size = sizeof(AFFTFiltContext),
478 .priv_class = &afftfilt_class,
481 .activate = activate,
482 .query_formats = query_formats,