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"
29 #include "window_func.h"
31 typedef struct AFFTFiltContext {
37 FFTContext *fft, *ifft;
38 FFTComplex **fft_data;
51 float *window_func_lut;
54 static const char *const var_names[] = { "sr", "b", "nb", "ch", "chs", "pts", NULL };
55 enum { VAR_SAMPLE_RATE, VAR_BIN, VAR_NBBINS, VAR_CHANNEL, VAR_CHANNELS, VAR_PTS, VAR_VARS_NB };
57 #define OFFSET(x) offsetof(AFFTFiltContext, x)
58 #define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
60 static const AVOption afftfilt_options[] = {
61 { "real", "set channels real expressions", OFFSET(real_str), AV_OPT_TYPE_STRING, {.str = "1" }, 0, 0, A },
62 { "imag", "set channels imaginary expressions", OFFSET(img_str), AV_OPT_TYPE_STRING, {.str = NULL }, 0, 0, A },
63 { "win_size", "set window size", OFFSET(fft_bits), AV_OPT_TYPE_INT, {.i64=12}, 4, 17, A, "fft" },
64 { "w16", 0, 0, AV_OPT_TYPE_CONST, {.i64=4}, 0, 0, A, "fft" },
65 { "w32", 0, 0, AV_OPT_TYPE_CONST, {.i64=5}, 0, 0, A, "fft" },
66 { "w64", 0, 0, AV_OPT_TYPE_CONST, {.i64=6}, 0, 0, A, "fft" },
67 { "w128", 0, 0, AV_OPT_TYPE_CONST, {.i64=7}, 0, 0, A, "fft" },
68 { "w256", 0, 0, AV_OPT_TYPE_CONST, {.i64=8}, 0, 0, A, "fft" },
69 { "w512", 0, 0, AV_OPT_TYPE_CONST, {.i64=9}, 0, 0, A, "fft" },
70 { "w1024", 0, 0, AV_OPT_TYPE_CONST, {.i64=10}, 0, 0, A, "fft" },
71 { "w2048", 0, 0, AV_OPT_TYPE_CONST, {.i64=11}, 0, 0, A, "fft" },
72 { "w4096", 0, 0, AV_OPT_TYPE_CONST, {.i64=12}, 0, 0, A, "fft" },
73 { "w8192", 0, 0, AV_OPT_TYPE_CONST, {.i64=13}, 0, 0, A, "fft" },
74 { "w16384", 0, 0, AV_OPT_TYPE_CONST, {.i64=14}, 0, 0, A, "fft" },
75 { "w32768", 0, 0, AV_OPT_TYPE_CONST, {.i64=15}, 0, 0, A, "fft" },
76 { "w65536", 0, 0, AV_OPT_TYPE_CONST, {.i64=16}, 0, 0, A, "fft" },
77 { "w131072",0, 0, AV_OPT_TYPE_CONST, {.i64=17}, 0, 0, A, "fft" },
78 { "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64 = WFUNC_HANNING}, 0, NB_WFUNC-1, A, "win_func" },
79 { "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, A, "win_func" },
80 { "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, A, "win_func" },
81 { "hann", "Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
82 { "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
83 { "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, A, "win_func" },
84 { "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, A, "win_func" },
85 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=0.75}, 0, 1, A },
89 AVFILTER_DEFINE_CLASS(afftfilt);
91 static int config_input(AVFilterLink *inlink)
93 AVFilterContext *ctx = inlink->dst;
94 AFFTFiltContext *s = ctx->priv;
99 const char *last_expr = "1";
101 s->fft = av_fft_init(s->fft_bits, 0);
102 s->ifft = av_fft_init(s->fft_bits, 1);
103 if (!s->fft || !s->ifft)
104 return AVERROR(ENOMEM);
106 s->window_size = 1 << s->fft_bits;
108 s->fft_data = av_calloc(inlink->channels, sizeof(*s->fft_data));
110 return AVERROR(ENOMEM);
112 for (ch = 0; ch < inlink->channels; ch++) {
113 s->fft_data[ch] = av_calloc(s->window_size, sizeof(**s->fft_data));
114 if (!s->fft_data[ch])
115 return AVERROR(ENOMEM);
118 s->real = av_calloc(inlink->channels, sizeof(*s->real));
120 return AVERROR(ENOMEM);
122 s->imag = av_calloc(inlink->channels, sizeof(*s->imag));
124 return AVERROR(ENOMEM);
126 args = av_strdup(s->real_str);
128 return AVERROR(ENOMEM);
130 for (ch = 0; ch < inlink->channels; ch++) {
131 char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
133 ret = av_expr_parse(&s->real[ch], arg ? arg : last_expr, var_names,
134 NULL, NULL, NULL, NULL, 0, ctx);
144 args = av_strdup(s->img_str ? s->img_str : s->real_str);
146 return AVERROR(ENOMEM);
148 for (ch = 0; ch < inlink->channels; ch++) {
149 char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
151 ret = av_expr_parse(&s->imag[ch], arg ? arg : last_expr, var_names,
152 NULL, NULL, NULL, NULL, 0, ctx);
161 s->fifo = av_audio_fifo_alloc(inlink->format, inlink->channels, s->window_size);
163 return AVERROR(ENOMEM);
165 s->window_func_lut = av_realloc_f(s->window_func_lut, s->window_size,
166 sizeof(*s->window_func_lut));
167 if (!s->window_func_lut)
168 return AVERROR(ENOMEM);
169 generate_window_func(s->window_func_lut, s->window_size, s->win_func, &overlap);
171 s->overlap = overlap;
173 for (s->win_scale = 0, i = 0; i < s->window_size; i++) {
174 s->win_scale += s->window_func_lut[i] * s->window_func_lut[i];
177 s->hop_size = s->window_size * (1 - s->overlap);
178 if (s->hop_size <= 0)
179 return AVERROR(EINVAL);
181 s->buffer = ff_get_audio_buffer(inlink, s->window_size * 2);
183 return AVERROR(ENOMEM);
188 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
190 AVFilterContext *ctx = inlink->dst;
191 AVFilterLink *outlink = ctx->outputs[0];
192 AFFTFiltContext *s = ctx->priv;
193 const int window_size = s->window_size;
194 const float f = 1. / s->win_scale;
195 double values[VAR_VARS_NB];
196 AVFrame *out, *in = NULL;
197 int ch, n, ret, i, j, k;
198 int start = s->start, end = s->end;
200 ret = av_audio_fifo_write(s->fifo, (void **)frame->extended_data, frame->nb_samples);
201 av_frame_free(&frame);
205 while (av_audio_fifo_size(s->fifo) >= window_size) {
207 in = ff_get_audio_buffer(outlink, window_size);
209 return AVERROR(ENOMEM);
212 ret = av_audio_fifo_peek(s->fifo, (void **)in->extended_data, window_size);
216 for (ch = 0; ch < inlink->channels; ch++) {
217 const float *src = (float *)in->extended_data[ch];
218 FFTComplex *fft_data = s->fft_data[ch];
220 for (n = 0; n < in->nb_samples; n++) {
221 fft_data[n].re = src[n] * s->window_func_lut[n];
225 for (; n < window_size; n++) {
231 values[VAR_PTS] = s->pts;
232 values[VAR_SAMPLE_RATE] = inlink->sample_rate;
233 values[VAR_NBBINS] = window_size / 2;
234 values[VAR_CHANNELS] = inlink->channels;
236 for (ch = 0; ch < inlink->channels; ch++) {
237 FFTComplex *fft_data = s->fft_data[ch];
238 float *buf = (float *)s->buffer->extended_data[ch];
241 values[VAR_CHANNEL] = ch;
243 av_fft_permute(s->fft, fft_data);
244 av_fft_calc(s->fft, fft_data);
246 for (n = 0; n < window_size / 2; n++) {
251 fr = av_expr_eval(s->real[ch], values, s);
252 fi = av_expr_eval(s->imag[ch], values, s);
254 fft_data[n].re *= fr;
255 fft_data[n].im *= fi;
258 for (n = window_size / 2 + 1, x = window_size / 2 - 1; n < window_size; n++, x--) {
259 fft_data[n].re = fft_data[x].re;
260 fft_data[n].im = -fft_data[x].im;
263 av_fft_permute(s->ifft, fft_data);
264 av_fft_calc(s->ifft, fft_data);
269 for (i = 0, j = start; j < k && i < window_size; i++, j++) {
270 buf[j] += s->fft_data[ch][i].re * f;
273 for (; i < window_size; i++, j++) {
274 buf[j] = s->fft_data[ch][i].re * f;
277 start += s->hop_size;
284 if (start >= window_size) {
287 start -= window_size;
293 out = ff_get_audio_buffer(outlink, window_size);
295 ret = AVERROR(ENOMEM);
300 s->pts += window_size;
302 for (ch = 0; ch < inlink->channels; ch++) {
303 dst = (float *)out->extended_data[ch];
304 buf = (float *)s->buffer->extended_data[ch];
306 for (n = 0; n < window_size; n++) {
307 dst[n] = buf[n] * (1 - s->overlap);
309 memmove(buf, buf + window_size, window_size * 4);
312 ret = ff_filter_frame(outlink, out);
317 av_audio_fifo_drain(s->fifo, s->hop_size);
321 return ret < 0 ? ret : 0;
324 static int query_formats(AVFilterContext *ctx)
326 AVFilterFormats *formats;
327 AVFilterChannelLayouts *layouts;
328 static const enum AVSampleFormat sample_fmts[] = {
334 layouts = ff_all_channel_counts();
336 return AVERROR(ENOMEM);
337 ret = ff_set_common_channel_layouts(ctx, layouts);
341 formats = ff_make_format_list(sample_fmts);
343 return AVERROR(ENOMEM);
344 ret = ff_set_common_formats(ctx, formats);
348 formats = ff_all_samplerates();
350 return AVERROR(ENOMEM);
351 return ff_set_common_samplerates(ctx, formats);
354 static av_cold void uninit(AVFilterContext *ctx)
356 AFFTFiltContext *s = ctx->priv;
362 for (i = 0; i < s->nb_exprs; i++) {
364 av_freep(&s->fft_data[i]);
366 av_freep(&s->fft_data);
368 for (i = 0; i < s->nb_exprs; i++) {
369 av_expr_free(s->real[i]);
370 av_expr_free(s->imag[i]);
375 av_frame_free(&s->buffer);
376 av_freep(&s->window_func_lut);
378 av_audio_fifo_free(s->fifo);
381 static const AVFilterPad inputs[] = {
384 .type = AVMEDIA_TYPE_AUDIO,
385 .config_props = config_input,
386 .filter_frame = filter_frame,
391 static const AVFilterPad outputs[] = {
394 .type = AVMEDIA_TYPE_AUDIO,
399 AVFilter ff_af_afftfilt = {
401 .description = NULL_IF_CONFIG_SMALL("Apply arbitrary expressions to samples in frequency domain."),
402 .priv_size = sizeof(AFFTFiltContext),
403 .priv_class = &afftfilt_class,
406 .query_formats = query_formats,