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;
39 FFTComplex **fft_temp;
52 float *window_func_lut;
55 static const char *const var_names[] = { "sr", "b", "nb", "ch", "chs", "pts", "re", "im", NULL };
56 enum { VAR_SAMPLE_RATE, VAR_BIN, VAR_NBBINS, VAR_CHANNEL, VAR_CHANNELS, VAR_PTS, VAR_REAL, VAR_IMAG, VAR_VARS_NB };
58 #define OFFSET(x) offsetof(AFFTFiltContext, x)
59 #define A AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
61 static const AVOption afftfilt_options[] = {
62 { "real", "set channels real expressions", OFFSET(real_str), AV_OPT_TYPE_STRING, {.str = "re" }, 0, 0, A },
63 { "imag", "set channels imaginary expressions", OFFSET(img_str), AV_OPT_TYPE_STRING, {.str = "im" }, 0, 0, A },
64 { "win_size", "set window size", OFFSET(fft_bits), AV_OPT_TYPE_INT, {.i64=12}, 4, 17, A, "fft" },
65 { "w16", 0, 0, AV_OPT_TYPE_CONST, {.i64=4}, 0, 0, A, "fft" },
66 { "w32", 0, 0, AV_OPT_TYPE_CONST, {.i64=5}, 0, 0, A, "fft" },
67 { "w64", 0, 0, AV_OPT_TYPE_CONST, {.i64=6}, 0, 0, A, "fft" },
68 { "w128", 0, 0, AV_OPT_TYPE_CONST, {.i64=7}, 0, 0, A, "fft" },
69 { "w256", 0, 0, AV_OPT_TYPE_CONST, {.i64=8}, 0, 0, A, "fft" },
70 { "w512", 0, 0, AV_OPT_TYPE_CONST, {.i64=9}, 0, 0, A, "fft" },
71 { "w1024", 0, 0, AV_OPT_TYPE_CONST, {.i64=10}, 0, 0, A, "fft" },
72 { "w2048", 0, 0, AV_OPT_TYPE_CONST, {.i64=11}, 0, 0, A, "fft" },
73 { "w4096", 0, 0, AV_OPT_TYPE_CONST, {.i64=12}, 0, 0, A, "fft" },
74 { "w8192", 0, 0, AV_OPT_TYPE_CONST, {.i64=13}, 0, 0, A, "fft" },
75 { "w16384", 0, 0, AV_OPT_TYPE_CONST, {.i64=14}, 0, 0, A, "fft" },
76 { "w32768", 0, 0, AV_OPT_TYPE_CONST, {.i64=15}, 0, 0, A, "fft" },
77 { "w65536", 0, 0, AV_OPT_TYPE_CONST, {.i64=16}, 0, 0, A, "fft" },
78 { "w131072",0, 0, AV_OPT_TYPE_CONST, {.i64=17}, 0, 0, A, "fft" },
79 { "win_func", "set window function", OFFSET(win_func), AV_OPT_TYPE_INT, {.i64 = WFUNC_HANNING}, 0, NB_WFUNC-1, A, "win_func" },
80 { "rect", "Rectangular", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_RECT}, 0, 0, A, "win_func" },
81 { "bartlett", "Bartlett", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_BARTLETT}, 0, 0, A, "win_func" },
82 { "hann", "Hann", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
83 { "hanning", "Hanning", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HANNING}, 0, 0, A, "win_func" },
84 { "hamming", "Hamming", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_HAMMING}, 0, 0, A, "win_func" },
85 { "sine", "Sine", 0, AV_OPT_TYPE_CONST, {.i64=WFUNC_SINE}, 0, 0, A, "win_func" },
86 { "overlap", "set window overlap", OFFSET(overlap), AV_OPT_TYPE_FLOAT, {.dbl=0.75}, 0, 1, A },
90 AVFILTER_DEFINE_CLASS(afftfilt);
92 static inline double getreal(void *priv, double x, double ch)
94 AFFTFiltContext *s = priv;
97 ich = av_clip(ch, 0, s->nb_exprs - 1);
98 ix = av_clip(x, 0, s->window_size / 2);
100 return s->fft_data[ich][ix].re;
103 static inline double getimag(void *priv, double x, double ch)
105 AFFTFiltContext *s = priv;
108 ich = av_clip(ch, 0, s->nb_exprs - 1);
109 ix = av_clip(x, 0, s->window_size / 2);
111 return s->fft_data[ich][ix].im;
114 static double realf(void *priv, double x, double ch) { return getreal(priv, x, ch); }
115 static double imagf(void *priv, double x, double ch) { return getimag(priv, x, ch); }
117 static const char *const func2_names[] = { "real", "imag", NULL };
118 double (*func2[])(void *, double, double) = { realf, imagf, NULL };
120 static int config_input(AVFilterLink *inlink)
122 AVFilterContext *ctx = inlink->dst;
123 AFFTFiltContext *s = ctx->priv;
124 char *saveptr = NULL;
128 const char *last_expr = "1";
130 s->fft = av_fft_init(s->fft_bits, 0);
131 s->ifft = av_fft_init(s->fft_bits, 1);
132 if (!s->fft || !s->ifft)
133 return AVERROR(ENOMEM);
135 s->window_size = 1 << s->fft_bits;
137 s->fft_data = av_calloc(inlink->channels, sizeof(*s->fft_data));
139 return AVERROR(ENOMEM);
141 s->fft_temp = av_calloc(inlink->channels, sizeof(*s->fft_temp));
143 return AVERROR(ENOMEM);
145 for (ch = 0; ch < inlink->channels; ch++) {
146 s->fft_data[ch] = av_calloc(s->window_size, sizeof(**s->fft_data));
147 if (!s->fft_data[ch])
148 return AVERROR(ENOMEM);
151 for (ch = 0; ch < inlink->channels; ch++) {
152 s->fft_temp[ch] = av_calloc(s->window_size, sizeof(**s->fft_temp));
153 if (!s->fft_temp[ch])
154 return AVERROR(ENOMEM);
157 s->real = av_calloc(inlink->channels, sizeof(*s->real));
159 return AVERROR(ENOMEM);
161 s->imag = av_calloc(inlink->channels, sizeof(*s->imag));
163 return AVERROR(ENOMEM);
165 args = av_strdup(s->real_str);
167 return AVERROR(ENOMEM);
169 for (ch = 0; ch < inlink->channels; ch++) {
170 char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
172 ret = av_expr_parse(&s->real[ch], arg ? arg : last_expr, var_names,
173 NULL, NULL, func2_names, func2, 0, ctx);
183 args = av_strdup(s->img_str ? s->img_str : s->real_str);
185 return AVERROR(ENOMEM);
187 for (ch = 0; ch < inlink->channels; ch++) {
188 char *arg = av_strtok(ch == 0 ? args : NULL, "|", &saveptr);
190 ret = av_expr_parse(&s->imag[ch], arg ? arg : last_expr, var_names,
191 NULL, NULL, func2_names, func2, 0, ctx);
200 s->fifo = av_audio_fifo_alloc(inlink->format, inlink->channels, s->window_size);
202 return AVERROR(ENOMEM);
204 s->window_func_lut = av_realloc_f(s->window_func_lut, s->window_size,
205 sizeof(*s->window_func_lut));
206 if (!s->window_func_lut)
207 return AVERROR(ENOMEM);
208 generate_window_func(s->window_func_lut, s->window_size, s->win_func, &overlap);
210 s->overlap = overlap;
212 for (s->win_scale = 0, i = 0; i < s->window_size; i++) {
213 s->win_scale += s->window_func_lut[i] * s->window_func_lut[i];
216 s->hop_size = s->window_size * (1 - s->overlap);
217 if (s->hop_size <= 0)
218 return AVERROR(EINVAL);
220 s->buffer = ff_get_audio_buffer(inlink, s->window_size * 2);
222 return AVERROR(ENOMEM);
227 static int filter_frame(AVFilterLink *inlink, AVFrame *frame)
229 AVFilterContext *ctx = inlink->dst;
230 AVFilterLink *outlink = ctx->outputs[0];
231 AFFTFiltContext *s = ctx->priv;
232 const int window_size = s->window_size;
233 const float f = 1. / s->win_scale;
234 double values[VAR_VARS_NB];
235 AVFrame *out, *in = NULL;
236 int ch, n, ret, i, j, k;
237 int start = s->start, end = s->end;
239 ret = av_audio_fifo_write(s->fifo, (void **)frame->extended_data, frame->nb_samples);
240 av_frame_free(&frame);
244 while (av_audio_fifo_size(s->fifo) >= window_size) {
246 in = ff_get_audio_buffer(outlink, window_size);
248 return AVERROR(ENOMEM);
251 ret = av_audio_fifo_peek(s->fifo, (void **)in->extended_data, window_size);
255 for (ch = 0; ch < inlink->channels; ch++) {
256 const float *src = (float *)in->extended_data[ch];
257 FFTComplex *fft_data = s->fft_data[ch];
259 for (n = 0; n < in->nb_samples; n++) {
260 fft_data[n].re = src[n] * s->window_func_lut[n];
264 for (; n < window_size; n++) {
270 values[VAR_PTS] = s->pts;
271 values[VAR_SAMPLE_RATE] = inlink->sample_rate;
272 values[VAR_NBBINS] = window_size / 2;
273 values[VAR_CHANNELS] = inlink->channels;
275 for (ch = 0; ch < inlink->channels; ch++) {
276 FFTComplex *fft_data = s->fft_data[ch];
277 FFTComplex *fft_temp = s->fft_temp[ch];
278 float *buf = (float *)s->buffer->extended_data[ch];
281 values[VAR_CHANNEL] = ch;
283 av_fft_permute(s->fft, fft_data);
284 av_fft_calc(s->fft, fft_data);
286 for (n = 0; n <= window_size / 2; n++) {
290 values[VAR_REAL] = fft_data[n].re;
291 values[VAR_IMAG] = fft_data[n].im;
293 fr = av_expr_eval(s->real[ch], values, s);
294 fi = av_expr_eval(s->imag[ch], values, s);
300 for (n = window_size / 2 + 1, x = window_size / 2 - 1; n < window_size; n++, x--) {
301 fft_temp[n].re = fft_temp[x].re;
302 fft_temp[n].im = -fft_temp[x].im;
305 av_fft_permute(s->ifft, fft_temp);
306 av_fft_calc(s->ifft, fft_temp);
311 for (i = 0, j = start; j < k && i < window_size; i++, j++) {
312 buf[j] += s->fft_temp[ch][i].re * f;
315 for (; i < window_size; i++, j++) {
316 buf[j] = s->fft_temp[ch][i].re * f;
319 start += s->hop_size;
326 if (start >= window_size) {
329 start -= window_size;
335 out = ff_get_audio_buffer(outlink, window_size);
337 ret = AVERROR(ENOMEM);
342 s->pts += window_size;
344 for (ch = 0; ch < inlink->channels; ch++) {
345 dst = (float *)out->extended_data[ch];
346 buf = (float *)s->buffer->extended_data[ch];
348 for (n = 0; n < window_size; n++) {
349 dst[n] = buf[n] * (1 - s->overlap);
351 memmove(buf, buf + window_size, window_size * 4);
354 ret = ff_filter_frame(outlink, out);
359 av_audio_fifo_drain(s->fifo, s->hop_size);
363 return ret < 0 ? ret : 0;
366 static int query_formats(AVFilterContext *ctx)
368 AVFilterFormats *formats;
369 AVFilterChannelLayouts *layouts;
370 static const enum AVSampleFormat sample_fmts[] = {
376 layouts = ff_all_channel_counts();
378 return AVERROR(ENOMEM);
379 ret = ff_set_common_channel_layouts(ctx, layouts);
383 formats = ff_make_format_list(sample_fmts);
385 return AVERROR(ENOMEM);
386 ret = ff_set_common_formats(ctx, formats);
390 formats = ff_all_samplerates();
392 return AVERROR(ENOMEM);
393 return ff_set_common_samplerates(ctx, formats);
396 static av_cold void uninit(AVFilterContext *ctx)
398 AFFTFiltContext *s = ctx->priv;
404 for (i = 0; i < s->nb_exprs; i++) {
406 av_freep(&s->fft_data[i]);
408 av_freep(&s->fft_temp[i]);
410 av_freep(&s->fft_data);
411 av_freep(&s->fft_temp);
413 for (i = 0; i < s->nb_exprs; i++) {
414 av_expr_free(s->real[i]);
415 av_expr_free(s->imag[i]);
420 av_frame_free(&s->buffer);
421 av_freep(&s->window_func_lut);
423 av_audio_fifo_free(s->fifo);
426 static const AVFilterPad inputs[] = {
429 .type = AVMEDIA_TYPE_AUDIO,
430 .config_props = config_input,
431 .filter_frame = filter_frame,
436 static const AVFilterPad outputs[] = {
439 .type = AVMEDIA_TYPE_AUDIO,
444 AVFilter ff_af_afftfilt = {
446 .description = NULL_IF_CONFIG_SMALL("Apply arbitrary expressions to samples in frequency domain."),
447 .priv_size = sizeof(AFFTFiltContext),
448 .priv_class = &afftfilt_class,
451 .query_formats = query_formats,