2 * Copyright (c) 2019 Paul B Mahol
4 * This file is part of FFmpeg.
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.
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/avassert.h"
22 #include "libavutil/audio_fifo.h"
23 #include "libavutil/channel_layout.h"
24 #include "libavutil/common.h"
25 #include "libavutil/opt.h"
33 typedef struct AudioXCorrelateContext {
47 int (*xcorrelate)(AVFilterContext *ctx, AVFrame *out);
48 } AudioXCorrelateContext;
50 static int query_formats(AVFilterContext *ctx)
52 AVFilterFormats *formats;
53 AVFilterChannelLayouts *layouts;
54 static const enum AVSampleFormat sample_fmts[] = {
60 layouts = ff_all_channel_counts();
62 return AVERROR(ENOMEM);
63 ret = ff_set_common_channel_layouts(ctx, layouts);
67 formats = ff_make_format_list(sample_fmts);
69 return AVERROR(ENOMEM);
70 ret = ff_set_common_formats(ctx, formats);
74 formats = ff_all_samplerates();
76 return AVERROR(ENOMEM);
77 return ff_set_common_samplerates(ctx, formats);
80 static float mean_sum(const float *in, int size)
84 for (int i = 0; i < size; i++)
90 static float square_sum(const float *x, const float *y, int size)
92 float square_sum = 0.f;
94 for (int i = 0; i < size; i++)
95 square_sum += x[i] * y[i];
100 static float xcorrelate(const float *x, const float *y, float sumx, float sumy, int size)
102 const float xm = sumx / size, ym = sumy / size;
103 float num = 0.f, den, den0 = 0.f, den1 = 0.f;
105 for (int i = 0; i < size; i++) {
106 float xd = x[i] - xm;
107 float yd = y[i] - ym;
115 den = sqrtf((den0 * den1) / (size * size));
117 return den <= 1e-6f ? 0.f : num / den;
120 static int xcorrelate_slow(AVFilterContext *ctx, AVFrame *out)
122 AudioXCorrelateContext *s = ctx->priv;
123 const int size = s->size;
126 for (int ch = 0; ch < out->channels; ch++) {
127 const float *x = (const float *)s->cache[0]->extended_data[ch];
128 const float *y = (const float *)s->cache[1]->extended_data[ch];
129 float *sumx = (float *)s->mean_sum[0]->extended_data[ch];
130 float *sumy = (float *)s->mean_sum[1]->extended_data[ch];
131 float *dst = (float *)out->extended_data[ch];
135 sumx[0] = mean_sum(x, size);
136 sumy[0] = mean_sum(y, size);
140 for (int n = 0; n < out->nb_samples; n++) {
141 dst[n] = xcorrelate(x + n, y + n, sumx[0], sumy[0], size);
144 sumx[0] += x[n + size];
146 sumy[0] += y[n + size];
153 static int xcorrelate_fast(AVFilterContext *ctx, AVFrame *out)
155 AudioXCorrelateContext *s = ctx->priv;
156 const int size = s->size;
159 for (int ch = 0; ch < out->channels; ch++) {
160 const float *x = (const float *)s->cache[0]->extended_data[ch];
161 const float *y = (const float *)s->cache[1]->extended_data[ch];
162 float *num_sum = (float *)s->num_sum->extended_data[ch];
163 float *den_sumx = (float *)s->den_sum[0]->extended_data[ch];
164 float *den_sumy = (float *)s->den_sum[1]->extended_data[ch];
165 float *dst = (float *)out->extended_data[ch];
169 num_sum[0] = square_sum(x, y, size);
170 den_sumx[0] = square_sum(x, x, size);
171 den_sumy[0] = square_sum(y, y, size);
175 for (int n = 0; n < out->nb_samples; n++) {
178 num = num_sum[0] / size;
179 den = sqrtf((den_sumx[0] * den_sumy[0]) / (size * size));
181 dst[n] = den <= 1e-6f ? 0.f : num / den;
183 num_sum[0] -= x[n] * y[n];
184 num_sum[0] += x[n + size] * y[n + size];
185 den_sumx[0] -= x[n] * x[n];
186 den_sumx[0] = FFMAX(den_sumx[0], 0.f);
187 den_sumx[0] += x[n + size] * x[n + size];
188 den_sumy[0] -= y[n] * y[n];
189 den_sumy[0] = FFMAX(den_sumy[0], 0.f);
190 den_sumy[0] += y[n + size] * y[n + size];
197 static int activate(AVFilterContext *ctx)
199 AudioXCorrelateContext *s = ctx->priv;
200 AVFrame *frame = NULL;
205 FF_FILTER_FORWARD_STATUS_BACK_ALL(ctx->outputs[0], ctx);
207 for (int i = 0; i < 2; i++) {
208 ret = ff_inlink_consume_frame(ctx->inputs[i], &frame);
210 if (s->pts == AV_NOPTS_VALUE)
212 ret = av_audio_fifo_write(s->fifo[i], (void **)frame->extended_data,
214 av_frame_free(&frame);
220 available = FFMIN(av_audio_fifo_size(s->fifo[0]), av_audio_fifo_size(s->fifo[1]));
221 if (available > s->size) {
222 const int out_samples = available - s->size;
225 if (!s->cache[0] || s->cache[0]->nb_samples < available) {
226 av_frame_free(&s->cache[0]);
227 s->cache[0] = ff_get_audio_buffer(ctx->outputs[0], available);
229 return AVERROR(ENOMEM);
232 if (!s->cache[1] || s->cache[1]->nb_samples < available) {
233 av_frame_free(&s->cache[1]);
234 s->cache[1] = ff_get_audio_buffer(ctx->outputs[0], available);
236 return AVERROR(ENOMEM);
239 ret = av_audio_fifo_peek(s->fifo[0], (void **)s->cache[0]->extended_data, available);
243 ret = av_audio_fifo_peek(s->fifo[1], (void **)s->cache[1]->extended_data, available);
247 out = ff_get_audio_buffer(ctx->outputs[0], out_samples);
249 return AVERROR(ENOMEM);
251 s->used = s->xcorrelate(ctx, out);
254 s->pts += out_samples;
256 av_audio_fifo_drain(s->fifo[0], out_samples);
257 av_audio_fifo_drain(s->fifo[1], out_samples);
259 return ff_filter_frame(ctx->outputs[0], out);
262 if (av_audio_fifo_size(s->fifo[0]) > s->size &&
263 av_audio_fifo_size(s->fifo[1]) > s->size) {
264 ff_filter_set_ready(ctx, 10);
268 for (int i = 0; i < 2; i++) {
269 if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
270 ff_outlink_set_status(ctx->outputs[0], status, pts);
275 if (ff_outlink_frame_wanted(ctx->outputs[0])) {
276 for (int i = 0; i < 2; i++) {
277 if (av_audio_fifo_size(s->fifo[i]) > s->size)
279 ff_inlink_request_frame(ctx->inputs[i]);
284 return FFERROR_NOT_READY;
287 static int config_output(AVFilterLink *outlink)
289 AVFilterContext *ctx = outlink->src;
290 AVFilterLink *inlink = ctx->inputs[0];
291 AudioXCorrelateContext *s = ctx->priv;
293 s->pts = AV_NOPTS_VALUE;
295 outlink->format = inlink->format;
296 outlink->channels = inlink->channels;
297 s->fifo[0] = av_audio_fifo_alloc(outlink->format, outlink->channels, s->size);
298 s->fifo[1] = av_audio_fifo_alloc(outlink->format, outlink->channels, s->size);
299 if (!s->fifo[0] || !s->fifo[1])
300 return AVERROR(ENOMEM);
302 s->mean_sum[0] = ff_get_audio_buffer(outlink, 1);
303 s->mean_sum[1] = ff_get_audio_buffer(outlink, 1);
304 s->num_sum = ff_get_audio_buffer(outlink, 1);
305 s->den_sum[0] = ff_get_audio_buffer(outlink, 1);
306 s->den_sum[1] = ff_get_audio_buffer(outlink, 1);
307 if (!s->mean_sum[0] || !s->mean_sum[1] || !s->num_sum ||
308 !s->den_sum[0] || !s->den_sum[1])
309 return AVERROR(ENOMEM);
312 case 0: s->xcorrelate = xcorrelate_slow; break;
313 case 1: s->xcorrelate = xcorrelate_fast; break;
319 static av_cold void uninit(AVFilterContext *ctx)
321 AudioXCorrelateContext *s = ctx->priv;
323 av_audio_fifo_free(s->fifo[0]);
324 av_audio_fifo_free(s->fifo[1]);
325 av_frame_free(&s->cache[0]);
326 av_frame_free(&s->cache[1]);
327 av_frame_free(&s->mean_sum[0]);
328 av_frame_free(&s->mean_sum[1]);
329 av_frame_free(&s->num_sum);
330 av_frame_free(&s->den_sum[0]);
331 av_frame_free(&s->den_sum[1]);
334 static const AVFilterPad inputs[] = {
336 .name = "axcorrelate0",
337 .type = AVMEDIA_TYPE_AUDIO,
340 .name = "axcorrelate1",
341 .type = AVMEDIA_TYPE_AUDIO,
346 static const AVFilterPad outputs[] = {
349 .type = AVMEDIA_TYPE_AUDIO,
350 .config_props = config_output,
355 #define AF AV_OPT_FLAG_AUDIO_PARAM|AV_OPT_FLAG_FILTERING_PARAM
356 #define OFFSET(x) offsetof(AudioXCorrelateContext, x)
358 static const AVOption axcorrelate_options[] = {
359 { "size", "set segment size", OFFSET(size), AV_OPT_TYPE_INT, {.i64=256}, 2, 131072, AF },
360 { "algo", "set alghorithm", OFFSET(algo), AV_OPT_TYPE_INT, {.i64=0}, 0, 1, AF, "algo" },
361 { "slow", "slow algorithm", 0, AV_OPT_TYPE_CONST, {.i64=0}, 0, 0, AF, "algo" },
362 { "fast", "fast algorithm", 0, AV_OPT_TYPE_CONST, {.i64=1}, 0, 0, AF, "algo" },
366 AVFILTER_DEFINE_CLASS(axcorrelate);
368 const AVFilter ff_af_axcorrelate = {
369 .name = "axcorrelate",
370 .description = NULL_IF_CONFIG_SMALL("Cross-correlate two audio streams."),
371 .priv_size = sizeof(AudioXCorrelateContext),
372 .priv_class = &axcorrelate_class,
373 .query_formats = query_formats,
374 .activate = activate,