3 * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
26 * Mixes audio from multiple sources into a single output. The channel layout,
27 * sample rate, and sample format will be the same for all inputs and the
31 #include "libavutil/attributes.h"
32 #include "libavutil/audio_fifo.h"
33 #include "libavutil/avassert.h"
34 #include "libavutil/avstring.h"
35 #include "libavutil/channel_layout.h"
36 #include "libavutil/common.h"
37 #include "libavutil/float_dsp.h"
38 #include "libavutil/mathematics.h"
39 #include "libavutil/opt.h"
40 #include "libavutil/samplefmt.h"
48 #define INPUT_ON 1 /**< input is active */
49 #define INPUT_EOF 2 /**< input has reached EOF (may still be active) */
51 #define DURATION_LONGEST 0
52 #define DURATION_SHORTEST 1
53 #define DURATION_FIRST 2
56 typedef struct FrameInfo {
59 struct FrameInfo *next;
63 * Linked list used to store timestamps and frame sizes of all frames in the
64 * FIFO for the first input.
66 * This is needed to keep timestamps synchronized for the case where multiple
67 * input frames are pushed to the filter for processing before a frame is
68 * requested by the output link.
70 typedef struct FrameList {
77 static void frame_list_clear(FrameList *frame_list)
80 while (frame_list->list) {
81 FrameInfo *info = frame_list->list;
82 frame_list->list = info->next;
85 frame_list->nb_frames = 0;
86 frame_list->nb_samples = 0;
87 frame_list->end = NULL;
91 static int frame_list_next_frame_size(FrameList *frame_list)
93 if (!frame_list->list)
95 return frame_list->list->nb_samples;
98 static int64_t frame_list_next_pts(FrameList *frame_list)
100 if (!frame_list->list)
101 return AV_NOPTS_VALUE;
102 return frame_list->list->pts;
105 static void frame_list_remove_samples(FrameList *frame_list, int nb_samples)
107 if (nb_samples >= frame_list->nb_samples) {
108 frame_list_clear(frame_list);
110 int samples = nb_samples;
111 while (samples > 0) {
112 FrameInfo *info = frame_list->list;
114 if (info->nb_samples <= samples) {
115 samples -= info->nb_samples;
116 frame_list->list = info->next;
117 if (!frame_list->list)
118 frame_list->end = NULL;
119 frame_list->nb_frames--;
120 frame_list->nb_samples -= info->nb_samples;
123 info->nb_samples -= samples;
124 info->pts += samples;
125 frame_list->nb_samples -= samples;
132 static int frame_list_add_frame(FrameList *frame_list, int nb_samples, int64_t pts)
134 FrameInfo *info = av_malloc(sizeof(*info));
136 return AVERROR(ENOMEM);
137 info->nb_samples = nb_samples;
141 if (!frame_list->list) {
142 frame_list->list = info;
143 frame_list->end = info;
145 av_assert0(frame_list->end);
146 frame_list->end->next = info;
147 frame_list->end = info;
149 frame_list->nb_frames++;
150 frame_list->nb_samples += nb_samples;
155 /* FIXME: use directly links fifo */
157 typedef struct MixContext {
158 const AVClass *class; /**< class for AVOptions */
159 AVFloatDSPContext *fdsp;
161 int nb_inputs; /**< number of inputs */
162 int active_inputs; /**< number of input currently active */
163 int duration_mode; /**< mode for determining duration */
164 float dropout_transition; /**< transition time when an input drops out */
165 char *weights_str; /**< string for custom weights for every input */
167 int nb_channels; /**< number of channels */
168 int sample_rate; /**< sample rate */
170 AVAudioFifo **fifos; /**< audio fifo for each input */
171 uint8_t *input_state; /**< current state of each input */
172 float *input_scale; /**< mixing scale factor for each input */
173 float *weights; /**< custom weights for every input */
174 float weight_sum; /**< sum of custom weights for every input */
175 float *scale_norm; /**< normalization factor for every input */
176 int64_t next_pts; /**< calculated pts for next output frame */
177 FrameList *frame_list; /**< list of frame info for the first input */
180 #define OFFSET(x) offsetof(MixContext, x)
181 #define A AV_OPT_FLAG_AUDIO_PARAM
182 #define F AV_OPT_FLAG_FILTERING_PARAM
183 static const AVOption amix_options[] = {
184 { "inputs", "Number of inputs.",
185 OFFSET(nb_inputs), AV_OPT_TYPE_INT, { .i64 = 2 }, 1, 1024, A|F },
186 { "duration", "How to determine the end-of-stream.",
187 OFFSET(duration_mode), AV_OPT_TYPE_INT, { .i64 = DURATION_LONGEST }, 0, 2, A|F, "duration" },
188 { "longest", "Duration of longest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_LONGEST }, 0, 0, A|F, "duration" },
189 { "shortest", "Duration of shortest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_SHORTEST }, 0, 0, A|F, "duration" },
190 { "first", "Duration of first input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_FIRST }, 0, 0, A|F, "duration" },
191 { "dropout_transition", "Transition time, in seconds, for volume "
192 "renormalization when an input stream ends.",
193 OFFSET(dropout_transition), AV_OPT_TYPE_FLOAT, { .dbl = 2.0 }, 0, INT_MAX, A|F },
194 { "weights", "Set weight for each input.",
195 OFFSET(weights_str), AV_OPT_TYPE_STRING, {.str="1 1"}, 0, 0, A|F },
199 AVFILTER_DEFINE_CLASS(amix);
202 * Update the scaling factors to apply to each input during mixing.
204 * This balances the full volume range between active inputs and handles
205 * volume transitions when EOF is encountered on an input but mixing continues
206 * with the remaining inputs.
208 static void calculate_scales(MixContext *s, int nb_samples)
210 float weight_sum = 0.f;
213 for (i = 0; i < s->nb_inputs; i++)
214 if (s->input_state[i] & INPUT_ON)
215 weight_sum += s->weights[i];
217 for (i = 0; i < s->nb_inputs; i++) {
218 if (s->input_state[i] & INPUT_ON) {
219 if (s->scale_norm[i] > weight_sum / s->weights[i]) {
220 s->scale_norm[i] -= ((s->weight_sum / s->weights[i]) / s->nb_inputs) *
221 nb_samples / (s->dropout_transition * s->sample_rate);
222 s->scale_norm[i] = FFMAX(s->scale_norm[i], weight_sum / s->weights[i]);
227 for (i = 0; i < s->nb_inputs; i++) {
228 if (s->input_state[i] & INPUT_ON)
229 s->input_scale[i] = 1.0f / s->scale_norm[i];
231 s->input_scale[i] = 0.0f;
235 static int config_output(AVFilterLink *outlink)
237 AVFilterContext *ctx = outlink->src;
238 MixContext *s = ctx->priv;
242 s->planar = av_sample_fmt_is_planar(outlink->format);
243 s->sample_rate = outlink->sample_rate;
244 outlink->time_base = (AVRational){ 1, outlink->sample_rate };
245 s->next_pts = AV_NOPTS_VALUE;
247 s->frame_list = av_mallocz(sizeof(*s->frame_list));
249 return AVERROR(ENOMEM);
251 s->fifos = av_mallocz_array(s->nb_inputs, sizeof(*s->fifos));
253 return AVERROR(ENOMEM);
255 s->nb_channels = outlink->channels;
256 for (i = 0; i < s->nb_inputs; i++) {
257 s->fifos[i] = av_audio_fifo_alloc(outlink->format, s->nb_channels, 1024);
259 return AVERROR(ENOMEM);
262 s->input_state = av_malloc(s->nb_inputs);
264 return AVERROR(ENOMEM);
265 memset(s->input_state, INPUT_ON, s->nb_inputs);
266 s->active_inputs = s->nb_inputs;
268 s->input_scale = av_mallocz_array(s->nb_inputs, sizeof(*s->input_scale));
269 s->scale_norm = av_mallocz_array(s->nb_inputs, sizeof(*s->scale_norm));
270 if (!s->input_scale || !s->scale_norm)
271 return AVERROR(ENOMEM);
272 for (i = 0; i < s->nb_inputs; i++)
273 s->scale_norm[i] = s->weight_sum / s->weights[i];
274 calculate_scales(s, 0);
276 av_get_channel_layout_string(buf, sizeof(buf), -1, outlink->channel_layout);
278 av_log(ctx, AV_LOG_VERBOSE,
279 "inputs:%d fmt:%s srate:%d cl:%s\n", s->nb_inputs,
280 av_get_sample_fmt_name(outlink->format), outlink->sample_rate, buf);
286 * Read samples from the input FIFOs, mix, and write to the output link.
288 static int output_frame(AVFilterLink *outlink)
290 AVFilterContext *ctx = outlink->src;
291 MixContext *s = ctx->priv;
292 AVFrame *out_buf, *in_buf;
293 int nb_samples, ns, i;
295 if (s->input_state[0] & INPUT_ON) {
296 /* first input live: use the corresponding frame size */
297 nb_samples = frame_list_next_frame_size(s->frame_list);
298 for (i = 1; i < s->nb_inputs; i++) {
299 if (s->input_state[i] & INPUT_ON) {
300 ns = av_audio_fifo_size(s->fifos[i]);
301 if (ns < nb_samples) {
302 if (!(s->input_state[i] & INPUT_EOF))
303 /* unclosed input with not enough samples */
305 /* closed input to drain */
311 /* first input closed: use the available samples */
312 nb_samples = INT_MAX;
313 for (i = 1; i < s->nb_inputs; i++) {
314 if (s->input_state[i] & INPUT_ON) {
315 ns = av_audio_fifo_size(s->fifos[i]);
316 nb_samples = FFMIN(nb_samples, ns);
319 if (nb_samples == INT_MAX) {
320 ff_outlink_set_status(outlink, AVERROR_EOF, s->next_pts);
325 s->next_pts = frame_list_next_pts(s->frame_list);
326 frame_list_remove_samples(s->frame_list, nb_samples);
328 calculate_scales(s, nb_samples);
333 out_buf = ff_get_audio_buffer(outlink, nb_samples);
335 return AVERROR(ENOMEM);
337 in_buf = ff_get_audio_buffer(outlink, nb_samples);
339 av_frame_free(&out_buf);
340 return AVERROR(ENOMEM);
343 for (i = 0; i < s->nb_inputs; i++) {
344 if (s->input_state[i] & INPUT_ON) {
345 int planes, plane_size, p;
347 av_audio_fifo_read(s->fifos[i], (void **)in_buf->extended_data,
350 planes = s->planar ? s->nb_channels : 1;
351 plane_size = nb_samples * (s->planar ? 1 : s->nb_channels);
352 plane_size = FFALIGN(plane_size, 16);
354 if (out_buf->format == AV_SAMPLE_FMT_FLT ||
355 out_buf->format == AV_SAMPLE_FMT_FLTP) {
356 for (p = 0; p < planes; p++) {
357 s->fdsp->vector_fmac_scalar((float *)out_buf->extended_data[p],
358 (float *) in_buf->extended_data[p],
359 s->input_scale[i], plane_size);
362 for (p = 0; p < planes; p++) {
363 s->fdsp->vector_dmac_scalar((double *)out_buf->extended_data[p],
364 (double *) in_buf->extended_data[p],
365 s->input_scale[i], plane_size);
370 av_frame_free(&in_buf);
372 out_buf->pts = s->next_pts;
373 if (s->next_pts != AV_NOPTS_VALUE)
374 s->next_pts += nb_samples;
376 return ff_filter_frame(outlink, out_buf);
380 * Requests a frame, if needed, from each input link other than the first.
382 static int request_samples(AVFilterContext *ctx, int min_samples)
384 MixContext *s = ctx->priv;
387 av_assert0(s->nb_inputs > 1);
389 for (i = 1; i < s->nb_inputs; i++) {
390 if (!(s->input_state[i] & INPUT_ON) ||
391 (s->input_state[i] & INPUT_EOF))
393 if (av_audio_fifo_size(s->fifos[i]) >= min_samples)
395 ff_inlink_request_frame(ctx->inputs[i]);
397 return output_frame(ctx->outputs[0]);
401 * Calculates the number of active inputs and determines EOF based on the
404 * @return 0 if mixing should continue, or AVERROR_EOF if mixing should stop.
406 static int calc_active_inputs(MixContext *s)
409 int active_inputs = 0;
410 for (i = 0; i < s->nb_inputs; i++)
411 active_inputs += !!(s->input_state[i] & INPUT_ON);
412 s->active_inputs = active_inputs;
414 if (!active_inputs ||
415 (s->duration_mode == DURATION_FIRST && !(s->input_state[0] & INPUT_ON)) ||
416 (s->duration_mode == DURATION_SHORTEST && active_inputs != s->nb_inputs))
421 static int activate(AVFilterContext *ctx)
423 AVFilterLink *outlink = ctx->outputs[0];
424 MixContext *s = ctx->priv;
428 FF_FILTER_FORWARD_STATUS_BACK_ALL(outlink, ctx);
430 for (i = 0; i < s->nb_inputs; i++) {
431 AVFilterLink *inlink = ctx->inputs[i];
433 if ((ret = ff_inlink_consume_frame(ctx->inputs[i], &buf)) > 0) {
435 int64_t pts = av_rescale_q(buf->pts, inlink->time_base,
437 ret = frame_list_add_frame(s->frame_list, buf->nb_samples, pts);
444 ret = av_audio_fifo_write(s->fifos[i], (void **)buf->extended_data,
453 ret = output_frame(outlink);
459 for (i = 0; i < s->nb_inputs; i++) {
463 if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
464 if (status == AVERROR_EOF) {
466 s->input_state[i] = 0;
467 if (s->nb_inputs == 1) {
468 ff_outlink_set_status(outlink, status, pts);
472 s->input_state[i] |= INPUT_EOF;
473 if (av_audio_fifo_size(s->fifos[i]) == 0) {
474 s->input_state[i] = 0;
481 if (calc_active_inputs(s)) {
482 ff_outlink_set_status(outlink, AVERROR_EOF, s->next_pts);
486 if (ff_outlink_frame_wanted(outlink)) {
489 if (!(s->input_state[0] & INPUT_ON))
490 return request_samples(ctx, 1);
492 if (s->frame_list->nb_frames == 0) {
493 ff_inlink_request_frame(ctx->inputs[0]);
496 av_assert0(s->frame_list->nb_frames > 0);
498 wanted_samples = frame_list_next_frame_size(s->frame_list);
500 return request_samples(ctx, wanted_samples);
506 static av_cold int init(AVFilterContext *ctx)
508 MixContext *s = ctx->priv;
509 char *p, *arg, *saveptr = NULL;
510 float last_weight = 1.f;
513 for (i = 0; i < s->nb_inputs; i++) {
514 AVFilterPad pad = { 0 };
516 pad.type = AVMEDIA_TYPE_AUDIO;
517 pad.name = av_asprintf("input%d", i);
519 return AVERROR(ENOMEM);
521 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
527 s->fdsp = avpriv_float_dsp_alloc(0);
529 return AVERROR(ENOMEM);
531 s->weights = av_mallocz_array(s->nb_inputs, sizeof(*s->weights));
533 return AVERROR(ENOMEM);
536 for (i = 0; i < s->nb_inputs; i++) {
537 if (!(arg = av_strtok(p, " ", &saveptr)))
541 sscanf(arg, "%f", &last_weight);
542 s->weights[i] = last_weight;
543 s->weight_sum += last_weight;
546 for (; i < s->nb_inputs; i++) {
547 s->weights[i] = last_weight;
548 s->weight_sum += last_weight;
554 static av_cold void uninit(AVFilterContext *ctx)
557 MixContext *s = ctx->priv;
560 for (i = 0; i < s->nb_inputs; i++)
561 av_audio_fifo_free(s->fifos[i]);
564 frame_list_clear(s->frame_list);
565 av_freep(&s->frame_list);
566 av_freep(&s->input_state);
567 av_freep(&s->input_scale);
568 av_freep(&s->scale_norm);
569 av_freep(&s->weights);
572 for (i = 0; i < ctx->nb_inputs; i++)
573 av_freep(&ctx->input_pads[i].name);
576 static int query_formats(AVFilterContext *ctx)
578 AVFilterFormats *formats = NULL;
579 AVFilterChannelLayouts *layouts;
582 layouts = ff_all_channel_counts();
584 ret = AVERROR(ENOMEM);
588 if ((ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLT )) < 0 ||
589 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_FLTP)) < 0 ||
590 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_DBL )) < 0 ||
591 (ret = ff_add_format(&formats, AV_SAMPLE_FMT_DBLP)) < 0 ||
592 (ret = ff_set_common_formats (ctx, formats)) < 0 ||
593 (ret = ff_set_common_channel_layouts(ctx, layouts)) < 0 ||
594 (ret = ff_set_common_samplerates(ctx, ff_all_samplerates())) < 0)
599 av_freep(&layouts->channel_layouts);
604 static const AVFilterPad avfilter_af_amix_outputs[] = {
607 .type = AVMEDIA_TYPE_AUDIO,
608 .config_props = config_output,
613 AVFilter ff_af_amix = {
615 .description = NULL_IF_CONFIG_SMALL("Audio mixing."),
616 .priv_size = sizeof(MixContext),
617 .priv_class = &amix_class,
620 .activate = activate,
621 .query_formats = query_formats,
623 .outputs = avfilter_af_amix_outputs,
624 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS,