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/eval.h"
38 #include "libavutil/float_dsp.h"
39 #include "libavutil/mathematics.h"
40 #include "libavutil/opt.h"
41 #include "libavutil/samplefmt.h"
49 #define INPUT_ON 1 /**< input is active */
50 #define INPUT_EOF 2 /**< input has reached EOF (may still be active) */
52 #define DURATION_LONGEST 0
53 #define DURATION_SHORTEST 1
54 #define DURATION_FIRST 2
57 typedef struct FrameInfo {
60 struct FrameInfo *next;
64 * Linked list used to store timestamps and frame sizes of all frames in the
65 * FIFO for the first input.
67 * This is needed to keep timestamps synchronized for the case where multiple
68 * input frames are pushed to the filter for processing before a frame is
69 * requested by the output link.
71 typedef struct FrameList {
78 static void frame_list_clear(FrameList *frame_list)
81 while (frame_list->list) {
82 FrameInfo *info = frame_list->list;
83 frame_list->list = info->next;
86 frame_list->nb_frames = 0;
87 frame_list->nb_samples = 0;
88 frame_list->end = NULL;
92 static int frame_list_next_frame_size(FrameList *frame_list)
94 if (!frame_list->list)
96 return frame_list->list->nb_samples;
99 static int64_t frame_list_next_pts(FrameList *frame_list)
101 if (!frame_list->list)
102 return AV_NOPTS_VALUE;
103 return frame_list->list->pts;
106 static void frame_list_remove_samples(FrameList *frame_list, int nb_samples)
108 if (nb_samples >= frame_list->nb_samples) {
109 frame_list_clear(frame_list);
111 int samples = nb_samples;
112 while (samples > 0) {
113 FrameInfo *info = frame_list->list;
115 if (info->nb_samples <= samples) {
116 samples -= info->nb_samples;
117 frame_list->list = info->next;
118 if (!frame_list->list)
119 frame_list->end = NULL;
120 frame_list->nb_frames--;
121 frame_list->nb_samples -= info->nb_samples;
124 info->nb_samples -= samples;
125 info->pts += samples;
126 frame_list->nb_samples -= samples;
133 static int frame_list_add_frame(FrameList *frame_list, int nb_samples, int64_t pts)
135 FrameInfo *info = av_malloc(sizeof(*info));
137 return AVERROR(ENOMEM);
138 info->nb_samples = nb_samples;
142 if (!frame_list->list) {
143 frame_list->list = info;
144 frame_list->end = info;
146 av_assert0(frame_list->end);
147 frame_list->end->next = info;
148 frame_list->end = info;
150 frame_list->nb_frames++;
151 frame_list->nb_samples += nb_samples;
156 /* FIXME: use directly links fifo */
158 typedef struct MixContext {
159 const AVClass *class; /**< class for AVOptions */
160 AVFloatDSPContext *fdsp;
162 int nb_inputs; /**< number of inputs */
163 int active_inputs; /**< number of input currently active */
164 int duration_mode; /**< mode for determining duration */
165 float dropout_transition; /**< transition time when an input drops out */
166 char *weights_str; /**< string for custom weights for every input */
168 int nb_channels; /**< number of channels */
169 int sample_rate; /**< sample rate */
171 AVAudioFifo **fifos; /**< audio fifo for each input */
172 uint8_t *input_state; /**< current state of each input */
173 float *input_scale; /**< mixing scale factor for each input */
174 float *weights; /**< custom weights for every input */
175 float weight_sum; /**< sum of custom weights for every input */
176 float *scale_norm; /**< normalization factor for every input */
177 int64_t next_pts; /**< calculated pts for next output frame */
178 FrameList *frame_list; /**< list of frame info for the first input */
181 #define OFFSET(x) offsetof(MixContext, x)
182 #define A AV_OPT_FLAG_AUDIO_PARAM
183 #define F AV_OPT_FLAG_FILTERING_PARAM
184 #define T AV_OPT_FLAG_RUNTIME_PARAM
185 static const AVOption amix_options[] = {
186 { "inputs", "Number of inputs.",
187 OFFSET(nb_inputs), AV_OPT_TYPE_INT, { .i64 = 2 }, 1, INT16_MAX, A|F },
188 { "duration", "How to determine the end-of-stream.",
189 OFFSET(duration_mode), AV_OPT_TYPE_INT, { .i64 = DURATION_LONGEST }, 0, 2, A|F, "duration" },
190 { "longest", "Duration of longest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_LONGEST }, 0, 0, A|F, "duration" },
191 { "shortest", "Duration of shortest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_SHORTEST }, 0, 0, A|F, "duration" },
192 { "first", "Duration of first input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_FIRST }, 0, 0, A|F, "duration" },
193 { "dropout_transition", "Transition time, in seconds, for volume "
194 "renormalization when an input stream ends.",
195 OFFSET(dropout_transition), AV_OPT_TYPE_FLOAT, { .dbl = 2.0 }, 0, INT_MAX, A|F },
196 { "weights", "Set weight for each input.",
197 OFFSET(weights_str), AV_OPT_TYPE_STRING, {.str="1 1"}, 0, 0, A|F|T },
201 AVFILTER_DEFINE_CLASS(amix);
204 * Update the scaling factors to apply to each input during mixing.
206 * This balances the full volume range between active inputs and handles
207 * volume transitions when EOF is encountered on an input but mixing continues
208 * with the remaining inputs.
210 static void calculate_scales(MixContext *s, int nb_samples)
212 float weight_sum = 0.f;
215 for (i = 0; i < s->nb_inputs; i++)
216 if (s->input_state[i] & INPUT_ON)
217 weight_sum += FFABS(s->weights[i]);
219 for (i = 0; i < s->nb_inputs; i++) {
220 if (s->input_state[i] & INPUT_ON) {
221 if (s->scale_norm[i] > weight_sum / FFABS(s->weights[i])) {
222 s->scale_norm[i] -= ((s->weight_sum / FFABS(s->weights[i])) / s->nb_inputs) *
223 nb_samples / (s->dropout_transition * s->sample_rate);
224 s->scale_norm[i] = FFMAX(s->scale_norm[i], weight_sum / FFABS(s->weights[i]));
229 for (i = 0; i < s->nb_inputs; i++) {
230 if (s->input_state[i] & INPUT_ON)
231 s->input_scale[i] = 1.0f / s->scale_norm[i] * FFSIGN(s->weights[i]);
233 s->input_scale[i] = 0.0f;
237 static int config_output(AVFilterLink *outlink)
239 AVFilterContext *ctx = outlink->src;
240 MixContext *s = ctx->priv;
244 s->planar = av_sample_fmt_is_planar(outlink->format);
245 s->sample_rate = outlink->sample_rate;
246 outlink->time_base = (AVRational){ 1, outlink->sample_rate };
247 s->next_pts = AV_NOPTS_VALUE;
249 s->frame_list = av_mallocz(sizeof(*s->frame_list));
251 return AVERROR(ENOMEM);
253 s->fifos = av_mallocz_array(s->nb_inputs, sizeof(*s->fifos));
255 return AVERROR(ENOMEM);
257 s->nb_channels = outlink->channels;
258 for (i = 0; i < s->nb_inputs; i++) {
259 s->fifos[i] = av_audio_fifo_alloc(outlink->format, s->nb_channels, 1024);
261 return AVERROR(ENOMEM);
264 s->input_state = av_malloc(s->nb_inputs);
266 return AVERROR(ENOMEM);
267 memset(s->input_state, INPUT_ON, s->nb_inputs);
268 s->active_inputs = s->nb_inputs;
270 s->input_scale = av_mallocz_array(s->nb_inputs, sizeof(*s->input_scale));
271 s->scale_norm = av_mallocz_array(s->nb_inputs, sizeof(*s->scale_norm));
272 if (!s->input_scale || !s->scale_norm)
273 return AVERROR(ENOMEM);
274 for (i = 0; i < s->nb_inputs; i++)
275 s->scale_norm[i] = s->weight_sum / FFABS(s->weights[i]);
276 calculate_scales(s, 0);
278 av_get_channel_layout_string(buf, sizeof(buf), -1, outlink->channel_layout);
280 av_log(ctx, AV_LOG_VERBOSE,
281 "inputs:%d fmt:%s srate:%d cl:%s\n", s->nb_inputs,
282 av_get_sample_fmt_name(outlink->format), outlink->sample_rate, buf);
288 * Read samples from the input FIFOs, mix, and write to the output link.
290 static int output_frame(AVFilterLink *outlink)
292 AVFilterContext *ctx = outlink->src;
293 MixContext *s = ctx->priv;
294 AVFrame *out_buf, *in_buf;
295 int nb_samples, ns, i;
297 if (s->input_state[0] & INPUT_ON) {
298 /* first input live: use the corresponding frame size */
299 nb_samples = frame_list_next_frame_size(s->frame_list);
300 for (i = 1; i < s->nb_inputs; i++) {
301 if (s->input_state[i] & INPUT_ON) {
302 ns = av_audio_fifo_size(s->fifos[i]);
303 if (ns < nb_samples) {
304 if (!(s->input_state[i] & INPUT_EOF))
305 /* unclosed input with not enough samples */
307 /* closed input to drain */
313 s->next_pts = frame_list_next_pts(s->frame_list);
315 /* first input closed: use the available samples */
316 nb_samples = INT_MAX;
317 for (i = 1; i < s->nb_inputs; i++) {
318 if (s->input_state[i] & INPUT_ON) {
319 ns = av_audio_fifo_size(s->fifos[i]);
320 nb_samples = FFMIN(nb_samples, ns);
323 if (nb_samples == INT_MAX) {
324 ff_outlink_set_status(outlink, AVERROR_EOF, s->next_pts);
329 frame_list_remove_samples(s->frame_list, nb_samples);
331 calculate_scales(s, nb_samples);
336 out_buf = ff_get_audio_buffer(outlink, nb_samples);
338 return AVERROR(ENOMEM);
340 in_buf = ff_get_audio_buffer(outlink, nb_samples);
342 av_frame_free(&out_buf);
343 return AVERROR(ENOMEM);
346 for (i = 0; i < s->nb_inputs; i++) {
347 if (s->input_state[i] & INPUT_ON) {
348 int planes, plane_size, p;
350 av_audio_fifo_read(s->fifos[i], (void **)in_buf->extended_data,
353 planes = s->planar ? s->nb_channels : 1;
354 plane_size = nb_samples * (s->planar ? 1 : s->nb_channels);
355 plane_size = FFALIGN(plane_size, 16);
357 if (out_buf->format == AV_SAMPLE_FMT_FLT ||
358 out_buf->format == AV_SAMPLE_FMT_FLTP) {
359 for (p = 0; p < planes; p++) {
360 s->fdsp->vector_fmac_scalar((float *)out_buf->extended_data[p],
361 (float *) in_buf->extended_data[p],
362 s->input_scale[i], plane_size);
365 for (p = 0; p < planes; p++) {
366 s->fdsp->vector_dmac_scalar((double *)out_buf->extended_data[p],
367 (double *) in_buf->extended_data[p],
368 s->input_scale[i], plane_size);
373 av_frame_free(&in_buf);
375 out_buf->pts = s->next_pts;
376 if (s->next_pts != AV_NOPTS_VALUE)
377 s->next_pts += nb_samples;
379 return ff_filter_frame(outlink, out_buf);
383 * Requests a frame, if needed, from each input link other than the first.
385 static int request_samples(AVFilterContext *ctx, int min_samples)
387 MixContext *s = ctx->priv;
390 av_assert0(s->nb_inputs > 1);
392 for (i = 1; i < s->nb_inputs; i++) {
393 if (!(s->input_state[i] & INPUT_ON) ||
394 (s->input_state[i] & INPUT_EOF))
396 if (av_audio_fifo_size(s->fifos[i]) >= min_samples)
398 ff_inlink_request_frame(ctx->inputs[i]);
400 return output_frame(ctx->outputs[0]);
404 * Calculates the number of active inputs and determines EOF based on the
407 * @return 0 if mixing should continue, or AVERROR_EOF if mixing should stop.
409 static int calc_active_inputs(MixContext *s)
412 int active_inputs = 0;
413 for (i = 0; i < s->nb_inputs; i++)
414 active_inputs += !!(s->input_state[i] & INPUT_ON);
415 s->active_inputs = active_inputs;
417 if (!active_inputs ||
418 (s->duration_mode == DURATION_FIRST && !(s->input_state[0] & INPUT_ON)) ||
419 (s->duration_mode == DURATION_SHORTEST && active_inputs != s->nb_inputs))
424 static int activate(AVFilterContext *ctx)
426 AVFilterLink *outlink = ctx->outputs[0];
427 MixContext *s = ctx->priv;
431 FF_FILTER_FORWARD_STATUS_BACK_ALL(outlink, ctx);
433 for (i = 0; i < s->nb_inputs; i++) {
434 AVFilterLink *inlink = ctx->inputs[i];
436 if ((ret = ff_inlink_consume_frame(ctx->inputs[i], &buf)) > 0) {
438 int64_t pts = av_rescale_q(buf->pts, inlink->time_base,
440 ret = frame_list_add_frame(s->frame_list, buf->nb_samples, pts);
447 ret = av_audio_fifo_write(s->fifos[i], (void **)buf->extended_data,
456 ret = output_frame(outlink);
462 for (i = 0; i < s->nb_inputs; i++) {
466 if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
467 if (status == AVERROR_EOF) {
469 s->input_state[i] = 0;
470 if (s->nb_inputs == 1) {
471 ff_outlink_set_status(outlink, status, pts);
475 s->input_state[i] |= INPUT_EOF;
476 if (av_audio_fifo_size(s->fifos[i]) == 0) {
477 s->input_state[i] = 0;
484 if (calc_active_inputs(s)) {
485 ff_outlink_set_status(outlink, AVERROR_EOF, s->next_pts);
489 if (ff_outlink_frame_wanted(outlink)) {
492 if (!(s->input_state[0] & INPUT_ON))
493 return request_samples(ctx, 1);
495 if (s->frame_list->nb_frames == 0) {
496 ff_inlink_request_frame(ctx->inputs[0]);
499 av_assert0(s->frame_list->nb_frames > 0);
501 wanted_samples = frame_list_next_frame_size(s->frame_list);
503 return request_samples(ctx, wanted_samples);
509 static void parse_weights(AVFilterContext *ctx)
511 MixContext *s = ctx->priv;
512 float last_weight = 1.f;
518 for (i = 0; i < s->nb_inputs; i++) {
519 last_weight = av_strtod(p, &p);
520 s->weights[i] = last_weight;
521 s->weight_sum += FFABS(last_weight);
530 for (; i < s->nb_inputs; i++) {
531 s->weights[i] = last_weight;
532 s->weight_sum += FFABS(last_weight);
536 static av_cold int init(AVFilterContext *ctx)
538 MixContext *s = ctx->priv;
541 for (i = 0; i < s->nb_inputs; i++) {
542 AVFilterPad pad = { 0 };
544 pad.type = AVMEDIA_TYPE_AUDIO;
545 pad.name = av_asprintf("input%d", i);
547 return AVERROR(ENOMEM);
549 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
555 s->fdsp = avpriv_float_dsp_alloc(0);
557 return AVERROR(ENOMEM);
559 s->weights = av_mallocz_array(s->nb_inputs, sizeof(*s->weights));
561 return AVERROR(ENOMEM);
568 static av_cold void uninit(AVFilterContext *ctx)
571 MixContext *s = ctx->priv;
574 for (i = 0; i < s->nb_inputs; i++)
575 av_audio_fifo_free(s->fifos[i]);
578 frame_list_clear(s->frame_list);
579 av_freep(&s->frame_list);
580 av_freep(&s->input_state);
581 av_freep(&s->input_scale);
582 av_freep(&s->scale_norm);
583 av_freep(&s->weights);
586 for (i = 0; i < ctx->nb_inputs; i++)
587 av_freep(&ctx->input_pads[i].name);
590 static int query_formats(AVFilterContext *ctx)
592 static const enum AVSampleFormat sample_fmts[] = {
593 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
594 AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
599 if ((ret = ff_set_common_formats(ctx, ff_make_format_list(sample_fmts))) < 0 ||
600 (ret = ff_set_common_samplerates(ctx, ff_all_samplerates())) < 0)
603 return ff_set_common_channel_layouts(ctx, ff_all_channel_counts());
606 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
607 char *res, int res_len, int flags)
609 MixContext *s = ctx->priv;
612 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
617 for (int i = 0; i < s->nb_inputs; i++)
618 s->scale_norm[i] = s->weight_sum / FFABS(s->weights[i]);
619 calculate_scales(s, 0);
624 static const AVFilterPad avfilter_af_amix_outputs[] = {
627 .type = AVMEDIA_TYPE_AUDIO,
628 .config_props = config_output,
633 AVFilter ff_af_amix = {
635 .description = NULL_IF_CONFIG_SMALL("Audio mixing."),
636 .priv_size = sizeof(MixContext),
637 .priv_class = &amix_class,
640 .activate = activate,
641 .query_formats = query_formats,
643 .outputs = avfilter_af_amix_outputs,
644 .process_command = process_command,
645 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS,