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 /* first input closed: use the available samples */
314 nb_samples = INT_MAX;
315 for (i = 1; i < s->nb_inputs; i++) {
316 if (s->input_state[i] & INPUT_ON) {
317 ns = av_audio_fifo_size(s->fifos[i]);
318 nb_samples = FFMIN(nb_samples, ns);
321 if (nb_samples == INT_MAX) {
322 ff_outlink_set_status(outlink, AVERROR_EOF, s->next_pts);
327 s->next_pts = frame_list_next_pts(s->frame_list);
328 frame_list_remove_samples(s->frame_list, nb_samples);
330 calculate_scales(s, nb_samples);
335 out_buf = ff_get_audio_buffer(outlink, nb_samples);
337 return AVERROR(ENOMEM);
339 in_buf = ff_get_audio_buffer(outlink, nb_samples);
341 av_frame_free(&out_buf);
342 return AVERROR(ENOMEM);
345 for (i = 0; i < s->nb_inputs; i++) {
346 if (s->input_state[i] & INPUT_ON) {
347 int planes, plane_size, p;
349 av_audio_fifo_read(s->fifos[i], (void **)in_buf->extended_data,
352 planes = s->planar ? s->nb_channels : 1;
353 plane_size = nb_samples * (s->planar ? 1 : s->nb_channels);
354 plane_size = FFALIGN(plane_size, 16);
356 if (out_buf->format == AV_SAMPLE_FMT_FLT ||
357 out_buf->format == AV_SAMPLE_FMT_FLTP) {
358 for (p = 0; p < planes; p++) {
359 s->fdsp->vector_fmac_scalar((float *)out_buf->extended_data[p],
360 (float *) in_buf->extended_data[p],
361 s->input_scale[i], plane_size);
364 for (p = 0; p < planes; p++) {
365 s->fdsp->vector_dmac_scalar((double *)out_buf->extended_data[p],
366 (double *) in_buf->extended_data[p],
367 s->input_scale[i], plane_size);
372 av_frame_free(&in_buf);
374 out_buf->pts = s->next_pts;
375 if (s->next_pts != AV_NOPTS_VALUE)
376 s->next_pts += nb_samples;
378 return ff_filter_frame(outlink, out_buf);
382 * Requests a frame, if needed, from each input link other than the first.
384 static int request_samples(AVFilterContext *ctx, int min_samples)
386 MixContext *s = ctx->priv;
389 av_assert0(s->nb_inputs > 1);
391 for (i = 1; i < s->nb_inputs; i++) {
392 if (!(s->input_state[i] & INPUT_ON) ||
393 (s->input_state[i] & INPUT_EOF))
395 if (av_audio_fifo_size(s->fifos[i]) >= min_samples)
397 ff_inlink_request_frame(ctx->inputs[i]);
399 return output_frame(ctx->outputs[0]);
403 * Calculates the number of active inputs and determines EOF based on the
406 * @return 0 if mixing should continue, or AVERROR_EOF if mixing should stop.
408 static int calc_active_inputs(MixContext *s)
411 int active_inputs = 0;
412 for (i = 0; i < s->nb_inputs; i++)
413 active_inputs += !!(s->input_state[i] & INPUT_ON);
414 s->active_inputs = active_inputs;
416 if (!active_inputs ||
417 (s->duration_mode == DURATION_FIRST && !(s->input_state[0] & INPUT_ON)) ||
418 (s->duration_mode == DURATION_SHORTEST && active_inputs != s->nb_inputs))
423 static int activate(AVFilterContext *ctx)
425 AVFilterLink *outlink = ctx->outputs[0];
426 MixContext *s = ctx->priv;
430 FF_FILTER_FORWARD_STATUS_BACK_ALL(outlink, ctx);
432 for (i = 0; i < s->nb_inputs; i++) {
433 AVFilterLink *inlink = ctx->inputs[i];
435 if ((ret = ff_inlink_consume_frame(ctx->inputs[i], &buf)) > 0) {
437 int64_t pts = av_rescale_q(buf->pts, inlink->time_base,
439 ret = frame_list_add_frame(s->frame_list, buf->nb_samples, pts);
446 ret = av_audio_fifo_write(s->fifos[i], (void **)buf->extended_data,
455 ret = output_frame(outlink);
461 for (i = 0; i < s->nb_inputs; i++) {
465 if (ff_inlink_acknowledge_status(ctx->inputs[i], &status, &pts)) {
466 if (status == AVERROR_EOF) {
468 s->input_state[i] = 0;
469 if (s->nb_inputs == 1) {
470 ff_outlink_set_status(outlink, status, pts);
474 s->input_state[i] |= INPUT_EOF;
475 if (av_audio_fifo_size(s->fifos[i]) == 0) {
476 s->input_state[i] = 0;
483 if (calc_active_inputs(s)) {
484 ff_outlink_set_status(outlink, AVERROR_EOF, s->next_pts);
488 if (ff_outlink_frame_wanted(outlink)) {
491 if (!(s->input_state[0] & INPUT_ON))
492 return request_samples(ctx, 1);
494 if (s->frame_list->nb_frames == 0) {
495 ff_inlink_request_frame(ctx->inputs[0]);
498 av_assert0(s->frame_list->nb_frames > 0);
500 wanted_samples = frame_list_next_frame_size(s->frame_list);
502 return request_samples(ctx, wanted_samples);
508 static void parse_weights(AVFilterContext *ctx)
510 MixContext *s = ctx->priv;
511 float last_weight = 1.f;
517 for (i = 0; i < s->nb_inputs; i++) {
518 last_weight = av_strtod(p, &p);
519 s->weights[i] = last_weight;
520 s->weight_sum += FFABS(last_weight);
529 for (; i < s->nb_inputs; i++) {
530 s->weights[i] = last_weight;
531 s->weight_sum += FFABS(last_weight);
535 static av_cold int init(AVFilterContext *ctx)
537 MixContext *s = ctx->priv;
540 for (i = 0; i < s->nb_inputs; i++) {
541 AVFilterPad pad = { 0 };
543 pad.type = AVMEDIA_TYPE_AUDIO;
544 pad.name = av_asprintf("input%d", i);
546 return AVERROR(ENOMEM);
548 if ((ret = ff_insert_inpad(ctx, i, &pad)) < 0) {
554 s->fdsp = avpriv_float_dsp_alloc(0);
556 return AVERROR(ENOMEM);
558 s->weights = av_mallocz_array(s->nb_inputs, sizeof(*s->weights));
560 return AVERROR(ENOMEM);
567 static av_cold void uninit(AVFilterContext *ctx)
570 MixContext *s = ctx->priv;
573 for (i = 0; i < s->nb_inputs; i++)
574 av_audio_fifo_free(s->fifos[i]);
577 frame_list_clear(s->frame_list);
578 av_freep(&s->frame_list);
579 av_freep(&s->input_state);
580 av_freep(&s->input_scale);
581 av_freep(&s->scale_norm);
582 av_freep(&s->weights);
585 for (i = 0; i < ctx->nb_inputs; i++)
586 av_freep(&ctx->input_pads[i].name);
589 static int query_formats(AVFilterContext *ctx)
591 static const enum AVSampleFormat sample_fmts[] = {
592 AV_SAMPLE_FMT_FLT, AV_SAMPLE_FMT_FLTP,
593 AV_SAMPLE_FMT_DBL, AV_SAMPLE_FMT_DBLP,
596 AVFilterChannelLayouts *layouts;
599 layouts = ff_all_channel_counts();
601 ret = AVERROR(ENOMEM);
605 if ((ret = ff_set_common_formats(ctx, ff_make_format_list(sample_fmts))) < 0 ||
606 (ret = ff_set_common_channel_layouts(ctx, layouts)) < 0 ||
607 (ret = ff_set_common_samplerates(ctx, ff_all_samplerates())) < 0)
612 av_freep(&layouts->channel_layouts);
617 static int process_command(AVFilterContext *ctx, const char *cmd, const char *args,
618 char *res, int res_len, int flags)
620 MixContext *s = ctx->priv;
623 ret = ff_filter_process_command(ctx, cmd, args, res, res_len, flags);
628 for (int i = 0; i < s->nb_inputs; i++)
629 s->scale_norm[i] = s->weight_sum / FFABS(s->weights[i]);
630 calculate_scales(s, 0);
635 static const AVFilterPad avfilter_af_amix_outputs[] = {
638 .type = AVMEDIA_TYPE_AUDIO,
639 .config_props = config_output,
644 AVFilter ff_af_amix = {
646 .description = NULL_IF_CONFIG_SMALL("Audio mixing."),
647 .priv_size = sizeof(MixContext),
648 .priv_class = &amix_class,
651 .activate = activate,
652 .query_formats = query_formats,
654 .outputs = avfilter_af_amix_outputs,
655 .process_command = process_command,
656 .flags = AVFILTER_FLAG_DYNAMIC_INPUTS,