3 * Copyright (c) 2012 Justin Ruggles <justin.ruggles@gmail.com>
5 * This file is part of Libav.
7 * Libav 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 * Libav 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 Libav; 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/audioconvert.h"
32 #include "libavutil/audio_fifo.h"
33 #include "libavutil/avassert.h"
34 #include "libavutil/avstring.h"
35 #include "libavutil/common.h"
36 #include "libavutil/float_dsp.h"
37 #include "libavutil/mathematics.h"
38 #include "libavutil/opt.h"
39 #include "libavutil/samplefmt.h"
46 #define INPUT_OFF 0 /**< input has reached EOF */
47 #define INPUT_ON 1 /**< input is active */
48 #define INPUT_INACTIVE 2 /**< input is on, but is currently inactive */
50 #define DURATION_LONGEST 0
51 #define DURATION_SHORTEST 1
52 #define DURATION_FIRST 2
55 typedef struct FrameInfo {
58 struct FrameInfo *next;
62 * Linked list used to store timestamps and frame sizes of all frames in the
63 * FIFO for the first input.
65 * This is needed to keep timestamps synchronized for the case where multiple
66 * input frames are pushed to the filter for processing before a frame is
67 * requested by the output link.
69 typedef struct FrameList {
76 static void frame_list_clear(FrameList *frame_list)
79 while (frame_list->list) {
80 FrameInfo *info = frame_list->list;
81 frame_list->list = info->next;
84 frame_list->nb_frames = 0;
85 frame_list->nb_samples = 0;
86 frame_list->end = NULL;
90 static int frame_list_next_frame_size(FrameList *frame_list)
92 if (!frame_list->list)
94 return frame_list->list->nb_samples;
97 static int64_t frame_list_next_pts(FrameList *frame_list)
99 if (!frame_list->list)
100 return AV_NOPTS_VALUE;
101 return frame_list->list->pts;
104 static void frame_list_remove_samples(FrameList *frame_list, int nb_samples)
106 if (nb_samples >= frame_list->nb_samples) {
107 frame_list_clear(frame_list);
109 int samples = nb_samples;
110 while (samples > 0) {
111 FrameInfo *info = frame_list->list;
112 av_assert0(info != NULL);
113 if (info->nb_samples <= samples) {
114 samples -= info->nb_samples;
115 frame_list->list = info->next;
116 if (!frame_list->list)
117 frame_list->end = NULL;
118 frame_list->nb_frames--;
119 frame_list->nb_samples -= info->nb_samples;
122 info->nb_samples -= samples;
123 info->pts += samples;
124 frame_list->nb_samples -= samples;
131 static int frame_list_add_frame(FrameList *frame_list, int nb_samples, int64_t pts)
133 FrameInfo *info = av_malloc(sizeof(*info));
135 return AVERROR(ENOMEM);
136 info->nb_samples = nb_samples;
140 if (!frame_list->list) {
141 frame_list->list = info;
142 frame_list->end = info;
144 av_assert0(frame_list->end != NULL);
145 frame_list->end->next = info;
146 frame_list->end = info;
148 frame_list->nb_frames++;
149 frame_list->nb_samples += nb_samples;
155 typedef struct MixContext {
156 const AVClass *class; /**< class for AVOptions */
157 AVFloatDSPContext fdsp;
159 int nb_inputs; /**< number of inputs */
160 int active_inputs; /**< number of input currently active */
161 int duration_mode; /**< mode for determining duration */
162 float dropout_transition; /**< transition time when an input drops out */
164 int nb_channels; /**< number of channels */
165 int sample_rate; /**< sample rate */
167 AVAudioFifo **fifos; /**< audio fifo for each input */
168 uint8_t *input_state; /**< current state of each input */
169 float *input_scale; /**< mixing scale factor for each input */
170 float scale_norm; /**< normalization factor for all inputs */
171 int64_t next_pts; /**< calculated pts for next output frame */
172 FrameList *frame_list; /**< list of frame info for the first input */
175 #define OFFSET(x) offsetof(MixContext, x)
176 #define A AV_OPT_FLAG_AUDIO_PARAM
177 static const AVOption options[] = {
178 { "inputs", "Number of inputs.",
179 OFFSET(nb_inputs), AV_OPT_TYPE_INT, { .i64 = 2 }, 1, 32, A },
180 { "duration", "How to determine the end-of-stream.",
181 OFFSET(duration_mode), AV_OPT_TYPE_INT, { .i64 = DURATION_LONGEST }, 0, 2, A, "duration" },
182 { "longest", "Duration of longest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_LONGEST }, INT_MIN, INT_MAX, A, "duration" },
183 { "shortest", "Duration of shortest input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_SHORTEST }, INT_MIN, INT_MAX, A, "duration" },
184 { "first", "Duration of first input.", 0, AV_OPT_TYPE_CONST, { .i64 = DURATION_FIRST }, INT_MIN, INT_MAX, A, "duration" },
185 { "dropout_transition", "Transition time, in seconds, for volume "
186 "renormalization when an input stream ends.",
187 OFFSET(dropout_transition), AV_OPT_TYPE_FLOAT, { .dbl = 2.0 }, 0, INT_MAX, A },
191 static const AVClass amix_class = {
192 .class_name = "amix filter",
193 .item_name = av_default_item_name,
195 .version = LIBAVUTIL_VERSION_INT,
200 * Update the scaling factors to apply to each input during mixing.
202 * This balances the full volume range between active inputs and handles
203 * volume transitions when EOF is encountered on an input but mixing continues
204 * with the remaining inputs.
206 static void calculate_scales(MixContext *s, int nb_samples)
210 if (s->scale_norm > s->active_inputs) {
211 s->scale_norm -= nb_samples / (s->dropout_transition * s->sample_rate);
212 s->scale_norm = FFMAX(s->scale_norm, s->active_inputs);
215 for (i = 0; i < s->nb_inputs; i++) {
216 if (s->input_state[i] == INPUT_ON)
217 s->input_scale[i] = 1.0f / s->scale_norm;
219 s->input_scale[i] = 0.0f;
223 static int config_output(AVFilterLink *outlink)
225 AVFilterContext *ctx = outlink->src;
226 MixContext *s = ctx->priv;
230 s->planar = av_sample_fmt_is_planar(outlink->format);
231 s->sample_rate = outlink->sample_rate;
232 outlink->time_base = (AVRational){ 1, outlink->sample_rate };
233 s->next_pts = AV_NOPTS_VALUE;
235 s->frame_list = av_mallocz(sizeof(*s->frame_list));
237 return AVERROR(ENOMEM);
239 s->fifos = av_mallocz(s->nb_inputs * sizeof(*s->fifos));
241 return AVERROR(ENOMEM);
243 s->nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
244 for (i = 0; i < s->nb_inputs; i++) {
245 s->fifos[i] = av_audio_fifo_alloc(outlink->format, s->nb_channels, 1024);
247 return AVERROR(ENOMEM);
250 s->input_state = av_malloc(s->nb_inputs);
252 return AVERROR(ENOMEM);
253 memset(s->input_state, INPUT_ON, s->nb_inputs);
254 s->active_inputs = s->nb_inputs;
256 s->input_scale = av_mallocz(s->nb_inputs * sizeof(*s->input_scale));
258 return AVERROR(ENOMEM);
259 s->scale_norm = s->active_inputs;
260 calculate_scales(s, 0);
262 av_get_channel_layout_string(buf, sizeof(buf), -1, outlink->channel_layout);
264 av_log(ctx, AV_LOG_VERBOSE,
265 "inputs:%d fmt:%s srate:%d cl:%s\n", s->nb_inputs,
266 av_get_sample_fmt_name(outlink->format), outlink->sample_rate, buf);
272 * Read samples from the input FIFOs, mix, and write to the output link.
274 static int output_frame(AVFilterLink *outlink, int nb_samples)
276 AVFilterContext *ctx = outlink->src;
277 MixContext *s = ctx->priv;
278 AVFilterBufferRef *out_buf, *in_buf;
281 calculate_scales(s, nb_samples);
283 out_buf = ff_get_audio_buffer(outlink, AV_PERM_WRITE, nb_samples);
285 return AVERROR(ENOMEM);
287 in_buf = ff_get_audio_buffer(outlink, AV_PERM_WRITE, nb_samples);
289 return AVERROR(ENOMEM);
291 for (i = 0; i < s->nb_inputs; i++) {
292 if (s->input_state[i] == INPUT_ON) {
293 int planes, plane_size, p;
295 av_audio_fifo_read(s->fifos[i], (void **)in_buf->extended_data,
298 planes = s->planar ? s->nb_channels : 1;
299 plane_size = nb_samples * (s->planar ? 1 : s->nb_channels);
300 plane_size = FFALIGN(plane_size, 16);
302 for (p = 0; p < planes; p++) {
303 s->fdsp.vector_fmac_scalar((float *)out_buf->extended_data[p],
304 (float *) in_buf->extended_data[p],
305 s->input_scale[i], plane_size);
309 avfilter_unref_buffer(in_buf);
311 out_buf->pts = s->next_pts;
312 if (s->next_pts != AV_NOPTS_VALUE)
313 s->next_pts += nb_samples;
315 return ff_filter_samples(outlink, out_buf);
319 * Returns the smallest number of samples available in the input FIFOs other
320 * than that of the first input.
322 static int get_available_samples(MixContext *s)
325 int available_samples = INT_MAX;
327 av_assert0(s->nb_inputs > 1);
329 for (i = 1; i < s->nb_inputs; i++) {
331 if (s->input_state[i] == INPUT_OFF)
333 nb_samples = av_audio_fifo_size(s->fifos[i]);
334 available_samples = FFMIN(available_samples, nb_samples);
336 if (available_samples == INT_MAX)
338 return available_samples;
342 * Requests a frame, if needed, from each input link other than the first.
344 static int request_samples(AVFilterContext *ctx, int min_samples)
346 MixContext *s = ctx->priv;
349 av_assert0(s->nb_inputs > 1);
351 for (i = 1; i < s->nb_inputs; i++) {
353 if (s->input_state[i] == INPUT_OFF)
355 while (!ret && av_audio_fifo_size(s->fifos[i]) < min_samples)
356 ret = ff_request_frame(ctx->inputs[i]);
357 if (ret == AVERROR_EOF) {
358 if (av_audio_fifo_size(s->fifos[i]) == 0) {
359 s->input_state[i] = INPUT_OFF;
369 * Calculates the number of active inputs and determines EOF based on the
372 * @return 0 if mixing should continue, or AVERROR_EOF if mixing should stop.
374 static int calc_active_inputs(MixContext *s)
377 int active_inputs = 0;
378 for (i = 0; i < s->nb_inputs; i++)
379 active_inputs += !!(s->input_state[i] != INPUT_OFF);
380 s->active_inputs = active_inputs;
382 if (!active_inputs ||
383 (s->duration_mode == DURATION_FIRST && s->input_state[0] == INPUT_OFF) ||
384 (s->duration_mode == DURATION_SHORTEST && active_inputs != s->nb_inputs))
389 static int request_frame(AVFilterLink *outlink)
391 AVFilterContext *ctx = outlink->src;
392 MixContext *s = ctx->priv;
394 int wanted_samples, available_samples;
396 ret = calc_active_inputs(s);
400 if (s->input_state[0] == INPUT_OFF) {
401 ret = request_samples(ctx, 1);
405 ret = calc_active_inputs(s);
409 available_samples = get_available_samples(s);
410 if (!available_samples)
411 return AVERROR(EAGAIN);
413 return output_frame(outlink, available_samples);
416 if (s->frame_list->nb_frames == 0) {
417 ret = ff_request_frame(ctx->inputs[0]);
418 if (ret == AVERROR_EOF) {
419 s->input_state[0] = INPUT_OFF;
420 if (s->nb_inputs == 1)
423 return AVERROR(EAGAIN);
427 av_assert0(s->frame_list->nb_frames > 0);
429 wanted_samples = frame_list_next_frame_size(s->frame_list);
431 if (s->active_inputs > 1) {
432 ret = request_samples(ctx, wanted_samples);
436 ret = calc_active_inputs(s);
441 if (s->active_inputs > 1) {
442 available_samples = get_available_samples(s);
443 if (!available_samples)
444 return AVERROR(EAGAIN);
445 available_samples = FFMIN(available_samples, wanted_samples);
447 available_samples = wanted_samples;
450 s->next_pts = frame_list_next_pts(s->frame_list);
451 frame_list_remove_samples(s->frame_list, available_samples);
453 return output_frame(outlink, available_samples);
456 static int filter_samples(AVFilterLink *inlink, AVFilterBufferRef *buf)
458 AVFilterContext *ctx = inlink->dst;
459 MixContext *s = ctx->priv;
460 AVFilterLink *outlink = ctx->outputs[0];
463 for (i = 0; i < ctx->nb_inputs; i++)
464 if (ctx->inputs[i] == inlink)
466 if (i >= ctx->nb_inputs) {
467 av_log(ctx, AV_LOG_ERROR, "unknown input link\n");
468 ret = AVERROR(EINVAL);
473 int64_t pts = av_rescale_q(buf->pts, inlink->time_base,
475 ret = frame_list_add_frame(s->frame_list, buf->audio->nb_samples, pts);
480 ret = av_audio_fifo_write(s->fifos[i], (void **)buf->extended_data,
481 buf->audio->nb_samples);
484 avfilter_unref_buffer(buf);
489 static int init(AVFilterContext *ctx, const char *args)
491 MixContext *s = ctx->priv;
494 s->class = &amix_class;
495 av_opt_set_defaults(s);
497 if ((ret = av_set_options_string(s, args, "=", ":")) < 0) {
498 av_log(ctx, AV_LOG_ERROR, "Error parsing options string '%s'.\n", args);
503 for (i = 0; i < s->nb_inputs; i++) {
505 AVFilterPad pad = { 0 };
507 snprintf(name, sizeof(name), "input%d", i);
508 pad.type = AVMEDIA_TYPE_AUDIO;
509 pad.name = av_strdup(name);
510 pad.filter_samples = filter_samples;
512 ff_insert_inpad(ctx, i, &pad);
515 avpriv_float_dsp_init(&s->fdsp, 0);
520 static void uninit(AVFilterContext *ctx)
523 MixContext *s = ctx->priv;
526 for (i = 0; i < s->nb_inputs; i++)
527 av_audio_fifo_free(s->fifos[i]);
530 frame_list_clear(s->frame_list);
531 av_freep(&s->frame_list);
532 av_freep(&s->input_state);
533 av_freep(&s->input_scale);
535 for (i = 0; i < ctx->nb_inputs; i++)
536 av_freep(&ctx->input_pads[i].name);
539 static int query_formats(AVFilterContext *ctx)
541 AVFilterFormats *formats = NULL;
542 ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
543 ff_add_format(&formats, AV_SAMPLE_FMT_FLTP);
544 ff_set_common_formats(ctx, formats);
545 ff_set_common_channel_layouts(ctx, ff_all_channel_layouts());
546 ff_set_common_samplerates(ctx, ff_all_samplerates());
550 AVFilter avfilter_af_amix = {
552 .description = NULL_IF_CONFIG_SMALL("Audio mixing."),
553 .priv_size = sizeof(MixContext),
557 .query_formats = query_formats,
559 .inputs = (const AVFilterPad[]) {{ .name = NULL}},
560 .outputs = (const AVFilterPad[]) {{ .name = "default",
561 .type = AVMEDIA_TYPE_AUDIO,
562 .config_props = config_output,
563 .request_frame = request_frame },