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/float_dsp.h"
36 #include "libavutil/mathematics.h"
37 #include "libavutil/opt.h"
38 #include "libavutil/samplefmt.h"
45 #define INPUT_OFF 0 /**< input has reached EOF */
46 #define INPUT_ON 1 /**< input is active */
47 #define INPUT_INACTIVE 2 /**< input is on, but is currently inactive */
49 #define DURATION_LONGEST 0
50 #define DURATION_SHORTEST 1
51 #define DURATION_FIRST 2
54 typedef struct FrameInfo {
57 struct FrameInfo *next;
61 * Linked list used to store timestamps and frame sizes of all frames in the
62 * FIFO for the first input.
64 * This is needed to keep timestamps synchronized for the case where multiple
65 * input frames are pushed to the filter for processing before a frame is
66 * requested by the output link.
68 typedef struct FrameList {
75 static void frame_list_clear(FrameList *frame_list)
78 while (frame_list->list) {
79 FrameInfo *info = frame_list->list;
80 frame_list->list = info->next;
83 frame_list->nb_frames = 0;
84 frame_list->nb_samples = 0;
85 frame_list->end = NULL;
89 static int frame_list_next_frame_size(FrameList *frame_list)
91 if (!frame_list->list)
93 return frame_list->list->nb_samples;
96 static int64_t frame_list_next_pts(FrameList *frame_list)
98 if (!frame_list->list)
99 return AV_NOPTS_VALUE;
100 return frame_list->list->pts;
103 static void frame_list_remove_samples(FrameList *frame_list, int nb_samples)
105 if (nb_samples >= frame_list->nb_samples) {
106 frame_list_clear(frame_list);
108 int samples = nb_samples;
109 while (samples > 0) {
110 FrameInfo *info = frame_list->list;
111 av_assert0(info != NULL);
112 if (info->nb_samples <= samples) {
113 samples -= info->nb_samples;
114 frame_list->list = info->next;
115 if (!frame_list->list)
116 frame_list->end = NULL;
117 frame_list->nb_frames--;
118 frame_list->nb_samples -= info->nb_samples;
121 info->nb_samples -= samples;
122 info->pts += samples;
123 frame_list->nb_samples -= samples;
130 static int frame_list_add_frame(FrameList *frame_list, int nb_samples, int64_t pts)
132 FrameInfo *info = av_malloc(sizeof(*info));
134 return AVERROR(ENOMEM);
135 info->nb_samples = nb_samples;
139 if (!frame_list->list) {
140 frame_list->list = info;
141 frame_list->end = info;
143 av_assert0(frame_list->end != NULL);
144 frame_list->end->next = info;
145 frame_list->end = info;
147 frame_list->nb_frames++;
148 frame_list->nb_samples += nb_samples;
154 typedef struct MixContext {
155 const AVClass *class; /**< class for AVOptions */
156 AVFloatDSPContext fdsp;
158 int nb_inputs; /**< number of inputs */
159 int active_inputs; /**< number of input currently active */
160 int duration_mode; /**< mode for determining duration */
161 float dropout_transition; /**< transition time when an input drops out */
163 int nb_channels; /**< number of channels */
164 int sample_rate; /**< sample rate */
166 AVAudioFifo **fifos; /**< audio fifo for each input */
167 uint8_t *input_state; /**< current state of each input */
168 float *input_scale; /**< mixing scale factor for each input */
169 float scale_norm; /**< normalization factor for all inputs */
170 int64_t next_pts; /**< calculated pts for next output frame */
171 FrameList *frame_list; /**< list of frame info for the first input */
174 #define OFFSET(x) offsetof(MixContext, x)
175 #define A AV_OPT_FLAG_AUDIO_PARAM
176 static const AVOption amix_options[] = {
177 { "inputs", "Number of inputs.",
178 OFFSET(nb_inputs), AV_OPT_TYPE_INT, { 2 }, 1, 32, A },
179 { "duration", "How to determine the end-of-stream.",
180 OFFSET(duration_mode), AV_OPT_TYPE_INT, { DURATION_LONGEST }, 0, 2, A, "duration" },
181 { "longest", "Duration of longest input.", 0, AV_OPT_TYPE_CONST, { DURATION_LONGEST }, INT_MIN, INT_MAX, A, "duration" },
182 { "shortest", "Duration of shortest input.", 0, AV_OPT_TYPE_CONST, { DURATION_SHORTEST }, INT_MIN, INT_MAX, A, "duration" },
183 { "first", "Duration of first input.", 0, AV_OPT_TYPE_CONST, { DURATION_FIRST }, INT_MIN, INT_MAX, A, "duration" },
184 { "dropout_transition", "Transition time, in seconds, for volume "
185 "renormalization when an input stream ends.",
186 OFFSET(dropout_transition), AV_OPT_TYPE_FLOAT, { 2.0 }, 0, INT_MAX, A },
190 AVFILTER_DEFINE_CLASS(amix);
193 * Update the scaling factors to apply to each input during mixing.
195 * This balances the full volume range between active inputs and handles
196 * volume transitions when EOF is encountered on an input but mixing continues
197 * with the remaining inputs.
199 static void calculate_scales(MixContext *s, int nb_samples)
203 if (s->scale_norm > s->active_inputs) {
204 s->scale_norm -= nb_samples / (s->dropout_transition * s->sample_rate);
205 s->scale_norm = FFMAX(s->scale_norm, s->active_inputs);
208 for (i = 0; i < s->nb_inputs; i++) {
209 if (s->input_state[i] == INPUT_ON)
210 s->input_scale[i] = 1.0f / s->scale_norm;
212 s->input_scale[i] = 0.0f;
216 static int config_output(AVFilterLink *outlink)
218 AVFilterContext *ctx = outlink->src;
219 MixContext *s = ctx->priv;
223 s->planar = av_sample_fmt_is_planar(outlink->format);
224 s->sample_rate = outlink->sample_rate;
225 outlink->time_base = (AVRational){ 1, outlink->sample_rate };
226 s->next_pts = AV_NOPTS_VALUE;
228 s->frame_list = av_mallocz(sizeof(*s->frame_list));
230 return AVERROR(ENOMEM);
232 s->fifos = av_mallocz(s->nb_inputs * sizeof(*s->fifos));
234 return AVERROR(ENOMEM);
236 s->nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
237 for (i = 0; i < s->nb_inputs; i++) {
238 s->fifos[i] = av_audio_fifo_alloc(outlink->format, s->nb_channels, 1024);
240 return AVERROR(ENOMEM);
243 s->input_state = av_malloc(s->nb_inputs);
245 return AVERROR(ENOMEM);
246 memset(s->input_state, INPUT_ON, s->nb_inputs);
247 s->active_inputs = s->nb_inputs;
249 s->input_scale = av_mallocz(s->nb_inputs * sizeof(*s->input_scale));
251 return AVERROR(ENOMEM);
252 s->scale_norm = s->active_inputs;
253 calculate_scales(s, 0);
255 av_get_channel_layout_string(buf, sizeof(buf), -1, outlink->channel_layout);
257 av_log(ctx, AV_LOG_VERBOSE,
258 "inputs:%d fmt:%s srate:%d cl:%s\n", s->nb_inputs,
259 av_get_sample_fmt_name(outlink->format), outlink->sample_rate, buf);
265 * Read samples from the input FIFOs, mix, and write to the output link.
267 static int output_frame(AVFilterLink *outlink, int nb_samples)
269 AVFilterContext *ctx = outlink->src;
270 MixContext *s = ctx->priv;
271 AVFilterBufferRef *out_buf, *in_buf;
274 calculate_scales(s, nb_samples);
276 out_buf = ff_get_audio_buffer(outlink, AV_PERM_WRITE, nb_samples);
278 return AVERROR(ENOMEM);
280 in_buf = ff_get_audio_buffer(outlink, AV_PERM_WRITE, nb_samples);
282 return AVERROR(ENOMEM);
284 for (i = 0; i < s->nb_inputs; i++) {
285 if (s->input_state[i] == INPUT_ON) {
286 int planes, plane_size, p;
288 av_audio_fifo_read(s->fifos[i], (void **)in_buf->extended_data,
291 planes = s->planar ? s->nb_channels : 1;
292 plane_size = nb_samples * (s->planar ? 1 : s->nb_channels);
293 plane_size = FFALIGN(plane_size, 16);
295 for (p = 0; p < planes; p++) {
296 s->fdsp.vector_fmac_scalar((float *)out_buf->extended_data[p],
297 (float *) in_buf->extended_data[p],
298 s->input_scale[i], plane_size);
302 avfilter_unref_buffer(in_buf);
304 out_buf->pts = s->next_pts;
305 if (s->next_pts != AV_NOPTS_VALUE)
306 s->next_pts += nb_samples;
308 return ff_filter_samples(outlink, out_buf);
312 * Returns the smallest number of samples available in the input FIFOs other
313 * than that of the first input.
315 static int get_available_samples(MixContext *s)
318 int available_samples = INT_MAX;
320 av_assert0(s->nb_inputs > 1);
322 for (i = 1; i < s->nb_inputs; i++) {
324 if (s->input_state[i] == INPUT_OFF)
326 nb_samples = av_audio_fifo_size(s->fifos[i]);
327 available_samples = FFMIN(available_samples, nb_samples);
329 if (available_samples == INT_MAX)
331 return available_samples;
335 * Requests a frame, if needed, from each input link other than the first.
337 static int request_samples(AVFilterContext *ctx, int min_samples)
339 MixContext *s = ctx->priv;
342 av_assert0(s->nb_inputs > 1);
344 for (i = 1; i < s->nb_inputs; i++) {
346 if (s->input_state[i] == INPUT_OFF)
348 while (!ret && av_audio_fifo_size(s->fifos[i]) < min_samples)
349 ret = ff_request_frame(ctx->inputs[i]);
350 if (ret == AVERROR_EOF) {
351 if (av_audio_fifo_size(s->fifos[i]) == 0) {
352 s->input_state[i] = INPUT_OFF;
362 * Calculates the number of active inputs and determines EOF based on the
365 * @return 0 if mixing should continue, or AVERROR_EOF if mixing should stop.
367 static int calc_active_inputs(MixContext *s)
370 int active_inputs = 0;
371 for (i = 0; i < s->nb_inputs; i++)
372 active_inputs += !!(s->input_state[i] != INPUT_OFF);
373 s->active_inputs = active_inputs;
375 if (!active_inputs ||
376 (s->duration_mode == DURATION_FIRST && s->input_state[0] == INPUT_OFF) ||
377 (s->duration_mode == DURATION_SHORTEST && active_inputs != s->nb_inputs))
382 static int request_frame(AVFilterLink *outlink)
384 AVFilterContext *ctx = outlink->src;
385 MixContext *s = ctx->priv;
387 int wanted_samples, available_samples;
389 ret = calc_active_inputs(s);
393 if (s->input_state[0] == INPUT_OFF) {
394 ret = request_samples(ctx, 1);
398 ret = calc_active_inputs(s);
402 available_samples = get_available_samples(s);
403 if (!available_samples)
404 return AVERROR(EAGAIN);
406 return output_frame(outlink, available_samples);
409 if (s->frame_list->nb_frames == 0) {
410 ret = ff_request_frame(ctx->inputs[0]);
411 if (ret == AVERROR_EOF) {
412 s->input_state[0] = INPUT_OFF;
413 if (s->nb_inputs == 1)
416 return AVERROR(EAGAIN);
420 av_assert0(s->frame_list->nb_frames > 0);
422 wanted_samples = frame_list_next_frame_size(s->frame_list);
424 if (s->active_inputs > 1) {
425 ret = request_samples(ctx, wanted_samples);
429 ret = calc_active_inputs(s);
434 if (s->active_inputs > 1) {
435 available_samples = get_available_samples(s);
436 if (!available_samples)
437 return AVERROR(EAGAIN);
438 available_samples = FFMIN(available_samples, wanted_samples);
440 available_samples = wanted_samples;
443 s->next_pts = frame_list_next_pts(s->frame_list);
444 frame_list_remove_samples(s->frame_list, available_samples);
446 return output_frame(outlink, available_samples);
449 static int filter_samples(AVFilterLink *inlink, AVFilterBufferRef *buf)
451 AVFilterContext *ctx = inlink->dst;
452 MixContext *s = ctx->priv;
453 AVFilterLink *outlink = ctx->outputs[0];
456 for (i = 0; i < ctx->nb_inputs; i++)
457 if (ctx->inputs[i] == inlink)
459 if (i >= ctx->nb_inputs) {
460 av_log(ctx, AV_LOG_ERROR, "unknown input link\n");
461 ret = AVERROR(EINVAL);
466 int64_t pts = av_rescale_q(buf->pts, inlink->time_base,
468 ret = frame_list_add_frame(s->frame_list, buf->audio->nb_samples, pts);
473 ret = av_audio_fifo_write(s->fifos[i], (void **)buf->extended_data,
474 buf->audio->nb_samples);
477 avfilter_unref_buffer(buf);
482 static int init(AVFilterContext *ctx, const char *args)
484 MixContext *s = ctx->priv;
487 s->class = &amix_class;
488 av_opt_set_defaults(s);
490 if ((ret = av_set_options_string(s, args, "=", ":")) < 0)
494 for (i = 0; i < s->nb_inputs; i++) {
496 AVFilterPad pad = { 0 };
498 snprintf(name, sizeof(name), "input%d", i);
499 pad.type = AVMEDIA_TYPE_AUDIO;
500 pad.name = av_strdup(name);
501 pad.filter_samples = filter_samples;
503 ff_insert_inpad(ctx, i, &pad);
506 avpriv_float_dsp_init(&s->fdsp, 0);
511 static void uninit(AVFilterContext *ctx)
514 MixContext *s = ctx->priv;
517 for (i = 0; i < s->nb_inputs; i++)
518 av_audio_fifo_free(s->fifos[i]);
521 frame_list_clear(s->frame_list);
522 av_freep(&s->frame_list);
523 av_freep(&s->input_state);
524 av_freep(&s->input_scale);
526 for (i = 0; i < ctx->nb_inputs; i++)
527 av_freep(&ctx->input_pads[i].name);
530 static int query_formats(AVFilterContext *ctx)
532 AVFilterFormats *formats = NULL;
533 ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
534 ff_add_format(&formats, AV_SAMPLE_FMT_FLTP);
535 ff_set_common_formats(ctx, formats);
536 ff_set_common_channel_layouts(ctx, ff_all_channel_layouts());
537 ff_set_common_samplerates(ctx, ff_all_samplerates());
541 AVFilter avfilter_af_amix = {
543 .description = NULL_IF_CONFIG_SMALL("Audio mixing."),
544 .priv_size = sizeof(MixContext),
548 .query_formats = query_formats,
550 .inputs = (const AVFilterPad[]) {{ .name = NULL}},
551 .outputs = (const AVFilterPad[]) {{ .name = "default",
552 .type = AVMEDIA_TYPE_AUDIO,
553 .config_props = config_output,
554 .request_frame = request_frame },