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/mathematics.h"
36 #include "libavutil/opt.h"
37 #include "libavutil/samplefmt.h"
44 #define INPUT_OFF 0 /**< input has reached EOF */
45 #define INPUT_ON 1 /**< input is active */
46 #define INPUT_INACTIVE 2 /**< input is on, but is currently inactive */
48 #define DURATION_LONGEST 0
49 #define DURATION_SHORTEST 1
50 #define DURATION_FIRST 2
53 typedef struct FrameInfo {
56 struct FrameInfo *next;
60 * Linked list used to store timestamps and frame sizes of all frames in the
61 * FIFO for the first input.
63 * This is needed to keep timestamps synchronized for the case where multiple
64 * input frames are pushed to the filter for processing before a frame is
65 * requested by the output link.
67 typedef struct FrameList {
74 static void frame_list_clear(FrameList *frame_list)
77 while (frame_list->list) {
78 FrameInfo *info = frame_list->list;
79 frame_list->list = info->next;
82 frame_list->nb_frames = 0;
83 frame_list->nb_samples = 0;
84 frame_list->end = NULL;
88 static int frame_list_next_frame_size(FrameList *frame_list)
90 if (!frame_list->list)
92 return frame_list->list->nb_samples;
95 static int64_t frame_list_next_pts(FrameList *frame_list)
97 if (!frame_list->list)
98 return AV_NOPTS_VALUE;
99 return frame_list->list->pts;
102 static void frame_list_remove_samples(FrameList *frame_list, int nb_samples)
104 if (nb_samples >= frame_list->nb_samples) {
105 frame_list_clear(frame_list);
107 int samples = nb_samples;
108 while (samples > 0) {
109 FrameInfo *info = frame_list->list;
110 av_assert0(info != NULL);
111 if (info->nb_samples <= samples) {
112 samples -= info->nb_samples;
113 frame_list->list = info->next;
114 if (!frame_list->list)
115 frame_list->end = NULL;
116 frame_list->nb_frames--;
117 frame_list->nb_samples -= info->nb_samples;
120 info->nb_samples -= samples;
121 info->pts += samples;
122 frame_list->nb_samples -= samples;
129 static int frame_list_add_frame(FrameList *frame_list, int nb_samples, int64_t pts)
131 FrameInfo *info = av_malloc(sizeof(*info));
133 return AVERROR(ENOMEM);
134 info->nb_samples = nb_samples;
138 if (!frame_list->list) {
139 frame_list->list = info;
140 frame_list->end = info;
142 av_assert0(frame_list->end != NULL);
143 frame_list->end->next = info;
144 frame_list->end = info;
146 frame_list->nb_frames++;
147 frame_list->nb_samples += nb_samples;
153 typedef struct MixContext {
154 const AVClass *class; /**< class for AVOptions */
156 int nb_inputs; /**< number of inputs */
157 int active_inputs; /**< number of input currently active */
158 int duration_mode; /**< mode for determining duration */
159 float dropout_transition; /**< transition time when an input drops out */
161 int nb_channels; /**< number of channels */
162 int sample_rate; /**< sample rate */
163 AVAudioFifo **fifos; /**< audio fifo for each input */
164 uint8_t *input_state; /**< current state of each input */
165 float *input_scale; /**< mixing scale factor for each input */
166 float scale_norm; /**< normalization factor for all inputs */
167 int64_t next_pts; /**< calculated pts for next output frame */
168 FrameList *frame_list; /**< list of frame info for the first input */
171 #define OFFSET(x) offsetof(MixContext, x)
172 #define A AV_OPT_FLAG_AUDIO_PARAM
173 static const AVOption options[] = {
174 { "inputs", "Number of inputs.",
175 OFFSET(nb_inputs), AV_OPT_TYPE_INT, { 2 }, 1, 32, A },
176 { "duration", "How to determine the end-of-stream.",
177 OFFSET(duration_mode), AV_OPT_TYPE_INT, { DURATION_LONGEST }, 0, 2, A, "duration" },
178 { "longest", "Duration of longest input.", 0, AV_OPT_TYPE_CONST, { DURATION_LONGEST }, INT_MIN, INT_MAX, A, "duration" },
179 { "shortest", "Duration of shortest input.", 0, AV_OPT_TYPE_CONST, { DURATION_SHORTEST }, INT_MIN, INT_MAX, A, "duration" },
180 { "first", "Duration of first input.", 0, AV_OPT_TYPE_CONST, { DURATION_FIRST }, INT_MIN, INT_MAX, A, "duration" },
181 { "dropout_transition", "Transition time, in seconds, for volume "
182 "renormalization when an input stream ends.",
183 OFFSET(dropout_transition), AV_OPT_TYPE_FLOAT, { 2.0 }, 0, INT_MAX, A },
187 static const AVClass amix_class = {
188 .class_name = "amix filter",
189 .item_name = av_default_item_name,
191 .version = LIBAVUTIL_VERSION_INT,
196 * Update the scaling factors to apply to each input during mixing.
198 * This balances the full volume range between active inputs and handles
199 * volume transitions when EOF is encountered on an input but mixing continues
200 * with the remaining inputs.
202 static void calculate_scales(MixContext *s, int nb_samples)
206 if (s->scale_norm > s->active_inputs) {
207 s->scale_norm -= nb_samples / (s->dropout_transition * s->sample_rate);
208 s->scale_norm = FFMAX(s->scale_norm, s->active_inputs);
211 for (i = 0; i < s->nb_inputs; i++) {
212 if (s->input_state[i] == INPUT_ON)
213 s->input_scale[i] = 1.0f / s->scale_norm;
215 s->input_scale[i] = 0.0f;
219 static int config_output(AVFilterLink *outlink)
221 AVFilterContext *ctx = outlink->src;
222 MixContext *s = ctx->priv;
226 s->sample_rate = outlink->sample_rate;
227 outlink->time_base = (AVRational){ 1, outlink->sample_rate };
228 s->next_pts = AV_NOPTS_VALUE;
230 s->frame_list = av_mallocz(sizeof(*s->frame_list));
232 return AVERROR(ENOMEM);
234 s->fifos = av_mallocz(s->nb_inputs * sizeof(*s->fifos));
236 return AVERROR(ENOMEM);
238 s->nb_channels = av_get_channel_layout_nb_channels(outlink->channel_layout);
239 for (i = 0; i < s->nb_inputs; i++) {
240 s->fifos[i] = av_audio_fifo_alloc(outlink->format, s->nb_channels, 1024);
242 return AVERROR(ENOMEM);
245 s->input_state = av_malloc(s->nb_inputs);
247 return AVERROR(ENOMEM);
248 memset(s->input_state, INPUT_ON, s->nb_inputs);
249 s->active_inputs = s->nb_inputs;
251 s->input_scale = av_mallocz(s->nb_inputs * sizeof(*s->input_scale));
253 return AVERROR(ENOMEM);
254 s->scale_norm = s->active_inputs;
255 calculate_scales(s, 0);
257 av_get_channel_layout_string(buf, sizeof(buf), -1, outlink->channel_layout);
259 av_log(ctx, AV_LOG_VERBOSE,
260 "inputs:%d fmt:%s srate:%"PRId64" cl:%s\n", s->nb_inputs,
261 av_get_sample_fmt_name(outlink->format), outlink->sample_rate, buf);
266 /* TODO: move optimized version from DSPContext to libavutil */
267 static void vector_fmac_scalar(float *dst, const float *src, float mul, int len)
270 for (i = 0; i < len; i++)
271 dst[i] += src[i] * mul;
275 * Read samples from the input FIFOs, mix, and write to the output link.
277 static int output_frame(AVFilterLink *outlink, int nb_samples)
279 AVFilterContext *ctx = outlink->src;
280 MixContext *s = ctx->priv;
281 AVFilterBufferRef *out_buf, *in_buf;
284 calculate_scales(s, nb_samples);
286 out_buf = ff_get_audio_buffer(outlink, AV_PERM_WRITE, nb_samples);
288 return AVERROR(ENOMEM);
290 in_buf = ff_get_audio_buffer(outlink, AV_PERM_WRITE, nb_samples);
292 return AVERROR(ENOMEM);
294 for (i = 0; i < s->nb_inputs; i++) {
295 if (s->input_state[i] == INPUT_ON) {
296 av_audio_fifo_read(s->fifos[i], (void **)in_buf->extended_data,
298 vector_fmac_scalar((float *)out_buf->extended_data[0],
299 (float *) in_buf->extended_data[0],
300 s->input_scale[i], nb_samples * s->nb_channels);
303 avfilter_unref_buffer(in_buf);
305 out_buf->pts = s->next_pts;
306 if (s->next_pts != AV_NOPTS_VALUE)
307 s->next_pts += nb_samples;
309 ff_filter_samples(outlink, out_buf);
315 * Returns the smallest number of samples available in the input FIFOs other
316 * than that of the first input.
318 static int get_available_samples(MixContext *s)
321 int available_samples = INT_MAX;
323 av_assert0(s->nb_inputs > 1);
325 for (i = 1; i < s->nb_inputs; i++) {
327 if (s->input_state[i] == INPUT_OFF)
329 nb_samples = av_audio_fifo_size(s->fifos[i]);
330 available_samples = FFMIN(available_samples, nb_samples);
332 if (available_samples == INT_MAX)
334 return available_samples;
338 * Requests a frame, if needed, from each input link other than the first.
340 static int request_samples(AVFilterContext *ctx, int min_samples)
342 MixContext *s = ctx->priv;
345 av_assert0(s->nb_inputs > 1);
347 for (i = 1; i < s->nb_inputs; i++) {
349 if (s->input_state[i] == INPUT_OFF)
351 while (!ret && av_audio_fifo_size(s->fifos[i]) < min_samples)
352 ret = avfilter_request_frame(ctx->inputs[i]);
353 if (ret == AVERROR_EOF) {
354 if (av_audio_fifo_size(s->fifos[i]) == 0) {
355 s->input_state[i] = INPUT_OFF;
365 * Calculates the number of active inputs and determines EOF based on the
368 * @return 0 if mixing should continue, or AVERROR_EOF if mixing should stop.
370 static int calc_active_inputs(MixContext *s)
373 int active_inputs = 0;
374 for (i = 0; i < s->nb_inputs; i++)
375 active_inputs += !!(s->input_state[i] != INPUT_OFF);
376 s->active_inputs = active_inputs;
378 if (!active_inputs ||
379 (s->duration_mode == DURATION_FIRST && s->input_state[0] == INPUT_OFF) ||
380 (s->duration_mode == DURATION_SHORTEST && active_inputs != s->nb_inputs))
385 static int request_frame(AVFilterLink *outlink)
387 AVFilterContext *ctx = outlink->src;
388 MixContext *s = ctx->priv;
390 int wanted_samples, available_samples;
392 ret = calc_active_inputs(s);
396 if (s->input_state[0] == INPUT_OFF) {
397 ret = request_samples(ctx, 1);
401 ret = calc_active_inputs(s);
405 available_samples = get_available_samples(s);
406 if (!available_samples)
409 return output_frame(outlink, available_samples);
412 if (s->frame_list->nb_frames == 0) {
413 ret = avfilter_request_frame(ctx->inputs[0]);
414 if (ret == AVERROR_EOF) {
415 s->input_state[0] = INPUT_OFF;
416 if (s->nb_inputs == 1)
419 return AVERROR(EAGAIN);
423 av_assert0(s->frame_list->nb_frames > 0);
425 wanted_samples = frame_list_next_frame_size(s->frame_list);
427 if (s->active_inputs > 1) {
428 ret = request_samples(ctx, wanted_samples);
432 ret = calc_active_inputs(s);
436 available_samples = get_available_samples(s);
437 if (!available_samples)
439 available_samples = FFMIN(available_samples, wanted_samples);
441 available_samples = wanted_samples;
444 s->next_pts = frame_list_next_pts(s->frame_list);
445 frame_list_remove_samples(s->frame_list, available_samples);
447 return output_frame(outlink, available_samples);
450 static void filter_samples(AVFilterLink *inlink, AVFilterBufferRef *buf)
452 AVFilterContext *ctx = inlink->dst;
453 MixContext *s = ctx->priv;
454 AVFilterLink *outlink = ctx->outputs[0];
457 for (i = 0; i < ctx->input_count; i++)
458 if (ctx->inputs[i] == inlink)
460 if (i >= ctx->input_count) {
461 av_log(ctx, AV_LOG_ERROR, "unknown input link\n");
466 int64_t pts = av_rescale_q(buf->pts, inlink->time_base,
468 frame_list_add_frame(s->frame_list, buf->audio->nb_samples, pts);
471 av_audio_fifo_write(s->fifos[i], (void **)buf->extended_data,
472 buf->audio->nb_samples);
474 avfilter_unref_buffer(buf);
477 static int init(AVFilterContext *ctx, const char *args, void *opaque)
479 MixContext *s = ctx->priv;
482 s->class = &amix_class;
483 av_opt_set_defaults(s);
485 if ((ret = av_set_options_string(s, args, "=", ":")) < 0) {
486 av_log(ctx, AV_LOG_ERROR, "Error parsing options string '%s'.\n", args);
491 for (i = 0; i < s->nb_inputs; i++) {
493 AVFilterPad pad = { 0 };
495 snprintf(name, sizeof(name), "input%d", i);
496 pad.type = AVMEDIA_TYPE_AUDIO;
497 pad.name = av_strdup(name);
498 pad.filter_samples = filter_samples;
500 avfilter_insert_inpad(ctx, i, &pad);
506 static void uninit(AVFilterContext *ctx)
509 MixContext *s = ctx->priv;
512 for (i = 0; i < s->nb_inputs; i++)
513 av_audio_fifo_free(s->fifos[i]);
516 frame_list_clear(s->frame_list);
517 av_freep(&s->frame_list);
518 av_freep(&s->input_state);
519 av_freep(&s->input_scale);
521 for (i = 0; i < ctx->input_count; i++)
522 av_freep(&ctx->input_pads[i].name);
525 static int query_formats(AVFilterContext *ctx)
527 AVFilterFormats *formats = NULL;
528 ff_add_format(&formats, AV_SAMPLE_FMT_FLT);
529 ff_set_common_formats(ctx, formats);
530 ff_set_common_channel_layouts(ctx, ff_all_channel_layouts());
531 ff_set_common_samplerates(ctx, ff_all_samplerates());
535 AVFilter avfilter_af_amix = {
537 .description = NULL_IF_CONFIG_SMALL("Audio mixing."),
538 .priv_size = sizeof(MixContext),
542 .query_formats = query_formats,
544 .inputs = (const AVFilterPad[]) {{ .name = NULL}},
545 .outputs = (const AVFilterPad[]) {{ .name = "default",
546 .type = AVMEDIA_TYPE_AUDIO,
547 .config_props = config_output,
548 .request_frame = request_frame },