2 * This file is part of Libav.
4 * Libav is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU Lesser General Public
6 * License as published by the Free Software Foundation; either
7 * version 2.1 of the License, or (at your option) any later version.
9 * Libav is distributed in the hope that it will be useful,
10 * but WITHOUT ANY WARRANTY; without even the implied warranty of
11 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
12 * Lesser General Public License for more details.
14 * You should have received a copy of the GNU Lesser General Public
15 * License along with Libav; if not, write to the Free Software
16 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
19 #include "libavresample/avresample.h"
20 #include "libavutil/audio_fifo.h"
21 #include "libavutil/common.h"
22 #include "libavutil/mathematics.h"
23 #include "libavutil/opt.h"
24 #include "libavutil/samplefmt.h"
30 typedef struct ASyncContext {
33 AVAudioResampleContext *avr;
34 int64_t pts; ///< timestamp in samples of the first sample in fifo
35 int min_delta; ///< pad/trim min threshold in samples
36 int first_frame; ///< 1 until filter_frame() has processed at least 1 frame with a pts != AV_NOPTS_VALUE
37 int64_t first_pts; ///< user-specified first expected pts, in samples
44 /* set by filter_frame() to signal an output frame to request_frame() */
48 #define OFFSET(x) offsetof(ASyncContext, x)
49 #define A AV_OPT_FLAG_AUDIO_PARAM
50 #define F AV_OPT_FLAG_FILTERING_PARAM
51 static const AVOption asyncts_options[] = {
52 { "compensate", "Stretch/squeeze the data to make it match the timestamps", OFFSET(resample), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, A|F },
53 { "min_delta", "Minimum difference between timestamps and audio data "
54 "(in seconds) to trigger padding/trimmin the data.", OFFSET(min_delta_sec), AV_OPT_TYPE_FLOAT, { .dbl = 0.1 }, 0, INT_MAX, A|F },
55 { "max_comp", "Maximum compensation in samples per second.", OFFSET(max_comp), AV_OPT_TYPE_INT, { .i64 = 500 }, 0, INT_MAX, A|F },
56 { "first_pts", "Assume the first pts should be this value.", OFFSET(first_pts), AV_OPT_TYPE_INT64, { .i64 = AV_NOPTS_VALUE }, INT64_MIN, INT64_MAX, A|F },
60 AVFILTER_DEFINE_CLASS(asyncts);
62 static int init(AVFilterContext *ctx, const char *args)
64 ASyncContext *s = ctx->priv;
67 s->class = &asyncts_class;
68 av_opt_set_defaults(s);
70 if ((ret = av_set_options_string(s, args, "=", ":")) < 0)
74 s->pts = AV_NOPTS_VALUE;
80 static void uninit(AVFilterContext *ctx)
82 ASyncContext *s = ctx->priv;
85 avresample_close(s->avr);
86 avresample_free(&s->avr);
90 static int config_props(AVFilterLink *link)
92 ASyncContext *s = link->src->priv;
95 s->min_delta = s->min_delta_sec * link->sample_rate;
96 link->time_base = (AVRational){1, link->sample_rate};
98 s->avr = avresample_alloc_context();
100 return AVERROR(ENOMEM);
102 av_opt_set_int(s->avr, "in_channel_layout", link->channel_layout, 0);
103 av_opt_set_int(s->avr, "out_channel_layout", link->channel_layout, 0);
104 av_opt_set_int(s->avr, "in_sample_fmt", link->format, 0);
105 av_opt_set_int(s->avr, "out_sample_fmt", link->format, 0);
106 av_opt_set_int(s->avr, "in_sample_rate", link->sample_rate, 0);
107 av_opt_set_int(s->avr, "out_sample_rate", link->sample_rate, 0);
110 av_opt_set_int(s->avr, "force_resampling", 1, 0);
112 if ((ret = avresample_open(s->avr)) < 0)
118 /* get amount of data currently buffered, in samples */
119 static int64_t get_delay(ASyncContext *s)
121 return avresample_available(s->avr) + avresample_get_delay(s->avr);
124 static void handle_trimming(AVFilterContext *ctx)
126 ASyncContext *s = ctx->priv;
128 if (s->pts < s->first_pts) {
129 int delta = FFMIN(s->first_pts - s->pts, avresample_available(s->avr));
130 av_log(ctx, AV_LOG_VERBOSE, "Trimming %d samples from start\n",
132 avresample_read(s->avr, NULL, delta);
134 } else if (s->first_frame)
135 s->pts = s->first_pts;
138 static int request_frame(AVFilterLink *link)
140 AVFilterContext *ctx = link->src;
141 ASyncContext *s = ctx->priv;
146 while (ret >= 0 && !s->got_output)
147 ret = ff_request_frame(ctx->inputs[0]);
150 if (ret == AVERROR_EOF) {
151 if (s->first_pts != AV_NOPTS_VALUE)
152 handle_trimming(ctx);
154 if (nb_samples = get_delay(s)) {
155 AVFilterBufferRef *buf = ff_get_audio_buffer(link, AV_PERM_WRITE,
158 return AVERROR(ENOMEM);
159 ret = avresample_convert(s->avr, buf->extended_data,
160 buf->linesize[0], nb_samples, NULL, 0, 0);
162 avfilter_unref_bufferp(&buf);
163 return (ret < 0) ? ret : AVERROR_EOF;
167 return ff_filter_frame(link, buf);
174 static int write_to_fifo(ASyncContext *s, AVFilterBufferRef *buf)
176 int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
177 buf->linesize[0], buf->audio->nb_samples);
178 avfilter_unref_buffer(buf);
182 static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *buf)
184 AVFilterContext *ctx = inlink->dst;
185 ASyncContext *s = ctx->priv;
186 AVFilterLink *outlink = ctx->outputs[0];
187 int nb_channels = av_get_channel_layout_nb_channels(buf->audio->channel_layout);
188 int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
189 av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
193 /* buffer data until we get the first timestamp */
194 if (s->pts == AV_NOPTS_VALUE) {
195 if (pts != AV_NOPTS_VALUE) {
196 s->pts = pts - get_delay(s);
198 return write_to_fifo(s, buf);
201 /* now wait for the next timestamp */
202 if (pts == AV_NOPTS_VALUE) {
203 return write_to_fifo(s, buf);
206 if (s->first_pts != AV_NOPTS_VALUE) {
207 handle_trimming(ctx);
208 if (!avresample_available(s->avr))
209 return write_to_fifo(s, buf);
212 /* when we have two timestamps, compute how many samples would we have
213 * to add/remove to get proper sync between data and timestamps */
214 delta = pts - s->pts - get_delay(s);
215 out_size = avresample_available(s->avr);
217 if (labs(delta) > s->min_delta || (s->first_frame && delta)) {
218 av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
219 out_size = av_clipl_int32((int64_t)out_size + delta);
222 int comp = av_clip(delta, -s->max_comp, s->max_comp);
223 av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
224 avresample_set_compensation(s->avr, comp, inlink->sample_rate);
230 AVFilterBufferRef *buf_out = ff_get_audio_buffer(outlink, AV_PERM_WRITE,
233 ret = AVERROR(ENOMEM);
237 if (s->first_frame && delta > 0) {
240 av_samples_set_silence(buf_out->extended_data, 0, delta,
241 nb_channels, buf->format);
243 for (ch = 0; ch < nb_channels; ch++)
244 buf_out->extended_data[ch] += delta;
246 avresample_read(s->avr, buf_out->extended_data, out_size);
248 for (ch = 0; ch < nb_channels; ch++)
249 buf_out->extended_data[ch] -= delta;
251 avresample_read(s->avr, buf_out->extended_data, out_size);
254 av_samples_set_silence(buf_out->extended_data, out_size - delta,
255 delta, nb_channels, buf->format);
258 buf_out->pts = s->pts;
259 ret = ff_filter_frame(outlink, buf_out);
263 } else if (avresample_available(s->avr)) {
264 av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
268 /* drain any remaining buffered data */
269 avresample_read(s->avr, NULL, avresample_available(s->avr));
271 s->pts = pts - avresample_get_delay(s->avr);
272 ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
273 buf->linesize[0], buf->audio->nb_samples);
277 avfilter_unref_buffer(buf);
282 static const AVFilterPad avfilter_af_asyncts_inputs[] = {
285 .type = AVMEDIA_TYPE_AUDIO,
286 .filter_frame = filter_frame
291 static const AVFilterPad avfilter_af_asyncts_outputs[] = {
294 .type = AVMEDIA_TYPE_AUDIO,
295 .config_props = config_props,
296 .request_frame = request_frame
301 AVFilter avfilter_af_asyncts = {
303 .description = NULL_IF_CONFIG_SMALL("Sync audio data to timestamps"),
308 .priv_size = sizeof(ASyncContext),
310 .inputs = avfilter_af_asyncts_inputs,
311 .outputs = avfilter_af_asyncts_outputs,
312 .priv_class = &asyncts_class,