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
38 int comp; ///< current resample compensation
45 /* set by filter_frame() to signal an output frame to request_frame() */
49 #define OFFSET(x) offsetof(ASyncContext, x)
50 #define A AV_OPT_FLAG_AUDIO_PARAM
51 static const AVOption options[] = {
52 { "compensate", "Stretch/squeeze the data to make it match the timestamps", OFFSET(resample), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 1, A },
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 },
55 { "max_comp", "Maximum compensation in samples per second.", OFFSET(max_comp), AV_OPT_TYPE_INT, { .i64 = 500 }, 0, INT_MAX, A },
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 },
60 static const AVClass async_class = {
61 .class_name = "asyncts filter",
62 .item_name = av_default_item_name,
64 .version = LIBAVUTIL_VERSION_INT,
67 static int init(AVFilterContext *ctx)
69 ASyncContext *s = ctx->priv;
71 s->pts = AV_NOPTS_VALUE;
77 static void uninit(AVFilterContext *ctx)
79 ASyncContext *s = ctx->priv;
82 avresample_close(s->avr);
83 avresample_free(&s->avr);
87 static int config_props(AVFilterLink *link)
89 ASyncContext *s = link->src->priv;
92 s->min_delta = s->min_delta_sec * link->sample_rate;
93 link->time_base = (AVRational){1, link->sample_rate};
95 s->avr = avresample_alloc_context();
97 return AVERROR(ENOMEM);
99 av_opt_set_int(s->avr, "in_channel_layout", link->channel_layout, 0);
100 av_opt_set_int(s->avr, "out_channel_layout", link->channel_layout, 0);
101 av_opt_set_int(s->avr, "in_sample_fmt", link->format, 0);
102 av_opt_set_int(s->avr, "out_sample_fmt", link->format, 0);
103 av_opt_set_int(s->avr, "in_sample_rate", link->sample_rate, 0);
104 av_opt_set_int(s->avr, "out_sample_rate", link->sample_rate, 0);
107 av_opt_set_int(s->avr, "force_resampling", 1, 0);
109 if ((ret = avresample_open(s->avr)) < 0)
115 /* get amount of data currently buffered, in samples */
116 static int64_t get_delay(ASyncContext *s)
118 return avresample_available(s->avr) + avresample_get_delay(s->avr);
121 static void handle_trimming(AVFilterContext *ctx)
123 ASyncContext *s = ctx->priv;
125 if (s->pts < s->first_pts) {
126 int delta = FFMIN(s->first_pts - s->pts, avresample_available(s->avr));
127 av_log(ctx, AV_LOG_VERBOSE, "Trimming %d samples from start\n",
129 avresample_read(s->avr, NULL, delta);
131 } else if (s->first_frame)
132 s->pts = s->first_pts;
135 static int request_frame(AVFilterLink *link)
137 AVFilterContext *ctx = link->src;
138 ASyncContext *s = ctx->priv;
143 while (ret >= 0 && !s->got_output)
144 ret = ff_request_frame(ctx->inputs[0]);
147 if (ret == AVERROR_EOF) {
148 if (s->first_pts != AV_NOPTS_VALUE)
149 handle_trimming(ctx);
151 if (nb_samples = get_delay(s)) {
152 AVFrame *buf = ff_get_audio_buffer(link, nb_samples);
154 return AVERROR(ENOMEM);
155 ret = avresample_convert(s->avr, buf->extended_data,
156 buf->linesize[0], nb_samples, NULL, 0, 0);
159 return (ret < 0) ? ret : AVERROR_EOF;
163 return ff_filter_frame(link, buf);
170 static int write_to_fifo(ASyncContext *s, AVFrame *buf)
172 int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
173 buf->linesize[0], buf->nb_samples);
178 static int filter_frame(AVFilterLink *inlink, AVFrame *buf)
180 AVFilterContext *ctx = inlink->dst;
181 ASyncContext *s = ctx->priv;
182 AVFilterLink *outlink = ctx->outputs[0];
183 int nb_channels = av_get_channel_layout_nb_channels(buf->channel_layout);
184 int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
185 av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
190 /* buffer data until we get the next timestamp */
191 if (s->pts == AV_NOPTS_VALUE || pts == AV_NOPTS_VALUE) {
192 if (pts != AV_NOPTS_VALUE) {
193 s->pts = pts - get_delay(s);
195 return write_to_fifo(s, buf);
198 if (s->first_pts != AV_NOPTS_VALUE) {
199 handle_trimming(ctx);
200 if (!avresample_available(s->avr))
201 return write_to_fifo(s, buf);
204 /* when we have two timestamps, compute how many samples would we have
205 * to add/remove to get proper sync between data and timestamps */
206 delta = pts - s->pts - get_delay(s);
207 out_size = avresample_available(s->avr);
209 if (labs(delta) > s->min_delta ||
210 (s->first_frame && delta && s->first_pts != AV_NOPTS_VALUE)) {
211 av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
212 out_size = av_clipl_int32((int64_t)out_size + delta);
215 // adjust the compensation if delta is non-zero
216 int delay = get_delay(s);
217 int comp = s->comp + av_clip(delta * inlink->sample_rate / delay,
218 -s->max_comp, s->max_comp);
219 if (comp != s->comp) {
220 av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
221 if (avresample_set_compensation(s->avr, comp, inlink->sample_rate) == 0) {
226 // adjust PTS to avoid monotonicity errors with input PTS jitter
232 AVFrame *buf_out = ff_get_audio_buffer(outlink, out_size);
234 ret = AVERROR(ENOMEM);
238 if (s->first_frame && delta > 0) {
241 av_samples_set_silence(buf_out->extended_data, 0, delta,
242 nb_channels, buf->format);
244 for (ch = 0; ch < nb_channels; ch++)
245 buf_out->extended_data[ch] += delta;
247 avresample_read(s->avr, buf_out->extended_data, out_size);
249 for (ch = 0; ch < nb_channels; ch++)
250 buf_out->extended_data[ch] -= delta;
252 avresample_read(s->avr, buf_out->extended_data, out_size);
255 av_samples_set_silence(buf_out->extended_data, out_size - delta,
256 delta, nb_channels, buf->format);
259 buf_out->pts = s->pts;
260 ret = ff_filter_frame(outlink, buf_out);
264 } else if (avresample_available(s->avr)) {
265 av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
269 /* drain any remaining buffered data */
270 avresample_read(s->avr, NULL, avresample_available(s->avr));
272 new_pts = pts - avresample_get_delay(s->avr);
273 /* check for s->pts monotonicity */
274 if (new_pts > s->pts) {
276 ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
277 buf->linesize[0], buf->nb_samples);
279 av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
291 static const AVFilterPad avfilter_af_asyncts_inputs[] = {
294 .type = AVMEDIA_TYPE_AUDIO,
295 .filter_frame = filter_frame,
300 static const AVFilterPad avfilter_af_asyncts_outputs[] = {
303 .type = AVMEDIA_TYPE_AUDIO,
304 .config_props = config_props,
305 .request_frame = request_frame
310 AVFilter avfilter_af_asyncts = {
312 .description = NULL_IF_CONFIG_SMALL("Sync audio data to timestamps"),
317 .priv_size = sizeof(ASyncContext),
318 .priv_class = &async_class,
320 .inputs = avfilter_af_asyncts_inputs,
321 .outputs = avfilter_af_asyncts_outputs,