FFmpeg
 All Data Structures Files Functions Variables Typedefs Enumerations Enumerator Macros Groups Pages
af_asyncts.c
Go to the documentation of this file.
1 /*
2  * This file is part of Libav.
3  *
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.
8  *
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.
13  *
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
17  */
18 
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"
25 
26 #include "audio.h"
27 #include "avfilter.h"
28 #include "internal.h"
29 
30 typedef struct ASyncContext {
31  const AVClass *class;
32 
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 
39  /* options */
40  int resample;
42  int max_comp;
43 
44  /* set by filter_frame() to signal an output frame to request_frame() */
46 } ASyncContext;
47 
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 },
57  { NULL },
58 };
59 
60 AVFILTER_DEFINE_CLASS(asyncts);
61 
62 static int init(AVFilterContext *ctx, const char *args)
63 {
64  ASyncContext *s = ctx->priv;
65  int ret;
66 
67  s->class = &asyncts_class;
69 
70  if ((ret = av_set_options_string(s, args, "=", ":")) < 0)
71  return ret;
72  av_opt_free(s);
73 
74  s->pts = AV_NOPTS_VALUE;
75  s->first_frame = 1;
76 
77  return 0;
78 }
79 
80 static void uninit(AVFilterContext *ctx)
81 {
82  ASyncContext *s = ctx->priv;
83 
84  if (s->avr) {
86  avresample_free(&s->avr);
87  }
88 }
89 
90 static int config_props(AVFilterLink *link)
91 {
92  ASyncContext *s = link->src->priv;
93  int ret;
94 
95  s->min_delta = s->min_delta_sec * link->sample_rate;
96  link->time_base = (AVRational){1, link->sample_rate};
97 
99  if (!s->avr)
100  return AVERROR(ENOMEM);
101 
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);
108 
109  if (s->resample)
110  av_opt_set_int(s->avr, "force_resampling", 1, 0);
111 
112  if ((ret = avresample_open(s->avr)) < 0)
113  return ret;
114 
115  return 0;
116 }
117 
118 /* get amount of data currently buffered, in samples */
119 static int64_t get_delay(ASyncContext *s)
120 {
122 }
123 
125 {
126  ASyncContext *s = ctx->priv;
127 
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",
131  delta);
132  avresample_read(s->avr, NULL, delta);
133  s->pts += delta;
134  } else if (s->first_frame)
135  s->pts = s->first_pts;
136 }
137 
138 static int request_frame(AVFilterLink *link)
139 {
140  AVFilterContext *ctx = link->src;
141  ASyncContext *s = ctx->priv;
142  int ret = 0;
143  int nb_samples;
144 
145  s->got_output = 0;
146  while (ret >= 0 && !s->got_output)
147  ret = ff_request_frame(ctx->inputs[0]);
148 
149  /* flush the fifo */
150  if (ret == AVERROR_EOF) {
151  if (s->first_pts != AV_NOPTS_VALUE)
152  handle_trimming(ctx);
153 
154  if (nb_samples = get_delay(s)) {
156  nb_samples);
157  if (!buf)
158  return AVERROR(ENOMEM);
159  ret = avresample_convert(s->avr, buf->extended_data,
160  buf->linesize[0], nb_samples, NULL, 0, 0);
161  if (ret <= 0) {
163  return (ret < 0) ? ret : AVERROR_EOF;
164  }
165 
166  buf->pts = s->pts;
167  return ff_filter_frame(link, buf);
168  }
169  }
170 
171  return ret;
172 }
173 
175 {
176  int ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
177  buf->linesize[0], buf->audio->nb_samples);
179  return ret;
180 }
181 
182 static int filter_frame(AVFilterLink *inlink, AVFilterBufferRef *buf)
183 {
184  AVFilterContext *ctx = inlink->dst;
185  ASyncContext *s = ctx->priv;
186  AVFilterLink *outlink = ctx->outputs[0];
188  int64_t pts = (buf->pts == AV_NOPTS_VALUE) ? buf->pts :
189  av_rescale_q(buf->pts, inlink->time_base, outlink->time_base);
190  int out_size, ret;
191  int64_t delta;
192 
193  /* buffer data until we get the next timestamp */
194  if (s->pts == AV_NOPTS_VALUE || pts == AV_NOPTS_VALUE) {
195  if (pts != AV_NOPTS_VALUE) {
196  s->pts = pts - get_delay(s);
197  }
198  return write_to_fifo(s, buf);
199  }
200 
201  if (s->first_pts != AV_NOPTS_VALUE) {
202  handle_trimming(ctx);
203  if (!avresample_available(s->avr))
204  return write_to_fifo(s, buf);
205  }
206 
207  /* when we have two timestamps, compute how many samples would we have
208  * to add/remove to get proper sync between data and timestamps */
209  delta = pts - s->pts - get_delay(s);
210  out_size = avresample_available(s->avr);
211 
212  if (labs(delta) > s->min_delta ||
213  (s->first_frame && delta && s->first_pts != AV_NOPTS_VALUE)) {
214  av_log(ctx, AV_LOG_VERBOSE, "Discontinuity - %"PRId64" samples.\n", delta);
215  out_size = av_clipl_int32((int64_t)out_size + delta);
216  } else {
217  if (s->resample) {
218  int comp = av_clip(delta, -s->max_comp, s->max_comp);
219  av_log(ctx, AV_LOG_VERBOSE, "Compensating %d samples per second.\n", comp);
220  avresample_set_compensation(s->avr, comp, inlink->sample_rate);
221  }
222  delta = 0;
223  }
224 
225  if (out_size > 0) {
227  out_size);
228  if (!buf_out) {
229  ret = AVERROR(ENOMEM);
230  goto fail;
231  }
232 
233  if (s->first_frame && delta > 0) {
234  int ch;
235 
236  av_samples_set_silence(buf_out->extended_data, 0, delta,
237  nb_channels, buf->format);
238 
239  for (ch = 0; ch < nb_channels; ch++)
240  buf_out->extended_data[ch] += delta;
241 
242  avresample_read(s->avr, buf_out->extended_data, out_size);
243 
244  for (ch = 0; ch < nb_channels; ch++)
245  buf_out->extended_data[ch] -= delta;
246  } else {
247  avresample_read(s->avr, buf_out->extended_data, out_size);
248 
249  if (delta > 0) {
250  av_samples_set_silence(buf_out->extended_data, out_size - delta,
251  delta, nb_channels, buf->format);
252  }
253  }
254  buf_out->pts = s->pts;
255  ret = ff_filter_frame(outlink, buf_out);
256  if (ret < 0)
257  goto fail;
258  s->got_output = 1;
259  } else if (avresample_available(s->avr)) {
260  av_log(ctx, AV_LOG_WARNING, "Non-monotonous timestamps, dropping "
261  "whole buffer.\n");
262  }
263 
264  /* drain any remaining buffered data */
266 
267  s->pts = pts - avresample_get_delay(s->avr);
268  ret = avresample_convert(s->avr, NULL, 0, 0, buf->extended_data,
269  buf->linesize[0], buf->audio->nb_samples);
270 
271  s->first_frame = 0;
272 fail:
274 
275  return ret;
276 }
277 
279  {
280  .name = "default",
281  .type = AVMEDIA_TYPE_AUDIO,
282  .filter_frame = filter_frame
283  },
284  { NULL }
285 };
286 
288  {
289  .name = "default",
290  .type = AVMEDIA_TYPE_AUDIO,
291  .config_props = config_props,
292  .request_frame = request_frame
293  },
294  { NULL }
295 };
296 
298  .name = "asyncts",
299  .description = NULL_IF_CONFIG_SMALL("Sync audio data to timestamps"),
300 
301  .init = init,
302  .uninit = uninit,
303 
304  .priv_size = sizeof(ASyncContext),
305 
306  .inputs = avfilter_af_asyncts_inputs,
307  .outputs = avfilter_af_asyncts_outputs,
308  .priv_class = &asyncts_class,
309 };