lavc: add a native Opus decoder.
[libav.git] / libavcodec / opusdec.c
1 /*
2 * Opus decoder
3 * Copyright (c) 2012 Andrew D'Addesio
4 * Copyright (c) 2013-2014 Mozilla Corporation
5 *
6 * This file is part of Libav.
7 *
8 * Libav is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU Lesser General Public
10 * License as published by the Free Software Foundation; either
11 * version 2.1 of the License, or (at your option) any later version.
12 *
13 * Libav is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
16 * Lesser General Public License for more details.
17 *
18 * You should have received a copy of the GNU Lesser General Public
19 * License along with Libav; if not, write to the Free Software
20 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
21 */
22
23 /**
24 * @file
25 * Opus decoder
26 * @author Andrew D'Addesio, Anton Khirnov
27 *
28 * Codec homepage: http://opus-codec.org/
29 * Specification: http://tools.ietf.org/html/rfc6716
30 * Ogg Opus specification: https://tools.ietf.org/html/draft-ietf-codec-oggopus-03
31 *
32 * Ogg-contained .opus files can be produced with opus-tools:
33 * http://git.xiph.org/?p=opus-tools.git
34 */
35
36 #include <stdint.h>
37
38 #include "libavutil/attributes.h"
39 #include "libavutil/audio_fifo.h"
40 #include "libavutil/channel_layout.h"
41 #include "libavutil/opt.h"
42
43 #include "libavresample/avresample.h"
44
45 #include "avcodec.h"
46 #include "celp_filters.h"
47 #include "fft.h"
48 #include "get_bits.h"
49 #include "internal.h"
50 #include "mathops.h"
51 #include "opus.h"
52
53 static const uint16_t silk_frame_duration_ms[16] = {
54 10, 20, 40, 60,
55 10, 20, 40, 60,
56 10, 20, 40, 60,
57 10, 20,
58 10, 20,
59 };
60
61 /* number of samples of silence to feed to the resampler
62 * at the beginning */
63 static const int silk_resample_delay[] = {
64 4, 8, 11, 11, 11
65 };
66
67 static const uint8_t celt_band_end[] = { 13, 17, 17, 19, 21 };
68
69 static int get_silk_samplerate(int config)
70 {
71 if (config < 4)
72 return 8000;
73 else if (config < 8)
74 return 12000;
75 return 16000;
76 }
77
78 /**
79 * Range decoder
80 */
81 static int opus_rc_init(OpusRangeCoder *rc, const uint8_t *data, int size)
82 {
83 int ret = init_get_bits8(&rc->gb, data, size);
84 if (ret < 0)
85 return ret;
86
87 rc->range = 128;
88 rc->value = 127 - get_bits(&rc->gb, 7);
89 rc->total_read_bits = 9;
90 opus_rc_normalize(rc);
91
92 return 0;
93 }
94
95 static void opus_raw_init(OpusRangeCoder *rc, const uint8_t *rightend,
96 unsigned int bytes)
97 {
98 rc->rb.position = rightend;
99 rc->rb.bytes = bytes;
100 rc->rb.cachelen = 0;
101 rc->rb.cacheval = 0;
102 }
103
104 static void opus_fade(float *out,
105 const float *in1, const float *in2,
106 const float *window, int len)
107 {
108 int i;
109 for (i = 0; i < len; i++)
110 out[i] = in2[i] * window[i] + in1[i] * (1.0 - window[i]);
111 }
112
113 static int opus_flush_resample(OpusStreamContext *s, int nb_samples)
114 {
115 int celt_size = av_audio_fifo_size(s->celt_delay);
116 int ret, i;
117
118 ret = avresample_convert(s->avr, (uint8_t**)s->out, s->out_size, nb_samples,
119 NULL, 0, 0);
120 if (ret < 0)
121 return ret;
122 else if (ret != nb_samples) {
123 av_log(s->avctx, AV_LOG_ERROR, "Wrong number of flushed samples: %d\n",
124 ret);
125 return AVERROR_BUG;
126 }
127
128 if (celt_size) {
129 if (celt_size != nb_samples) {
130 av_log(s->avctx, AV_LOG_ERROR, "Wrong number of CELT delay samples.\n");
131 return AVERROR_BUG;
132 }
133 av_audio_fifo_read(s->celt_delay, (void**)s->celt_output, nb_samples);
134 for (i = 0; i < s->output_channels; i++) {
135 s->fdsp->vector_fmac_scalar(s->out[i],
136 s->celt_output[i], 1.0,
137 nb_samples);
138 }
139 }
140
141 if (s->redundancy_idx) {
142 for (i = 0; i < s->output_channels; i++)
143 opus_fade(s->out[i], s->out[i],
144 s->redundancy_output[i] + 120 + s->redundancy_idx,
145 ff_celt_window2 + s->redundancy_idx, 120 - s->redundancy_idx);
146 s->redundancy_idx = 0;
147 }
148
149 s->out[0] += nb_samples;
150 s->out[1] += nb_samples;
151 s->out_size -= nb_samples * sizeof(float);
152
153 return 0;
154 }
155
156 static int opus_init_resample(OpusStreamContext *s)
157 {
158 float delay[16] = { 0.0 };
159 uint8_t *delayptr[2] = { (uint8_t*)delay, (uint8_t*)delay };
160 int ret;
161
162 av_opt_set_int(s->avr, "in_sample_rate", s->silk_samplerate, 0);
163 ret = avresample_open(s->avr);
164 if (ret < 0) {
165 av_log(s->avctx, AV_LOG_ERROR, "Error opening the resampler.\n");
166 return ret;
167 }
168
169 ret = avresample_convert(s->avr, NULL, 0, 0, delayptr, sizeof(delay),
170 silk_resample_delay[s->packet.bandwidth]);
171 if (ret < 0) {
172 av_log(s->avctx, AV_LOG_ERROR,
173 "Error feeding initial silence to the resampler.\n");
174 return ret;
175 }
176
177 return 0;
178 }
179
180 static int opus_decode_redundancy(OpusStreamContext *s, const uint8_t *data, int size)
181 {
182 int ret;
183 enum OpusBandwidth bw = s->packet.bandwidth;
184
185 if (s->packet.mode == OPUS_MODE_SILK &&
186 bw == OPUS_BANDWIDTH_MEDIUMBAND)
187 bw = OPUS_BANDWIDTH_WIDEBAND;
188
189 ret = opus_rc_init(&s->redundancy_rc, data, size);
190 if (ret < 0)
191 goto fail;
192 opus_raw_init(&s->redundancy_rc, data + size, size);
193
194 ret = ff_celt_decode_frame(s->celt, &s->redundancy_rc,
195 s->redundancy_output,
196 s->packet.stereo + 1, 240,
197 0, celt_band_end[s->packet.bandwidth]);
198 if (ret < 0)
199 goto fail;
200
201 return 0;
202 fail:
203 av_log(s->avctx, AV_LOG_ERROR, "Error decoding the redundancy frame.\n");
204 return ret;
205 }
206
207 static int opus_decode_frame(OpusStreamContext *s, const uint8_t *data, int size)
208 {
209 int samples = s->packet.frame_duration;
210 int redundancy = 0;
211 int redundancy_size, redundancy_pos;
212 int ret, i, consumed;
213 int delayed_samples = s->delayed_samples;
214
215 ret = opus_rc_init(&s->rc, data, size);
216 if (ret < 0)
217 return ret;
218
219 /* decode the silk frame */
220 if (s->packet.mode == OPUS_MODE_SILK || s->packet.mode == OPUS_MODE_HYBRID) {
221 if (!avresample_is_open(s->avr)) {
222 ret = opus_init_resample(s);
223 if (ret < 0)
224 return ret;
225 }
226
227 samples = ff_silk_decode_superframe(s->silk, &s->rc, s->silk_output,
228 FFMIN(s->packet.bandwidth, OPUS_BANDWIDTH_WIDEBAND),
229 s->packet.stereo + 1,
230 silk_frame_duration_ms[s->packet.config]);
231 if (samples < 0) {
232 av_log(s->avctx, AV_LOG_ERROR, "Error decoding a SILK frame.\n");
233 return samples;
234 }
235
236 samples = avresample_convert(s->avr, (uint8_t**)s->out, s->out_size,
237 s->packet.frame_duration,
238 (uint8_t**)s->silk_output,
239 sizeof(s->silk_buf[0]),
240 samples);
241 if (samples < 0) {
242 av_log(s->avctx, AV_LOG_ERROR, "Error resampling SILK data.\n");
243 return samples;
244 }
245 s->delayed_samples += s->packet.frame_duration - samples;
246 } else
247 ff_silk_flush(s->silk);
248
249 // decode redundancy information
250 consumed = opus_rc_tell(&s->rc);
251 if (s->packet.mode == OPUS_MODE_HYBRID && consumed + 37 <= size * 8)
252 redundancy = opus_rc_p2model(&s->rc, 12);
253 else if (s->packet.mode == OPUS_MODE_SILK && consumed + 17 <= size * 8)
254 redundancy = 1;
255
256 if (redundancy) {
257 redundancy_pos = opus_rc_p2model(&s->rc, 1);
258
259 if (s->packet.mode == OPUS_MODE_HYBRID)
260 redundancy_size = opus_rc_unimodel(&s->rc, 256) + 2;
261 else
262 redundancy_size = size - (consumed + 7) / 8;
263 size -= redundancy_size;
264 if (size < 0) {
265 av_log(s->avctx, AV_LOG_ERROR, "Invalid redundancy frame size.\n");
266 return AVERROR_INVALIDDATA;
267 }
268
269 if (redundancy_pos) {
270 ret = opus_decode_redundancy(s, data + size, redundancy_size);
271 if (ret < 0)
272 return ret;
273 ff_celt_flush(s->celt);
274 }
275 }
276
277 /* decode the CELT frame */
278 if (s->packet.mode == OPUS_MODE_CELT || s->packet.mode == OPUS_MODE_HYBRID) {
279 float *out_tmp[2] = { s->out[0], s->out[1] };
280 float **dst = (s->packet.mode == OPUS_MODE_CELT) ?
281 out_tmp : s->celt_output;
282 int celt_output_samples = samples;
283 int delay_samples = av_audio_fifo_size(s->celt_delay);
284
285 if (delay_samples) {
286 if (s->packet.mode == OPUS_MODE_HYBRID) {
287 av_audio_fifo_read(s->celt_delay, (void**)s->celt_output, delay_samples);
288
289 for (i = 0; i < s->output_channels; i++) {
290 s->fdsp->vector_fmac_scalar(out_tmp[i], s->celt_output[i], 1.0,
291 delay_samples);
292 out_tmp[i] += delay_samples;
293 }
294 celt_output_samples -= delay_samples;
295 } else {
296 av_log(s->avctx, AV_LOG_WARNING,
297 "Spurious CELT delay samples present.\n");
298 av_audio_fifo_drain(s->celt_delay, delay_samples);
299 if (s->avctx->err_recognition & AV_EF_EXPLODE)
300 return AVERROR_BUG;
301 }
302 }
303
304 opus_raw_init(&s->rc, data + size, size);
305
306 ret = ff_celt_decode_frame(s->celt, &s->rc, dst,
307 s->packet.stereo + 1,
308 s->packet.frame_duration,
309 (s->packet.mode == OPUS_MODE_HYBRID) ? 17 : 0,
310 celt_band_end[s->packet.bandwidth]);
311 if (ret < 0)
312 return ret;
313
314 if (s->packet.mode == OPUS_MODE_HYBRID) {
315 int celt_delay = s->packet.frame_duration - celt_output_samples;
316 void *delaybuf[2] = { s->celt_output[0] + celt_output_samples,
317 s->celt_output[1] + celt_output_samples };
318
319 for (i = 0; i < s->output_channels; i++) {
320 s->fdsp->vector_fmac_scalar(out_tmp[i],
321 s->celt_output[i], 1.0,
322 celt_output_samples);
323 }
324
325 ret = av_audio_fifo_write(s->celt_delay, delaybuf, celt_delay);
326 if (ret < 0)
327 return ret;
328 }
329 } else
330 ff_celt_flush(s->celt);
331
332 if (s->redundancy_idx) {
333 for (i = 0; i < s->output_channels; i++)
334 opus_fade(s->out[i], s->out[i],
335 s->redundancy_output[i] + 120 + s->redundancy_idx,
336 ff_celt_window2 + s->redundancy_idx, 120 - s->redundancy_idx);
337 s->redundancy_idx = 0;
338 }
339 if (redundancy) {
340 if (!redundancy_pos) {
341 ff_celt_flush(s->celt);
342 ret = opus_decode_redundancy(s, data + size, redundancy_size);
343 if (ret < 0)
344 return ret;
345
346 for (i = 0; i < s->output_channels; i++) {
347 opus_fade(s->out[i] + samples - 120 + delayed_samples,
348 s->out[i] + samples - 120 + delayed_samples,
349 s->redundancy_output[i] + 120,
350 ff_celt_window2, 120 - delayed_samples);
351 if (delayed_samples)
352 s->redundancy_idx = 120 - delayed_samples;
353 }
354 } else {
355 for (i = 0; i < s->output_channels; i++) {
356 memcpy(s->out[i] + delayed_samples, s->redundancy_output[i], 120 * sizeof(float));
357 opus_fade(s->out[i] + 120 + delayed_samples,
358 s->redundancy_output[i] + 120,
359 s->out[i] + 120 + delayed_samples,
360 ff_celt_window2, 120);
361 }
362 }
363 }
364
365 return samples;
366 }
367
368 static int opus_decode_subpacket(OpusStreamContext *s,
369 const uint8_t *buf, int buf_size,
370 int nb_samples)
371 {
372 int output_samples = 0;
373 int flush_needed = 0;
374 int i, j, ret;
375
376 /* check if we need to flush the resampler */
377 if (avresample_is_open(s->avr)) {
378 if (buf) {
379 int64_t cur_samplerate;
380 av_opt_get_int(s->avr, "in_sample_rate", 0, &cur_samplerate);
381 flush_needed = (s->packet.mode == OPUS_MODE_CELT) || (cur_samplerate != s->silk_samplerate);
382 } else {
383 flush_needed = !!s->delayed_samples;
384 }
385 }
386
387 if (!buf && !flush_needed)
388 return 0;
389
390 /* use dummy output buffers if the channel is not mapped to anything */
391 if (!s->out[0] ||
392 (s->output_channels == 2 && !s->out[1])) {
393 av_fast_malloc(&s->out_dummy, &s->out_dummy_allocated_size, s->out_size);
394 if (!s->out_dummy)
395 return AVERROR(ENOMEM);
396 if (!s->out[0])
397 s->out[0] = s->out_dummy;
398 if (!s->out[1])
399 s->out[1] = s->out_dummy;
400 }
401
402 /* flush the resampler if necessary */
403 if (flush_needed) {
404 ret = opus_flush_resample(s, s->delayed_samples);
405 if (ret < 0) {
406 av_log(s->avctx, AV_LOG_ERROR, "Error flushing the resampler.\n");
407 return ret;
408 }
409 avresample_close(s->avr);
410 output_samples += s->delayed_samples;
411 s->delayed_samples = 0;
412
413 if (!buf)
414 goto finish;
415 }
416
417 /* decode all the frames in the packet */
418 for (i = 0; i < s->packet.frame_count; i++) {
419 int size = s->packet.frame_size[i];
420 int samples = opus_decode_frame(s, buf + s->packet.frame_offset[i], size);
421
422 if (samples < 0) {
423 av_log(s->avctx, AV_LOG_ERROR, "Error decoding an Opus frame.\n");
424 if (s->avctx->err_recognition & AV_EF_EXPLODE)
425 return samples;
426
427 for (j = 0; j < s->output_channels; j++)
428 memset(s->out[j], 0, s->packet.frame_duration * sizeof(float));
429 samples = s->packet.frame_duration;
430 }
431 output_samples += samples;
432
433 for (j = 0; j < s->output_channels; j++)
434 s->out[j] += samples;
435 s->out_size -= samples * sizeof(float);
436 }
437
438 finish:
439 s->out[0] = s->out[1] = NULL;
440 s->out_size = 0;
441
442 return output_samples;
443 }
444
445 static int opus_decode_packet(AVCodecContext *avctx, void *data,
446 int *got_frame_ptr, AVPacket *avpkt)
447 {
448 OpusContext *c = avctx->priv_data;
449 AVFrame *frame = data;
450 const uint8_t *buf = avpkt->data;
451 int buf_size = avpkt->size;
452 int coded_samples = 0;
453 int decoded_samples = 0;
454 int i, ret;
455
456 /* decode the header of the first sub-packet to find out the sample count */
457 if (buf) {
458 OpusPacket *pkt = &c->streams[0].packet;
459 ret = ff_opus_parse_packet(pkt, buf, buf_size, c->nb_streams > 1);
460 if (ret < 0) {
461 av_log(avctx, AV_LOG_ERROR, "Error parsing the packet header.\n");
462 return ret;
463 }
464 coded_samples += pkt->frame_count * pkt->frame_duration;
465 c->streams[0].silk_samplerate = get_silk_samplerate(pkt->config);
466 }
467
468 frame->nb_samples = coded_samples + c->streams[0].delayed_samples;
469
470 /* no input or buffered data => nothing to do */
471 if (!frame->nb_samples) {
472 *got_frame_ptr = 0;
473 return 0;
474 }
475
476 /* setup the data buffers */
477 ret = ff_get_buffer(avctx, frame, 0);
478 if (ret < 0) {
479 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n");
480 return ret;
481 }
482 frame->nb_samples = 0;
483
484 for (i = 0; i < avctx->channels; i++) {
485 ChannelMap *map = &c->channel_maps[i];
486 if (!map->copy)
487 c->streams[map->stream_idx].out[map->channel_idx] = (float*)frame->extended_data[i];
488 }
489
490 for (i = 0; i < c->nb_streams; i++)
491 c->streams[i].out_size = frame->linesize[0];
492
493 /* decode each sub-packet */
494 for (i = 0; i < c->nb_streams; i++) {
495 OpusStreamContext *s = &c->streams[i];
496
497 if (i && buf) {
498 ret = ff_opus_parse_packet(&s->packet, buf, buf_size, i != c->nb_streams - 1);
499 if (ret < 0) {
500 av_log(avctx, AV_LOG_ERROR, "Error parsing the packet header.\n");
501 return ret;
502 }
503 s->silk_samplerate = get_silk_samplerate(s->packet.config);
504 }
505
506 ret = opus_decode_subpacket(&c->streams[i], buf,
507 s->packet.data_size, coded_samples);
508 if (ret < 0)
509 return ret;
510 if (decoded_samples && ret != decoded_samples) {
511 av_log(avctx, AV_LOG_ERROR, "Different numbers of decoded samples "
512 "in a multi-channel stream\n");
513 return AVERROR_INVALIDDATA;
514 }
515 decoded_samples = ret;
516 buf += s->packet.packet_size;
517 buf_size -= s->packet.packet_size;
518 }
519
520 for (i = 0; i < avctx->channels; i++) {
521 ChannelMap *map = &c->channel_maps[i];
522
523 /* handle copied channels */
524 if (map->copy) {
525 memcpy(frame->extended_data[i],
526 frame->extended_data[map->copy_idx],
527 frame->linesize[0]);
528 } else if (map->silence) {
529 memset(frame->extended_data[i], 0, frame->linesize[0]);
530 }
531
532 if (c->gain_i) {
533 c->fdsp.vector_fmul_scalar((float*)frame->extended_data[i],
534 (float*)frame->extended_data[i],
535 c->gain, FFALIGN(decoded_samples, 8));
536 }
537 }
538
539 frame->nb_samples = decoded_samples;
540 *got_frame_ptr = !!decoded_samples;
541
542 return avpkt->size;
543 }
544
545 static av_cold void opus_decode_flush(AVCodecContext *ctx)
546 {
547 OpusContext *c = ctx->priv_data;
548 int i;
549
550 for (i = 0; i < c->nb_streams; i++) {
551 OpusStreamContext *s = &c->streams[i];
552
553 memset(&s->packet, 0, sizeof(s->packet));
554 s->delayed_samples = 0;
555
556 if (s->celt_delay)
557 av_audio_fifo_drain(s->celt_delay, av_audio_fifo_size(s->celt_delay));
558 avresample_close(s->avr);
559
560 ff_silk_flush(s->silk);
561 ff_celt_flush(s->celt);
562 }
563 }
564
565 static av_cold int opus_decode_close(AVCodecContext *avctx)
566 {
567 OpusContext *c = avctx->priv_data;
568 int i;
569
570 for (i = 0; i < c->nb_streams; i++) {
571 OpusStreamContext *s = &c->streams[i];
572
573 ff_silk_free(&s->silk);
574 ff_celt_free(&s->celt);
575
576 av_freep(&s->out_dummy);
577 s->out_dummy_allocated_size = 0;
578
579 av_audio_fifo_free(s->celt_delay);
580 avresample_free(&s->avr);
581 }
582
583 av_freep(&c->streams);
584 c->nb_streams = 0;
585
586 av_freep(&c->channel_maps);
587
588 return 0;
589 }
590
591 static av_cold int opus_decode_init(AVCodecContext *avctx)
592 {
593 OpusContext *c = avctx->priv_data;
594 int ret, i, j;
595
596 avctx->sample_fmt = AV_SAMPLE_FMT_FLTP;
597 avctx->sample_rate = 48000;
598
599 avpriv_float_dsp_init(&c->fdsp, 0);
600
601 /* find out the channel configuration */
602 ret = ff_opus_parse_extradata(avctx, c);
603 if (ret < 0)
604 return ret;
605
606 /* allocate and init each independent decoder */
607 c->streams = av_mallocz_array(c->nb_streams, sizeof(*c->streams));
608 if (!c->streams) {
609 c->nb_streams = 0;
610 ret = AVERROR(ENOMEM);
611 goto fail;
612 }
613
614 for (i = 0; i < c->nb_streams; i++) {
615 OpusStreamContext *s = &c->streams[i];
616 uint64_t layout;
617
618 s->output_channels = (i < c->nb_stereo_streams) ? 2 : 1;
619
620 s->avctx = avctx;
621
622 for (j = 0; j < s->output_channels; j++) {
623 s->silk_output[j] = s->silk_buf[j];
624 s->celt_output[j] = s->celt_buf[j];
625 s->redundancy_output[j] = s->redundancy_buf[j];
626 }
627
628 s->fdsp = &c->fdsp;
629
630 s->avr = avresample_alloc_context();
631 if (!s->avr)
632 goto fail;
633
634 layout = (s->output_channels == 1) ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
635 av_opt_set_int(s->avr, "in_sample_fmt", avctx->sample_fmt, 0);
636 av_opt_set_int(s->avr, "out_sample_fmt", avctx->sample_fmt, 0);
637 av_opt_set_int(s->avr, "in_channel_layout", layout, 0);
638 av_opt_set_int(s->avr, "out_channel_layout", layout, 0);
639 av_opt_set_int(s->avr, "out_sample_rate", avctx->sample_rate, 0);
640
641 ret = ff_silk_init(avctx, &s->silk, s->output_channels);
642 if (ret < 0)
643 goto fail;
644
645 ret = ff_celt_init(avctx, &s->celt, s->output_channels);
646 if (ret < 0)
647 goto fail;
648
649 s->celt_delay = av_audio_fifo_alloc(avctx->sample_fmt,
650 s->output_channels, 1024);
651 if (!s->celt_delay) {
652 ret = AVERROR(ENOMEM);
653 goto fail;
654 }
655 }
656
657 return 0;
658 fail:
659 opus_decode_close(avctx);
660 return ret;
661 }
662
663 AVCodec ff_opus_decoder = {
664 .name = "opus",
665 .long_name = NULL_IF_CONFIG_SMALL("Opus"),
666 .type = AVMEDIA_TYPE_AUDIO,
667 .id = AV_CODEC_ID_OPUS,
668 .priv_data_size = sizeof(OpusContext),
669 .init = opus_decode_init,
670 .close = opus_decode_close,
671 .decode = opus_decode_packet,
672 .flush = opus_decode_flush,
673 .capabilities = CODEC_CAP_DR1 | CODEC_CAP_DELAY,
674 };