xref: /third_party/ffmpeg/libavcodec/opusdec.c (revision cabdff1a)
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 FFmpeg.
7 *
8 * FFmpeg 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 * FFmpeg 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 FFmpeg; 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 "libswresample/swresample.h"
44
45#include "avcodec.h"
46#include "codec_internal.h"
47#include "get_bits.h"
48#include "internal.h"
49#include "mathops.h"
50#include "opus.h"
51#include "opustab.h"
52#include "opus_celt.h"
53
54static const uint16_t silk_frame_duration_ms[16] = {
55    10, 20, 40, 60,
56    10, 20, 40, 60,
57    10, 20, 40, 60,
58    10, 20,
59    10, 20,
60};
61
62/* number of samples of silence to feed to the resampler
63 * at the beginning */
64static const int silk_resample_delay[] = {
65    4, 8, 11, 11, 11
66};
67
68static int get_silk_samplerate(int config)
69{
70    if (config < 4)
71        return 8000;
72    else if (config < 8)
73        return 12000;
74    return 16000;
75}
76
77static void opus_fade(float *out,
78                      const float *in1, const float *in2,
79                      const float *window, int len)
80{
81    int i;
82    for (i = 0; i < len; i++)
83        out[i] = in2[i] * window[i] + in1[i] * (1.0 - window[i]);
84}
85
86static int opus_flush_resample(OpusStreamContext *s, int nb_samples)
87{
88    int celt_size = av_audio_fifo_size(s->celt_delay);
89    int ret, i;
90    ret = swr_convert(s->swr,
91                      (uint8_t**)s->cur_out, nb_samples,
92                      NULL, 0);
93    if (ret < 0)
94        return ret;
95    else if (ret != nb_samples) {
96        av_log(s->avctx, AV_LOG_ERROR, "Wrong number of flushed samples: %d\n",
97               ret);
98        return AVERROR_BUG;
99    }
100
101    if (celt_size) {
102        if (celt_size != nb_samples) {
103            av_log(s->avctx, AV_LOG_ERROR, "Wrong number of CELT delay samples.\n");
104            return AVERROR_BUG;
105        }
106        av_audio_fifo_read(s->celt_delay, (void**)s->celt_output, nb_samples);
107        for (i = 0; i < s->output_channels; i++) {
108            s->fdsp->vector_fmac_scalar(s->cur_out[i],
109                                        s->celt_output[i], 1.0,
110                                        nb_samples);
111        }
112    }
113
114    if (s->redundancy_idx) {
115        for (i = 0; i < s->output_channels; i++)
116            opus_fade(s->cur_out[i], s->cur_out[i],
117                      s->redundancy_output[i] + 120 + s->redundancy_idx,
118                      ff_celt_window2 + s->redundancy_idx, 120 - s->redundancy_idx);
119        s->redundancy_idx = 0;
120    }
121
122    s->cur_out[0]         += nb_samples;
123    s->cur_out[1]         += nb_samples;
124    s->remaining_out_size -= nb_samples * sizeof(float);
125
126    return 0;
127}
128
129static int opus_init_resample(OpusStreamContext *s)
130{
131    static const float delay[16] = { 0.0 };
132    const uint8_t *delayptr[2] = { (uint8_t*)delay, (uint8_t*)delay };
133    int ret;
134
135    av_opt_set_int(s->swr, "in_sample_rate", s->silk_samplerate, 0);
136    ret = swr_init(s->swr);
137    if (ret < 0) {
138        av_log(s->avctx, AV_LOG_ERROR, "Error opening the resampler.\n");
139        return ret;
140    }
141
142    ret = swr_convert(s->swr,
143                      NULL, 0,
144                      delayptr, silk_resample_delay[s->packet.bandwidth]);
145    if (ret < 0) {
146        av_log(s->avctx, AV_LOG_ERROR,
147               "Error feeding initial silence to the resampler.\n");
148        return ret;
149    }
150
151    return 0;
152}
153
154static int opus_decode_redundancy(OpusStreamContext *s, const uint8_t *data, int size)
155{
156    int ret = ff_opus_rc_dec_init(&s->redundancy_rc, data, size);
157    if (ret < 0)
158        goto fail;
159    ff_opus_rc_dec_raw_init(&s->redundancy_rc, data + size, size);
160
161    ret = ff_celt_decode_frame(s->celt, &s->redundancy_rc,
162                               s->redundancy_output,
163                               s->packet.stereo + 1, 240,
164                               0, ff_celt_band_end[s->packet.bandwidth]);
165    if (ret < 0)
166        goto fail;
167
168    return 0;
169fail:
170    av_log(s->avctx, AV_LOG_ERROR, "Error decoding the redundancy frame.\n");
171    return ret;
172}
173
174static int opus_decode_frame(OpusStreamContext *s, const uint8_t *data, int size)
175{
176    int samples    = s->packet.frame_duration;
177    int redundancy = 0;
178    int redundancy_size, redundancy_pos;
179    int ret, i, consumed;
180    int delayed_samples = s->delayed_samples;
181
182    ret = ff_opus_rc_dec_init(&s->rc, data, size);
183    if (ret < 0)
184        return ret;
185
186    /* decode the silk frame */
187    if (s->packet.mode == OPUS_MODE_SILK || s->packet.mode == OPUS_MODE_HYBRID) {
188        if (!swr_is_initialized(s->swr)) {
189            ret = opus_init_resample(s);
190            if (ret < 0)
191                return ret;
192        }
193
194        samples = ff_silk_decode_superframe(s->silk, &s->rc, s->silk_output,
195                                            FFMIN(s->packet.bandwidth, OPUS_BANDWIDTH_WIDEBAND),
196                                            s->packet.stereo + 1,
197                                            silk_frame_duration_ms[s->packet.config]);
198        if (samples < 0) {
199            av_log(s->avctx, AV_LOG_ERROR, "Error decoding a SILK frame.\n");
200            return samples;
201        }
202        samples = swr_convert(s->swr,
203                              (uint8_t**)s->cur_out, s->packet.frame_duration,
204                              (const uint8_t**)s->silk_output, samples);
205        if (samples < 0) {
206            av_log(s->avctx, AV_LOG_ERROR, "Error resampling SILK data.\n");
207            return samples;
208        }
209        av_assert2((samples & 7) == 0);
210        s->delayed_samples += s->packet.frame_duration - samples;
211    } else
212        ff_silk_flush(s->silk);
213
214    // decode redundancy information
215    consumed = opus_rc_tell(&s->rc);
216    if (s->packet.mode == OPUS_MODE_HYBRID && consumed + 37 <= size * 8)
217        redundancy = ff_opus_rc_dec_log(&s->rc, 12);
218    else if (s->packet.mode == OPUS_MODE_SILK && consumed + 17 <= size * 8)
219        redundancy = 1;
220
221    if (redundancy) {
222        redundancy_pos = ff_opus_rc_dec_log(&s->rc, 1);
223
224        if (s->packet.mode == OPUS_MODE_HYBRID)
225            redundancy_size = ff_opus_rc_dec_uint(&s->rc, 256) + 2;
226        else
227            redundancy_size = size - (consumed + 7) / 8;
228        size -= redundancy_size;
229        if (size < 0) {
230            av_log(s->avctx, AV_LOG_ERROR, "Invalid redundancy frame size.\n");
231            return AVERROR_INVALIDDATA;
232        }
233
234        if (redundancy_pos) {
235            ret = opus_decode_redundancy(s, data + size, redundancy_size);
236            if (ret < 0)
237                return ret;
238            ff_celt_flush(s->celt);
239        }
240    }
241
242    /* decode the CELT frame */
243    if (s->packet.mode == OPUS_MODE_CELT || s->packet.mode == OPUS_MODE_HYBRID) {
244        float *out_tmp[2] = { s->cur_out[0], s->cur_out[1] };
245        float **dst = (s->packet.mode == OPUS_MODE_CELT) ?
246                      out_tmp : s->celt_output;
247        int celt_output_samples = samples;
248        int delay_samples = av_audio_fifo_size(s->celt_delay);
249
250        if (delay_samples) {
251            if (s->packet.mode == OPUS_MODE_HYBRID) {
252                av_audio_fifo_read(s->celt_delay, (void**)s->celt_output, delay_samples);
253
254                for (i = 0; i < s->output_channels; i++) {
255                    s->fdsp->vector_fmac_scalar(out_tmp[i], s->celt_output[i], 1.0,
256                                                delay_samples);
257                    out_tmp[i] += delay_samples;
258                }
259                celt_output_samples -= delay_samples;
260            } else {
261                av_log(s->avctx, AV_LOG_WARNING,
262                       "Spurious CELT delay samples present.\n");
263                av_audio_fifo_drain(s->celt_delay, delay_samples);
264                if (s->avctx->err_recognition & AV_EF_EXPLODE)
265                    return AVERROR_BUG;
266            }
267        }
268
269        ff_opus_rc_dec_raw_init(&s->rc, data + size, size);
270
271        ret = ff_celt_decode_frame(s->celt, &s->rc, dst,
272                                   s->packet.stereo + 1,
273                                   s->packet.frame_duration,
274                                   (s->packet.mode == OPUS_MODE_HYBRID) ? 17 : 0,
275                                   ff_celt_band_end[s->packet.bandwidth]);
276        if (ret < 0)
277            return ret;
278
279        if (s->packet.mode == OPUS_MODE_HYBRID) {
280            int celt_delay = s->packet.frame_duration - celt_output_samples;
281            void *delaybuf[2] = { s->celt_output[0] + celt_output_samples,
282                                  s->celt_output[1] + celt_output_samples };
283
284            for (i = 0; i < s->output_channels; i++) {
285                s->fdsp->vector_fmac_scalar(out_tmp[i],
286                                            s->celt_output[i], 1.0,
287                                            celt_output_samples);
288            }
289
290            ret = av_audio_fifo_write(s->celt_delay, delaybuf, celt_delay);
291            if (ret < 0)
292                return ret;
293        }
294    } else
295        ff_celt_flush(s->celt);
296
297    if (s->redundancy_idx) {
298        for (i = 0; i < s->output_channels; i++)
299            opus_fade(s->cur_out[i], s->cur_out[i],
300                      s->redundancy_output[i] + 120 + s->redundancy_idx,
301                      ff_celt_window2 + s->redundancy_idx, 120 - s->redundancy_idx);
302        s->redundancy_idx = 0;
303    }
304    if (redundancy) {
305        if (!redundancy_pos) {
306            ff_celt_flush(s->celt);
307            ret = opus_decode_redundancy(s, data + size, redundancy_size);
308            if (ret < 0)
309                return ret;
310
311            for (i = 0; i < s->output_channels; i++) {
312                opus_fade(s->cur_out[i] + samples - 120 + delayed_samples,
313                          s->cur_out[i] + samples - 120 + delayed_samples,
314                          s->redundancy_output[i] + 120,
315                          ff_celt_window2, 120 - delayed_samples);
316                if (delayed_samples)
317                    s->redundancy_idx = 120 - delayed_samples;
318            }
319        } else {
320            for (i = 0; i < s->output_channels; i++) {
321                memcpy(s->cur_out[i] + delayed_samples, s->redundancy_output[i], 120 * sizeof(float));
322                opus_fade(s->cur_out[i] + 120 + delayed_samples,
323                          s->redundancy_output[i] + 120,
324                          s->cur_out[i] + 120 + delayed_samples,
325                          ff_celt_window2, 120);
326            }
327        }
328    }
329
330    return samples;
331}
332
333static int opus_decode_subpacket(OpusStreamContext *s,
334                                 const uint8_t *buf, int buf_size,
335                                 int nb_samples)
336{
337    int output_samples = 0;
338    int flush_needed   = 0;
339    int i, j, ret;
340
341    s->cur_out[0]         = s->out[0];
342    s->cur_out[1]         = s->out[1];
343    s->remaining_out_size = s->out_size;
344
345    /* check if we need to flush the resampler */
346    if (swr_is_initialized(s->swr)) {
347        if (buf) {
348            int64_t cur_samplerate;
349            av_opt_get_int(s->swr, "in_sample_rate", 0, &cur_samplerate);
350            flush_needed = (s->packet.mode == OPUS_MODE_CELT) || (cur_samplerate != s->silk_samplerate);
351        } else {
352            flush_needed = !!s->delayed_samples;
353        }
354    }
355
356    if (!buf && !flush_needed)
357        return 0;
358
359    /* use dummy output buffers if the channel is not mapped to anything */
360    if (!s->cur_out[0] ||
361        (s->output_channels == 2 && !s->cur_out[1])) {
362        av_fast_malloc(&s->out_dummy, &s->out_dummy_allocated_size,
363                       s->remaining_out_size);
364        if (!s->out_dummy)
365            return AVERROR(ENOMEM);
366        if (!s->cur_out[0])
367            s->cur_out[0] = s->out_dummy;
368        if (!s->cur_out[1])
369            s->cur_out[1] = s->out_dummy;
370    }
371
372    /* flush the resampler if necessary */
373    if (flush_needed) {
374        ret = opus_flush_resample(s, s->delayed_samples);
375        if (ret < 0) {
376            av_log(s->avctx, AV_LOG_ERROR, "Error flushing the resampler.\n");
377            return ret;
378        }
379        swr_close(s->swr);
380        output_samples += s->delayed_samples;
381        s->delayed_samples = 0;
382
383        if (!buf)
384            goto finish;
385    }
386
387    /* decode all the frames in the packet */
388    for (i = 0; i < s->packet.frame_count; i++) {
389        int size = s->packet.frame_size[i];
390        int samples = opus_decode_frame(s, buf + s->packet.frame_offset[i], size);
391
392        if (samples < 0) {
393            av_log(s->avctx, AV_LOG_ERROR, "Error decoding an Opus frame.\n");
394            if (s->avctx->err_recognition & AV_EF_EXPLODE)
395                return samples;
396
397            for (j = 0; j < s->output_channels; j++)
398                memset(s->cur_out[j], 0, s->packet.frame_duration * sizeof(float));
399            samples = s->packet.frame_duration;
400        }
401        output_samples += samples;
402
403        for (j = 0; j < s->output_channels; j++)
404            s->cur_out[j] += samples;
405        s->remaining_out_size -= samples * sizeof(float);
406    }
407
408finish:
409    s->cur_out[0] = s->cur_out[1] = NULL;
410    s->remaining_out_size = 0;
411
412    return output_samples;
413}
414
415static int opus_decode_packet(AVCodecContext *avctx, AVFrame *frame,
416                              int *got_frame_ptr, AVPacket *avpkt)
417{
418    OpusContext *c      = avctx->priv_data;
419    const uint8_t *buf  = avpkt->data;
420    int buf_size        = avpkt->size;
421    int coded_samples   = 0;
422    int decoded_samples = INT_MAX;
423    int delayed_samples = 0;
424    int i, ret;
425
426    /* calculate the number of delayed samples */
427    for (i = 0; i < c->nb_streams; i++) {
428        OpusStreamContext *s = &c->streams[i];
429        s->out[0] =
430        s->out[1] = NULL;
431        delayed_samples = FFMAX(delayed_samples,
432                                s->delayed_samples + av_audio_fifo_size(s->sync_buffer));
433    }
434
435    /* decode the header of the first sub-packet to find out the sample count */
436    if (buf) {
437        OpusPacket *pkt = &c->streams[0].packet;
438        ret = ff_opus_parse_packet(pkt, buf, buf_size, c->nb_streams > 1);
439        if (ret < 0) {
440            av_log(avctx, AV_LOG_ERROR, "Error parsing the packet header.\n");
441            return ret;
442        }
443        coded_samples += pkt->frame_count * pkt->frame_duration;
444        c->streams[0].silk_samplerate = get_silk_samplerate(pkt->config);
445    }
446
447    frame->nb_samples = coded_samples + delayed_samples;
448
449    /* no input or buffered data => nothing to do */
450    if (!frame->nb_samples) {
451        *got_frame_ptr = 0;
452        return 0;
453    }
454
455    /* setup the data buffers */
456    ret = ff_get_buffer(avctx, frame, 0);
457    if (ret < 0)
458        return ret;
459    frame->nb_samples = 0;
460
461    for (i = 0; i < avctx->ch_layout.nb_channels; i++) {
462        ChannelMap *map = &c->channel_maps[i];
463        if (!map->copy)
464            c->streams[map->stream_idx].out[map->channel_idx] = (float*)frame->extended_data[i];
465    }
466
467    /* read the data from the sync buffers */
468    for (i = 0; i < c->nb_streams; i++) {
469        OpusStreamContext *s = &c->streams[i];
470        float          **out = s->out;
471        int sync_size = av_audio_fifo_size(s->sync_buffer);
472
473        float sync_dummy[32];
474        int out_dummy = (!out[0]) | ((!out[1]) << 1);
475
476        if (!out[0])
477            out[0] = sync_dummy;
478        if (!out[1])
479            out[1] = sync_dummy;
480        if (out_dummy && sync_size > FF_ARRAY_ELEMS(sync_dummy))
481            return AVERROR_BUG;
482
483        ret = av_audio_fifo_read(s->sync_buffer, (void**)out, sync_size);
484        if (ret < 0)
485            return ret;
486
487        if (out_dummy & 1)
488            out[0] = NULL;
489        else
490            out[0] += ret;
491        if (out_dummy & 2)
492            out[1] = NULL;
493        else
494            out[1] += ret;
495
496        s->out_size = frame->linesize[0] - ret * sizeof(float);
497    }
498
499    /* decode each sub-packet */
500    for (i = 0; i < c->nb_streams; i++) {
501        OpusStreamContext *s = &c->streams[i];
502
503        if (i && buf) {
504            ret = ff_opus_parse_packet(&s->packet, buf, buf_size, i != c->nb_streams - 1);
505            if (ret < 0) {
506                av_log(avctx, AV_LOG_ERROR, "Error parsing the packet header.\n");
507                return ret;
508            }
509            if (coded_samples != s->packet.frame_count * s->packet.frame_duration) {
510                av_log(avctx, AV_LOG_ERROR,
511                       "Mismatching coded sample count in substream %d.\n", i);
512                return AVERROR_INVALIDDATA;
513            }
514
515            s->silk_samplerate = get_silk_samplerate(s->packet.config);
516        }
517
518        ret = opus_decode_subpacket(&c->streams[i], buf, s->packet.data_size,
519                                    coded_samples);
520        if (ret < 0)
521            return ret;
522        s->decoded_samples = ret;
523        decoded_samples       = FFMIN(decoded_samples, ret);
524
525        buf      += s->packet.packet_size;
526        buf_size -= s->packet.packet_size;
527    }
528
529    /* buffer the extra samples */
530    for (i = 0; i < c->nb_streams; i++) {
531        OpusStreamContext *s = &c->streams[i];
532        int   buffer_samples = s->decoded_samples - decoded_samples;
533        if (buffer_samples) {
534            float *buf[2] = { s->out[0] ? s->out[0] : (float*)frame->extended_data[0],
535                              s->out[1] ? s->out[1] : (float*)frame->extended_data[0] };
536            buf[0] += decoded_samples;
537            buf[1] += decoded_samples;
538            ret = av_audio_fifo_write(s->sync_buffer, (void**)buf, buffer_samples);
539            if (ret < 0)
540                return ret;
541        }
542    }
543
544    for (i = 0; i < avctx->ch_layout.nb_channels; i++) {
545        ChannelMap *map = &c->channel_maps[i];
546
547        /* handle copied channels */
548        if (map->copy) {
549            memcpy(frame->extended_data[i],
550                   frame->extended_data[map->copy_idx],
551                   frame->linesize[0]);
552        } else if (map->silence) {
553            memset(frame->extended_data[i], 0, frame->linesize[0]);
554        }
555
556        if (c->gain_i && decoded_samples > 0) {
557            c->fdsp->vector_fmul_scalar((float*)frame->extended_data[i],
558                                       (float*)frame->extended_data[i],
559                                       c->gain, FFALIGN(decoded_samples, 8));
560        }
561    }
562
563    frame->nb_samples = decoded_samples;
564    *got_frame_ptr    = !!decoded_samples;
565
566    return avpkt->size;
567}
568
569static av_cold void opus_decode_flush(AVCodecContext *ctx)
570{
571    OpusContext *c = ctx->priv_data;
572    int i;
573
574    for (i = 0; i < c->nb_streams; i++) {
575        OpusStreamContext *s = &c->streams[i];
576
577        memset(&s->packet, 0, sizeof(s->packet));
578        s->delayed_samples = 0;
579
580        av_audio_fifo_drain(s->celt_delay, av_audio_fifo_size(s->celt_delay));
581        swr_close(s->swr);
582
583        av_audio_fifo_drain(s->sync_buffer, av_audio_fifo_size(s->sync_buffer));
584
585        ff_silk_flush(s->silk);
586        ff_celt_flush(s->celt);
587    }
588}
589
590static av_cold int opus_decode_close(AVCodecContext *avctx)
591{
592    OpusContext *c = avctx->priv_data;
593    int i;
594
595    for (i = 0; i < c->nb_streams; i++) {
596        OpusStreamContext *s = &c->streams[i];
597
598        ff_silk_free(&s->silk);
599        ff_celt_free(&s->celt);
600
601        av_freep(&s->out_dummy);
602        s->out_dummy_allocated_size = 0;
603
604        av_audio_fifo_free(s->sync_buffer);
605        av_audio_fifo_free(s->celt_delay);
606        swr_free(&s->swr);
607    }
608
609    av_freep(&c->streams);
610
611    c->nb_streams = 0;
612
613    av_freep(&c->channel_maps);
614    av_freep(&c->fdsp);
615
616    return 0;
617}
618
619static av_cold int opus_decode_init(AVCodecContext *avctx)
620{
621    OpusContext *c = avctx->priv_data;
622    int ret, i, j;
623
624    avctx->sample_fmt  = AV_SAMPLE_FMT_FLTP;
625    avctx->sample_rate = 48000;
626
627    c->fdsp = avpriv_float_dsp_alloc(0);
628    if (!c->fdsp)
629        return AVERROR(ENOMEM);
630
631    /* find out the channel configuration */
632    ret = ff_opus_parse_extradata(avctx, c);
633    if (ret < 0)
634        return ret;
635
636    /* allocate and init each independent decoder */
637    c->streams = av_calloc(c->nb_streams, sizeof(*c->streams));
638    if (!c->streams) {
639        c->nb_streams = 0;
640        return AVERROR(ENOMEM);
641    }
642
643    for (i = 0; i < c->nb_streams; i++) {
644        OpusStreamContext *s = &c->streams[i];
645        uint64_t layout;
646
647        s->output_channels = (i < c->nb_stereo_streams) ? 2 : 1;
648
649        s->avctx = avctx;
650
651        for (j = 0; j < s->output_channels; j++) {
652            s->silk_output[j]       = s->silk_buf[j];
653            s->celt_output[j]       = s->celt_buf[j];
654            s->redundancy_output[j] = s->redundancy_buf[j];
655        }
656
657        s->fdsp = c->fdsp;
658
659        s->swr =swr_alloc();
660        if (!s->swr)
661            return AVERROR(ENOMEM);
662
663        layout = (s->output_channels == 1) ? AV_CH_LAYOUT_MONO : AV_CH_LAYOUT_STEREO;
664        av_opt_set_int(s->swr, "in_sample_fmt",      avctx->sample_fmt,  0);
665        av_opt_set_int(s->swr, "out_sample_fmt",     avctx->sample_fmt,  0);
666        av_opt_set_int(s->swr, "in_channel_layout",  layout,             0);
667        av_opt_set_int(s->swr, "out_channel_layout", layout,             0);
668        av_opt_set_int(s->swr, "out_sample_rate",    avctx->sample_rate, 0);
669        av_opt_set_int(s->swr, "filter_size",        16,                 0);
670
671        ret = ff_silk_init(avctx, &s->silk, s->output_channels);
672        if (ret < 0)
673            return ret;
674
675        ret = ff_celt_init(avctx, &s->celt, s->output_channels, c->apply_phase_inv);
676        if (ret < 0)
677            return ret;
678
679        s->celt_delay = av_audio_fifo_alloc(avctx->sample_fmt,
680                                            s->output_channels, 1024);
681        if (!s->celt_delay)
682            return AVERROR(ENOMEM);
683
684        s->sync_buffer = av_audio_fifo_alloc(avctx->sample_fmt,
685                                             s->output_channels, 32);
686        if (!s->sync_buffer)
687            return AVERROR(ENOMEM);
688    }
689
690    return 0;
691}
692
693#define OFFSET(x) offsetof(OpusContext, x)
694#define AD AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_DECODING_PARAM
695static const AVOption opus_options[] = {
696    { "apply_phase_inv", "Apply intensity stereo phase inversion", OFFSET(apply_phase_inv), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, AD },
697    { NULL },
698};
699
700static const AVClass opus_class = {
701    .class_name = "Opus Decoder",
702    .item_name  = av_default_item_name,
703    .option     = opus_options,
704    .version    = LIBAVUTIL_VERSION_INT,
705};
706
707const FFCodec ff_opus_decoder = {
708    .p.name          = "opus",
709    .p.long_name     = NULL_IF_CONFIG_SMALL("Opus"),
710    .p.priv_class    = &opus_class,
711    .p.type          = AVMEDIA_TYPE_AUDIO,
712    .p.id            = AV_CODEC_ID_OPUS,
713    .priv_data_size  = sizeof(OpusContext),
714    .init            = opus_decode_init,
715    .close           = opus_decode_close,
716    FF_CODEC_DECODE_CB(opus_decode_packet),
717    .flush           = opus_decode_flush,
718    .p.capabilities  = AV_CODEC_CAP_DR1 | AV_CODEC_CAP_DELAY | AV_CODEC_CAP_CHANNEL_CONF,
719    .caps_internal   = FF_CODEC_CAP_INIT_THREADSAFE | FF_CODEC_CAP_INIT_CLEANUP,
720};
721