2 * Opus encoder using libopus
3 * Copyright (c) 2012 Nathan Caldwell
5 * This file is part of FFmpeg.
7 * FFmpeg is free software; you can redistribute it and/or
8 * modify it under the terms of the GNU Lesser General Public
9 * License as published by the Free Software Foundation; either
10 * version 2.1 of the License, or (at your option) any later version.
12 * FFmpeg is distributed in the hope that it will be useful,
13 * but WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * Lesser General Public License for more details.
17 * You should have received a copy of the GNU Lesser General Public
18 * License along with FFmpeg; if not, write to the Free Software
19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA
23 #include <opus_multistream.h>
25 #include "libavutil/opt.h"
27 #include "bytestream.h"
31 #include "audio_frame_queue.h"
33 typedef struct LibopusEncOpts {
43 #ifdef OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST
48 typedef struct LibopusEncContext {
55 const uint8_t *encoder_channel_map;
58 static const uint8_t opus_coupled_streams[8] = {
59 0, 1, 1, 2, 2, 2, 2, 3
62 /* Opus internal to Vorbis channel order mapping written in the header */
63 static const uint8_t opus_vorbis_channel_map[8][8] = {
70 { 0, 4, 1, 2, 3, 5, 6 },
71 { 0, 6, 1, 2, 3, 4, 5, 7 },
74 /* libavcodec to libopus channel order mapping, passed to libopus */
75 static const uint8_t libavcodec_libopus_channel_map[8][8] = {
82 { 0, 1, 5, 6, 2, 4, 3 },
83 { 0, 1, 6, 7, 4, 5, 2, 3 },
86 static void libopus_write_header(AVCodecContext *avctx, int stream_count,
87 int coupled_stream_count,
89 const uint8_t *channel_mapping)
91 uint8_t *p = avctx->extradata;
92 int channels = avctx->channels;
94 bytestream_put_buffer(&p, "OpusHead", 8);
95 bytestream_put_byte(&p, 1); /* Version */
96 bytestream_put_byte(&p, channels);
97 bytestream_put_le16(&p, avctx->initial_padding * 48000 / avctx->sample_rate); /* Lookahead samples at 48kHz */
98 bytestream_put_le32(&p, avctx->sample_rate); /* Original sample rate */
99 bytestream_put_le16(&p, 0); /* Gain of 0dB is recommended. */
101 /* Channel mapping */
102 bytestream_put_byte(&p, mapping_family);
103 if (mapping_family != 0) {
104 bytestream_put_byte(&p, stream_count);
105 bytestream_put_byte(&p, coupled_stream_count);
106 bytestream_put_buffer(&p, channel_mapping, channels);
110 static int libopus_configure_encoder(AVCodecContext *avctx, OpusMSEncoder *enc,
111 LibopusEncOpts *opts)
115 if (avctx->global_quality) {
116 av_log(avctx, AV_LOG_ERROR,
117 "Quality-based encoding not supported, "
118 "please specify a bitrate and VBR setting.\n");
119 return AVERROR(EINVAL);
122 ret = opus_multistream_encoder_ctl(enc, OPUS_SET_BITRATE(avctx->bit_rate));
123 if (ret != OPUS_OK) {
124 av_log(avctx, AV_LOG_ERROR,
125 "Failed to set bitrate: %s\n", opus_strerror(ret));
129 ret = opus_multistream_encoder_ctl(enc,
130 OPUS_SET_COMPLEXITY(opts->complexity));
132 av_log(avctx, AV_LOG_WARNING,
133 "Unable to set complexity: %s\n", opus_strerror(ret));
135 ret = opus_multistream_encoder_ctl(enc, OPUS_SET_VBR(!!opts->vbr));
137 av_log(avctx, AV_LOG_WARNING,
138 "Unable to set VBR: %s\n", opus_strerror(ret));
140 ret = opus_multistream_encoder_ctl(enc,
141 OPUS_SET_VBR_CONSTRAINT(opts->vbr == 2));
143 av_log(avctx, AV_LOG_WARNING,
144 "Unable to set constrained VBR: %s\n", opus_strerror(ret));
146 ret = opus_multistream_encoder_ctl(enc,
147 OPUS_SET_PACKET_LOSS_PERC(opts->packet_loss));
149 av_log(avctx, AV_LOG_WARNING,
150 "Unable to set expected packet loss percentage: %s\n",
153 ret = opus_multistream_encoder_ctl(enc,
154 OPUS_SET_INBAND_FEC(opts->fec));
156 av_log(avctx, AV_LOG_WARNING,
157 "Unable to set inband FEC: %s\n",
161 ret = opus_multistream_encoder_ctl(enc,
162 OPUS_SET_MAX_BANDWIDTH(opts->max_bandwidth));
164 av_log(avctx, AV_LOG_WARNING,
165 "Unable to set maximum bandwidth: %s\n", opus_strerror(ret));
168 #ifdef OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST
169 ret = opus_multistream_encoder_ctl(enc,
170 OPUS_SET_PHASE_INVERSION_DISABLED(!opts->apply_phase_inv));
172 av_log(avctx, AV_LOG_WARNING,
173 "Unable to set phase inversion: %s\n",
179 static int libopus_check_max_channels(AVCodecContext *avctx,
181 if (avctx->channels > max_channels) {
182 av_log(avctx, AV_LOG_ERROR, "Opus mapping family undefined for %d channels.\n",
184 return AVERROR(EINVAL);
190 static int libopus_check_vorbis_layout(AVCodecContext *avctx, int mapping_family) {
191 av_assert2(avctx->channels < FF_ARRAY_ELEMS(ff_vorbis_channel_layouts));
193 if (!avctx->channel_layout) {
194 av_log(avctx, AV_LOG_WARNING,
195 "No channel layout specified. Opus encoder will use Vorbis "
196 "channel layout for %d channels.\n", avctx->channels);
197 } else if (avctx->channel_layout != ff_vorbis_channel_layouts[avctx->channels - 1]) {
199 av_get_channel_layout_string(name, sizeof(name), avctx->channels,
200 avctx->channel_layout);
201 av_log(avctx, AV_LOG_ERROR,
202 "Invalid channel layout %s for specified mapping family %d.\n",
203 name, mapping_family);
205 return AVERROR(EINVAL);
211 static int libopus_validate_layout_and_get_channel_map(
212 AVCodecContext *avctx,
214 const uint8_t ** channel_map_result)
216 const uint8_t * channel_map = NULL;
219 switch (mapping_family) {
221 ret = libopus_check_max_channels(avctx, 8);
223 ret = libopus_check_vorbis_layout(avctx, mapping_family);
224 /* Channels do not need to be reordered. */
229 ret = libopus_check_max_channels(avctx, 2);
231 ret = libopus_check_vorbis_layout(avctx, mapping_family);
235 /* Opus expects channels to be in Vorbis order. */
236 ret = libopus_check_max_channels(avctx, 8);
238 ret = libopus_check_vorbis_layout(avctx, mapping_family);
239 channel_map = ff_vorbis_channel_layout_offsets[avctx->channels - 1];
243 ret = libopus_check_max_channels(avctx, 254);
246 av_log(avctx, AV_LOG_WARNING,
247 "Unknown channel mapping family %d. Output channel layout may be invalid.\n",
252 *channel_map_result = channel_map;
256 static av_cold int libopus_encode_init(AVCodecContext *avctx)
258 LibopusEncContext *opus = avctx->priv_data;
260 uint8_t libopus_channel_mapping[255];
263 int coupled_stream_count, header_size, frame_size;
266 frame_size = opus->opts.frame_duration * 48000 / 1000;
267 switch (frame_size) {
270 if (opus->opts.application != OPUS_APPLICATION_RESTRICTED_LOWDELAY)
271 av_log(avctx, AV_LOG_WARNING,
272 "LPC mode cannot be used with a frame duration of less "
273 "than 10ms. Enabling restricted low-delay mode.\n"
274 "Use a longer frame duration if this is not what you want.\n");
275 /* Frame sizes less than 10 ms can only use MDCT mode, so switching to
276 * RESTRICTED_LOWDELAY avoids an unnecessary extra 2.5ms lookahead. */
277 opus->opts.application = OPUS_APPLICATION_RESTRICTED_LOWDELAY;
282 #ifdef OPUS_FRAMESIZE_120_MS
287 opus->opts.packet_size =
288 avctx->frame_size = frame_size * avctx->sample_rate / 48000;
291 av_log(avctx, AV_LOG_ERROR, "Invalid frame duration: %g.\n"
292 "Frame duration must be exactly one of: 2.5, 5, 10, 20, 40"
293 #ifdef OPUS_FRAMESIZE_120_MS
294 ", 60, 80, 100 or 120.\n",
298 opus->opts.frame_duration);
299 return AVERROR(EINVAL);
302 if (avctx->compression_level < 0 || avctx->compression_level > 10) {
303 av_log(avctx, AV_LOG_WARNING,
304 "Compression level must be in the range 0 to 10. "
305 "Defaulting to 10.\n");
306 opus->opts.complexity = 10;
308 opus->opts.complexity = avctx->compression_level;
312 switch (avctx->cutoff) {
314 opus->opts.max_bandwidth = OPUS_BANDWIDTH_NARROWBAND;
317 opus->opts.max_bandwidth = OPUS_BANDWIDTH_MEDIUMBAND;
320 opus->opts.max_bandwidth = OPUS_BANDWIDTH_WIDEBAND;
323 opus->opts.max_bandwidth = OPUS_BANDWIDTH_SUPERWIDEBAND;
326 opus->opts.max_bandwidth = OPUS_BANDWIDTH_FULLBAND;
329 av_log(avctx, AV_LOG_WARNING,
330 "Invalid frequency cutoff: %d. Using default maximum bandwidth.\n"
331 "Cutoff frequency must be exactly one of: 4000, 6000, 8000, 12000 or 20000.\n",
337 /* Channels may need to be reordered to match opus mapping. */
338 av_ret = libopus_validate_layout_and_get_channel_map(avctx, opus->opts.mapping_family,
339 &opus->encoder_channel_map);
344 if (opus->opts.mapping_family == -1) {
345 /* By default, use mapping family 1 for the header but use the older
346 * libopus multistream API to avoid surround masking. */
348 /* Set the mapping family so that the value is correct in the header */
349 mapping_family = avctx->channels > 2 ? 1 : 0;
350 coupled_stream_count = opus_coupled_streams[avctx->channels - 1];
351 opus->stream_count = avctx->channels - coupled_stream_count;
352 memcpy(libopus_channel_mapping,
353 opus_vorbis_channel_map[avctx->channels - 1],
354 avctx->channels * sizeof(*libopus_channel_mapping));
356 enc = opus_multistream_encoder_create(
357 avctx->sample_rate, avctx->channels, opus->stream_count,
358 coupled_stream_count,
359 libavcodec_libopus_channel_map[avctx->channels - 1],
360 opus->opts.application, &ret);
362 /* Use the newer multistream API. The encoder will set the channel
363 * mapping and coupled stream counts to its internal defaults and will
364 * use surround masking analysis to save bits. */
365 mapping_family = opus->opts.mapping_family;
366 enc = opus_multistream_surround_encoder_create(
367 avctx->sample_rate, avctx->channels, mapping_family,
368 &opus->stream_count, &coupled_stream_count, libopus_channel_mapping,
369 opus->opts.application, &ret);
372 if (ret != OPUS_OK) {
373 av_log(avctx, AV_LOG_ERROR,
374 "Failed to create encoder: %s\n", opus_strerror(ret));
375 return ff_opus_error_to_averror(ret);
378 if (!avctx->bit_rate) {
379 /* Sane default copied from opusenc */
380 avctx->bit_rate = 64000 * opus->stream_count +
381 32000 * coupled_stream_count;
382 av_log(avctx, AV_LOG_WARNING,
383 "No bit rate set. Defaulting to %"PRId64" bps.\n", avctx->bit_rate);
386 if (avctx->bit_rate < 500 || avctx->bit_rate > 256000 * avctx->channels) {
387 av_log(avctx, AV_LOG_ERROR, "The bit rate %"PRId64" bps is unsupported. "
388 "Please choose a value between 500 and %d.\n", avctx->bit_rate,
389 256000 * avctx->channels);
390 ret = AVERROR(EINVAL);
394 ret = libopus_configure_encoder(avctx, enc, &opus->opts);
395 if (ret != OPUS_OK) {
396 ret = ff_opus_error_to_averror(ret);
400 /* Header includes channel mapping table if and only if mapping family is NOT 0 */
401 header_size = 19 + (mapping_family == 0 ? 0 : 2 + avctx->channels);
402 avctx->extradata = av_malloc(header_size + AV_INPUT_BUFFER_PADDING_SIZE);
403 if (!avctx->extradata) {
404 av_log(avctx, AV_LOG_ERROR, "Failed to allocate extradata.\n");
405 ret = AVERROR(ENOMEM);
408 avctx->extradata_size = header_size;
410 opus->samples = av_mallocz_array(frame_size, avctx->channels *
411 av_get_bytes_per_sample(avctx->sample_fmt));
412 if (!opus->samples) {
413 av_log(avctx, AV_LOG_ERROR, "Failed to allocate samples buffer.\n");
414 ret = AVERROR(ENOMEM);
418 ret = opus_multistream_encoder_ctl(enc, OPUS_GET_LOOKAHEAD(&avctx->initial_padding));
420 av_log(avctx, AV_LOG_WARNING,
421 "Unable to get number of lookahead samples: %s\n",
424 libopus_write_header(avctx, opus->stream_count, coupled_stream_count,
425 mapping_family, libopus_channel_mapping);
427 ff_af_queue_init(avctx, &opus->afq);
434 opus_multistream_encoder_destroy(enc);
435 av_freep(&avctx->extradata);
439 static void libopus_copy_samples_with_channel_map(
440 uint8_t *dst, const uint8_t *src, const uint8_t *channel_map,
441 int nb_channels, int nb_samples, int bytes_per_sample) {
443 for (sample = 0; sample < nb_samples; ++sample) {
444 for (channel = 0; channel < nb_channels; ++channel) {
445 const size_t src_pos = bytes_per_sample * (nb_channels * sample + channel);
446 const size_t dst_pos = bytes_per_sample * (nb_channels * sample + channel_map[channel]);
448 memcpy(&dst[dst_pos], &src[src_pos], bytes_per_sample);
453 static int libopus_encode(AVCodecContext *avctx, AVPacket *avpkt,
454 const AVFrame *frame, int *got_packet_ptr)
456 LibopusEncContext *opus = avctx->priv_data;
457 const int bytes_per_sample = av_get_bytes_per_sample(avctx->sample_fmt);
458 const int sample_size = avctx->channels * bytes_per_sample;
464 ret = ff_af_queue_add(&opus->afq, frame);
467 if (opus->encoder_channel_map != NULL) {
468 audio = opus->samples;
469 libopus_copy_samples_with_channel_map(
470 audio, frame->data[0], opus->encoder_channel_map,
471 avctx->channels, frame->nb_samples, bytes_per_sample);
472 } else if (frame->nb_samples < opus->opts.packet_size) {
473 audio = opus->samples;
474 memcpy(audio, frame->data[0], frame->nb_samples * sample_size);
476 audio = frame->data[0];
478 if (!opus->afq.remaining_samples || (!opus->afq.frame_alloc && !opus->afq.frame_count))
480 audio = opus->samples;
481 memset(audio, 0, opus->opts.packet_size * sample_size);
484 /* Maximum packet size taken from opusenc in opus-tools. 120ms packets
485 * consist of 6 frames in one packet. The maximum frame size is 1275
486 * bytes along with the largest possible packet header of 7 bytes. */
487 if ((ret = ff_alloc_packet2(avctx, avpkt, (1275 * 6 + 7) * opus->stream_count, 0)) < 0)
490 if (avctx->sample_fmt == AV_SAMPLE_FMT_FLT)
491 ret = opus_multistream_encode_float(opus->enc, (float *)audio,
492 opus->opts.packet_size,
493 avpkt->data, avpkt->size);
495 ret = opus_multistream_encode(opus->enc, (opus_int16 *)audio,
496 opus->opts.packet_size,
497 avpkt->data, avpkt->size);
500 av_log(avctx, AV_LOG_ERROR,
501 "Error encoding frame: %s\n", opus_strerror(ret));
502 return ff_opus_error_to_averror(ret);
505 av_shrink_packet(avpkt, ret);
507 ff_af_queue_remove(&opus->afq, opus->opts.packet_size,
508 &avpkt->pts, &avpkt->duration);
510 discard_padding = opus->opts.packet_size - avpkt->duration;
511 // Check if subtraction resulted in an overflow
512 if ((discard_padding < opus->opts.packet_size) != (avpkt->duration > 0)) {
513 av_packet_unref(avpkt);
514 return AVERROR(EINVAL);
516 if (discard_padding > 0) {
517 uint8_t* side_data = av_packet_new_side_data(avpkt,
518 AV_PKT_DATA_SKIP_SAMPLES,
521 av_packet_unref(avpkt);
522 return AVERROR(ENOMEM);
524 AV_WL32(side_data + 4, discard_padding);
532 static av_cold int libopus_encode_close(AVCodecContext *avctx)
534 LibopusEncContext *opus = avctx->priv_data;
536 opus_multistream_encoder_destroy(opus->enc);
538 ff_af_queue_close(&opus->afq);
540 av_freep(&opus->samples);
541 av_freep(&avctx->extradata);
546 #define OFFSET(x) offsetof(LibopusEncContext, opts.x)
547 #define FLAGS AV_OPT_FLAG_AUDIO_PARAM | AV_OPT_FLAG_ENCODING_PARAM
548 static const AVOption libopus_options[] = {
549 { "application", "Intended application type", OFFSET(application), AV_OPT_TYPE_INT, { .i64 = OPUS_APPLICATION_AUDIO }, OPUS_APPLICATION_VOIP, OPUS_APPLICATION_RESTRICTED_LOWDELAY, FLAGS, "application" },
550 { "voip", "Favor improved speech intelligibility", 0, AV_OPT_TYPE_CONST, { .i64 = OPUS_APPLICATION_VOIP }, 0, 0, FLAGS, "application" },
551 { "audio", "Favor faithfulness to the input", 0, AV_OPT_TYPE_CONST, { .i64 = OPUS_APPLICATION_AUDIO }, 0, 0, FLAGS, "application" },
552 { "lowdelay", "Restrict to only the lowest delay modes", 0, AV_OPT_TYPE_CONST, { .i64 = OPUS_APPLICATION_RESTRICTED_LOWDELAY }, 0, 0, FLAGS, "application" },
553 { "frame_duration", "Duration of a frame in milliseconds", OFFSET(frame_duration), AV_OPT_TYPE_FLOAT, { .dbl = 20.0 }, 2.5, 120.0, FLAGS },
554 { "packet_loss", "Expected packet loss percentage", OFFSET(packet_loss), AV_OPT_TYPE_INT, { .i64 = 0 }, 0, 100, FLAGS },
555 { "fec", "Enable inband FEC. Expected packet loss must be non-zero", OFFSET(fec), AV_OPT_TYPE_BOOL, { .i64 = 0 }, 0, 1, FLAGS },
556 { "vbr", "Variable bit rate mode", OFFSET(vbr), AV_OPT_TYPE_INT, { .i64 = 1 }, 0, 2, FLAGS, "vbr" },
557 { "off", "Use constant bit rate", 0, AV_OPT_TYPE_CONST, { .i64 = 0 }, 0, 0, FLAGS, "vbr" },
558 { "on", "Use variable bit rate", 0, AV_OPT_TYPE_CONST, { .i64 = 1 }, 0, 0, FLAGS, "vbr" },
559 { "constrained", "Use constrained VBR", 0, AV_OPT_TYPE_CONST, { .i64 = 2 }, 0, 0, FLAGS, "vbr" },
560 { "mapping_family", "Channel Mapping Family", OFFSET(mapping_family), AV_OPT_TYPE_INT, { .i64 = -1 }, -1, 255, FLAGS, "mapping_family" },
561 #ifdef OPUS_SET_PHASE_INVERSION_DISABLED_REQUEST
562 { "apply_phase_inv", "Apply intensity stereo phase inversion", OFFSET(apply_phase_inv), AV_OPT_TYPE_BOOL, { .i64 = 1 }, 0, 1, FLAGS },
567 static const AVClass libopus_class = {
568 .class_name = "libopus",
569 .item_name = av_default_item_name,
570 .option = libopus_options,
571 .version = LIBAVUTIL_VERSION_INT,
574 static const AVCodecDefault libopus_defaults[] = {
576 { "compression_level", "10" },
580 static const int libopus_sample_rates[] = {
581 48000, 24000, 16000, 12000, 8000, 0,
584 const AVCodec ff_libopus_encoder = {
586 .long_name = NULL_IF_CONFIG_SMALL("libopus Opus"),
587 .type = AVMEDIA_TYPE_AUDIO,
588 .id = AV_CODEC_ID_OPUS,
589 .priv_data_size = sizeof(LibopusEncContext),
590 .init = libopus_encode_init,
591 .encode2 = libopus_encode,
592 .close = libopus_encode_close,
593 .capabilities = AV_CODEC_CAP_DELAY | AV_CODEC_CAP_SMALL_LAST_FRAME,
594 .sample_fmts = (const enum AVSampleFormat[]){ AV_SAMPLE_FMT_S16,
596 AV_SAMPLE_FMT_NONE },
597 .supported_samplerates = libopus_sample_rates,
598 .priv_class = &libopus_class,
599 .defaults = libopus_defaults,
600 .wrapper_name = "libopus",