1 #include "audio_encoder.h"
4 #include <libavcodec/avcodec.h>
5 #include <libavformat/avformat.h>
6 #include <libswresample/swresample.h>
7 #include <libavutil/channel_layout.h>
8 #include <libavutil/error.h>
9 #include <libavutil/frame.h>
10 #include <libavutil/mem.h>
11 #include <libavutil/opt.h>
12 #include <libavutil/rational.h>
13 #include <libavutil/samplefmt.h>
25 #include "shared/mux.h"
26 #include "shared/timebase.h"
30 AudioEncoder::AudioEncoder(const string &codec_name, int bit_rate, const AVOutputFormat *oformat)
32 const AVCodec *codec = avcodec_find_encoder_by_name(codec_name.c_str());
33 if (codec == nullptr) {
34 fprintf(stderr, "ERROR: Could not find codec '%s'\n", codec_name.c_str());
38 ctx = avcodec_alloc_context3(codec);
39 ctx->bit_rate = bit_rate;
40 ctx->sample_rate = OUTPUT_FREQUENCY;
41 ctx->sample_fmt = codec->sample_fmts[0];
42 ctx->ch_layout.order = AV_CHANNEL_ORDER_NATIVE;
43 ctx->ch_layout.nb_channels = 2;
44 ctx->ch_layout.u.mask = AV_CH_LAYOUT_STEREO;
45 ctx->time_base = AVRational{1, TIMEBASE};
46 if (oformat->flags & AVFMT_GLOBALHEADER) {
47 ctx->flags |= AV_CODEC_FLAG_GLOBAL_HEADER;
49 if (avcodec_open2(ctx, codec, NULL) < 0) {
50 fprintf(stderr, "Could not open codec '%s'\n", codec_name.c_str());
55 int ok = swr_alloc_set_opts2(&resampler,
56 /*out_ch_layout=*/&ctx->ch_layout,
57 /*out_sample_fmt=*/ctx->sample_fmt,
58 /*out_sample_rate=*/OUTPUT_FREQUENCY,
59 /*in_ch_layout=*/&ctx->ch_layout,
60 /*in_sample_fmt=*/AV_SAMPLE_FMT_FLT,
61 /*in_sample_rate=*/OUTPUT_FREQUENCY,
65 fprintf(stderr, "Allocating resampler failed.\n");
69 if (swr_init(resampler) < 0) {
70 fprintf(stderr, "Could not open resample context.\n");
74 audio_frame = av_frame_alloc();
77 AudioEncoder::~AudioEncoder()
79 av_frame_free(&audio_frame);
81 avcodec_free_context(&ctx);
84 void AudioEncoder::encode_audio(const vector<float> &audio, int64_t audio_pts)
86 if (ctx->frame_size == 0) {
87 // No queueing needed.
88 assert(audio_queue.empty());
89 assert(audio.size() % 2 == 0);
90 encode_audio_one_frame(&audio[0], audio.size() / 2, audio_pts);
94 int64_t sample_offset = audio_queue.size();
96 audio_queue.insert(audio_queue.end(), audio.begin(), audio.end());
99 sample_num + ctx->frame_size * 2 <= audio_queue.size();
100 sample_num += ctx->frame_size * 2) {
101 int64_t adjusted_audio_pts = audio_pts + (int64_t(sample_num) - sample_offset) * TIMEBASE / (OUTPUT_FREQUENCY * 2);
102 encode_audio_one_frame(&audio_queue[sample_num],
106 audio_queue.erase(audio_queue.begin(), audio_queue.begin() + sample_num);
108 last_pts = audio_pts + audio.size() * TIMEBASE / (OUTPUT_FREQUENCY * 2);
111 void AudioEncoder::encode_audio_one_frame(const float *audio, size_t num_samples, int64_t audio_pts)
113 audio_frame->pts = audio_pts;
114 audio_frame->nb_samples = num_samples;
115 audio_frame->ch_layout.order = AV_CHANNEL_ORDER_NATIVE;
116 audio_frame->ch_layout.nb_channels = 2;
117 audio_frame->ch_layout.u.mask = AV_CH_LAYOUT_STEREO;
118 audio_frame->format = ctx->sample_fmt;
119 audio_frame->sample_rate = OUTPUT_FREQUENCY;
121 if (av_samples_alloc(audio_frame->data, nullptr, 2, num_samples, ctx->sample_fmt, 0) < 0) {
122 fprintf(stderr, "Could not allocate %zu samples.\n", num_samples);
126 if (swr_convert(resampler, audio_frame->data, num_samples, reinterpret_cast<const uint8_t **>(&audio), num_samples) < 0) {
127 fprintf(stderr, "Audio conversion failed.\n");
131 int err = avcodec_send_frame(ctx, audio_frame);
133 fprintf(stderr, "avcodec_send_frame() failed with error %d\n", err);
137 for ( ;; ) { // Termination condition within loop.
138 AVPacketWithDeleter pkt = av_packet_alloc_unique();
141 int err = avcodec_receive_packet(ctx, pkt.get());
143 pkt->stream_index = 1;
145 for (Mux *mux : muxes) {
146 mux->add_packet(*pkt, pkt->pts, pkt->dts);
148 } else if (err == AVERROR(EAGAIN)) {
151 fprintf(stderr, "avcodec_receive_frame() failed with error %d\n", err);
156 av_freep(&audio_frame->data[0]);
157 av_frame_unref(audio_frame);
160 void AudioEncoder::encode_last_audio()
162 if (!audio_queue.empty()) {
163 // Last frame can be whatever size we want.
164 assert(audio_queue.size() % 2 == 0);
165 encode_audio_one_frame(&audio_queue[0], audio_queue.size() / 2, last_pts);
169 if (ctx->codec->capabilities & AV_CODEC_CAP_DELAY) {
170 // Collect any delayed frames.
172 AVPacketWithDeleter pkt = av_packet_alloc_unique();
175 int err = avcodec_receive_packet(ctx, pkt.get());
177 pkt->stream_index = 1;
179 for (Mux *mux : muxes) {
180 mux->add_packet(*pkt, pkt->pts, pkt->dts);
182 } else if (err == AVERROR_EOF) {
185 fprintf(stderr, "avcodec_receive_frame() failed with error %d\n", err);
192 AVCodecParametersWithDeleter AudioEncoder::get_codec_parameters()
194 AVCodecParameters *codecpar = avcodec_parameters_alloc();
195 avcodec_parameters_from_context(codecpar, ctx);
196 return AVCodecParametersWithDeleter(codecpar);