+void FFmpegCapture::convert_audio(const AVFrame *audio_avframe, FrameAllocator::Frame *audio_frame, AudioFormat *audio_format)
+{
+ // Decide on a format. If there already is one in this audio frame,
+ // we're pretty much forced to use it. If not, we try to find an exact match.
+ // If that still doesn't work, we default to 32-bit signed chunked
+ // (float would be nice, but there's really no way to signal that yet).
+ AVSampleFormat dst_format;
+ if (audio_format->bits_per_sample == 0) {
+ switch (audio_avframe->format) {
+ case AV_SAMPLE_FMT_S16:
+ case AV_SAMPLE_FMT_S16P:
+ audio_format->bits_per_sample = 16;
+ dst_format = AV_SAMPLE_FMT_S16;
+ break;
+ case AV_SAMPLE_FMT_S32:
+ case AV_SAMPLE_FMT_S32P:
+ default:
+ audio_format->bits_per_sample = 32;
+ dst_format = AV_SAMPLE_FMT_S32;
+ break;
+ }
+ } else if (audio_format->bits_per_sample == 16) {
+ dst_format = AV_SAMPLE_FMT_S16;
+ } else if (audio_format->bits_per_sample == 32) {
+ dst_format = AV_SAMPLE_FMT_S32;
+ } else {
+ assert(false);
+ }
+ audio_format->num_channels = 2;
+
+ int64_t channel_layout = audio_avframe->channel_layout;
+ if (channel_layout == 0) {
+ channel_layout = av_get_default_channel_layout(audio_avframe->channels);
+ }
+
+ if (resampler == nullptr ||
+ audio_avframe->format != last_src_format ||
+ dst_format != last_dst_format ||
+ channel_layout != last_channel_layout ||
+ av_frame_get_sample_rate(audio_avframe) != last_sample_rate) {
+ avresample_free(&resampler);
+ resampler = avresample_alloc_context();
+ if (resampler == nullptr) {
+ fprintf(stderr, "Allocating resampler failed.\n");
+ exit(1);
+ }
+
+ av_opt_set_int(resampler, "in_channel_layout", channel_layout, 0);
+ av_opt_set_int(resampler, "out_channel_layout", AV_CH_LAYOUT_STEREO_DOWNMIX, 0);
+ av_opt_set_int(resampler, "in_sample_rate", av_frame_get_sample_rate(audio_avframe), 0);
+ av_opt_set_int(resampler, "out_sample_rate", OUTPUT_FREQUENCY, 0);
+ av_opt_set_int(resampler, "in_sample_fmt", audio_avframe->format, 0);
+ av_opt_set_int(resampler, "out_sample_fmt", dst_format, 0);
+
+ if (avresample_open(resampler) < 0) {
+ fprintf(stderr, "Could not open resample context.\n");
+ exit(1);
+ }
+
+ last_src_format = AVSampleFormat(audio_avframe->format);
+ last_dst_format = dst_format;
+ last_channel_layout = channel_layout;
+ last_sample_rate = av_frame_get_sample_rate(audio_avframe);
+ }
+
+ size_t bytes_per_sample = (audio_format->bits_per_sample / 8) * 2;
+ size_t num_samples_room = (audio_frame->size - audio_frame->len) / bytes_per_sample;
+
+ uint8_t *data = audio_frame->data + audio_frame->len;
+ int out_samples = avresample_convert(resampler, &data, 0, num_samples_room,
+ const_cast<uint8_t **>(audio_avframe->data), audio_avframe->linesize[0], audio_avframe->nb_samples);
+ if (out_samples < 0) {
+ fprintf(stderr, "Audio conversion failed.\n");
+ exit(1);
+ }
+
+ audio_frame->len += out_samples * bytes_per_sample;
+}
+