: httpd("test.ts", WIDTH, HEIGHT),
num_cards(num_cards),
mixer_surface(create_surface(format)),
- h264_encoder_surface(create_surface(format))
+ h264_encoder_surface(create_surface(format)),
+ compressor(48000.0f)
{
httpd.start(9095);
card_copy[card_index].new_data_ready = card->new_data_ready;
card_copy[card_index].new_frame = card->new_frame;
card_copy[card_index].new_data_ready_fence = card->new_data_ready_fence;
- card_copy[card_index].new_frame_audio = move(card->new_frame_audio);
card_copy[card_index].dropped_frames = card->dropped_frames;
card->new_data_ready = false;
card->new_data_ready_changed.notify_all();
void Mixer::process_audio_one_frame()
{
- // TODO: Allow using audio from the other card(s) as well.
+ vector<float> samples_card;
vector<float> samples_out;
for (unsigned card_index = 0; card_index < num_cards; ++card_index) {
- samples_out.resize((48000 / 60) * 2);
+ samples_card.resize((48000 / 60) * 2);
{
unique_lock<mutex> lock(cards[card_index].audio_mutex);
- if (!cards[card_index].resampler->get_output_samples(pts(), &samples_out[0], 48000 / 60)) {
+ if (!cards[card_index].resampler->get_output_samples(pts(), &samples_card[0], 48000 / 60)) {
printf("Card %d reported previous underrun.\n", card_index);
}
}
+ // TODO: Allow using audio from the other card(s) as well.
if (card_index == 0) {
- vector<float> left, right;
- peak = std::max(peak, find_peak(samples_out));
- deinterleave_samples(samples_out, &left, &right);
- float *ptrs[] = { left.data(), right.data() };
- r128.process(left.size(), ptrs);
- h264_encoder->add_audio(pts_int, move(samples_out));
+ samples_out = move(samples_card);
}
}
+
+ // Apply a level compressor to get the general level right.
+ // Basically, if it's over about -40 dBFS, we squeeze it down to that level
+ // (or more precisely, near it, since we don't use infinite ratio),
+ // then apply a makeup gain to get it to -12 dBFS. -12 dBFS is, of course,
+ // entirely arbitrary, but from practical tests with speech, it seems to
+ // put ut around -23 LUFS, so it's a reasonable starting point for later use.
+ //
+ // TODO: Hook this up to a UI, so we can see the effects, and/or turn it off
+ // to control the gain manually instead. For now, there's only the #if-ed out
+ // code below.
+ //
+ // TODO: Add the actual compressors/limiters (for taking care of transients)
+ // later in the chain.
+ float threshold = 0.01f; // -40 dBFS.
+ float ratio = 20.0f;
+ float attack_time = 0.1f;
+ float release_time = 10.0f;
+ float makeup_gain = pow(10.0f, 28.0f / 20.0f); // +28 dB takes us to -12 dBFS.
+ compressor.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain);
+
+#if 0
+ printf("level=%f (%+5.2f dBFS) attenuation=%f (%+5.2f dB) end_result=%+5.2f dB\n",
+ compressor.get_level(), 20.0 * log10(compressor.get_level()),
+ compressor.get_attenuation(), 20.0 * log10(compressor.get_attenuation()),
+ 20.0 * log10(compressor.get_level() * compressor.get_attenuation() * makeup_gain));
+#endif
+
+ // Find peak and R128 levels.
+ peak = std::max(peak, find_peak(samples_out));
+ vector<float> left, right;
+ deinterleave_samples(samples_out, &left, &right);
+ float *ptrs[] = { left.data(), right.data() };
+ r128.process(left.size(), ptrs);
+
+ // Actually add the samples to the output.
+ h264_encoder->add_audio(pts_int, move(samples_out));
}
void Mixer::subsample_chroma(GLuint src_tex, GLuint dst_tex)