X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=mixer.cpp;h=6bf81192fdfd99421398dedc189e23be75344a5d;hb=e2d886719370306464e2c67574bb6eef62a0a64e;hp=3ebfbeeecdbd788695109ba3f385842d49e09df0;hpb=4ed1495550e603bfb7584e1b1b9a0c74e57de223;p=nageru diff --git a/mixer.cpp b/mixer.cpp index 3ebfbee..6bf8119 100644 --- a/mixer.cpp +++ b/mixer.cpp @@ -33,6 +33,7 @@ #include "bmusb/bmusb.h" #include "context.h" +#include "defs.h" #include "h264encode.h" #include "pbo_frame_allocator.h" #include "ref_counted_gl_sync.h" @@ -68,7 +69,10 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) : 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)), + level_compressor(OUTPUT_FREQUENCY), + limiter(OUTPUT_FREQUENCY), + compressor(OUTPUT_FREQUENCY) { httpd.start(9095); @@ -120,7 +124,7 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) [this]{ resource_pool->clean_context(); }); - card->resampler.reset(new Resampler(48000.0, 48000.0, 2)); + card->resampling_queue.reset(new ResamplingQueue(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2)); card->usb->configure_card(); } @@ -145,8 +149,10 @@ Mixer::Mixer(const QSurfaceFormat &format, unsigned num_cards) "} \n"; cbcr_program_num = resource_pool->compile_glsl_program(cbcr_vert_shader, cbcr_frag_shader); - r128.init(2, 48000); + r128.init(2, OUTPUT_FREQUENCY); r128.integr_start(); + + locut.init(FILTER_HPF, 2); } Mixer::~Mixer() @@ -240,21 +246,21 @@ void Mixer::bm_frame(unsigned card_index, uint16_t timecode, unique_lock lock(card->audio_mutex); int unwrapped_timecode = timecode; - if (dropped_frames > 60 * 2) { + if (dropped_frames > FPS * 2) { fprintf(stderr, "Card %d lost more than two seconds (or time code jumping around), resetting resampler\n", card_index); - card->resampler.reset(new Resampler(48000.0, 48000.0, 2)); + card->resampling_queue.reset(new ResamplingQueue(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2)); } else if (dropped_frames > 0) { // Insert silence as needed. fprintf(stderr, "Card %d dropped %d frame(s) (before timecode 0x%04x), inserting silence.\n", card_index, dropped_frames, timecode); vector silence; - silence.resize((48000 / 60) * 2); + silence.resize((OUTPUT_FREQUENCY / FPS) * 2); for (int i = 0; i < dropped_frames; ++i) { - card->resampler->add_input_samples((unwrapped_timecode - dropped_frames + i) / 60.0, silence.data(), (48000 / 60)); + card->resampling_queue->add_input_samples((unwrapped_timecode - dropped_frames + i) / double(FPS), silence.data(), (OUTPUT_FREQUENCY / FPS)); } } - card->resampler->add_input_samples(unwrapped_timecode / 60.0, audio.data(), num_samples); + card->resampling_queue->add_input_samples(unwrapped_timecode / double(FPS), audio.data(), num_samples); } // Done with the audio, so release it. @@ -357,7 +363,6 @@ void Mixer::thread_func() 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(); @@ -370,7 +375,7 @@ void Mixer::thread_func() if (frame_num != card_copy[0].dropped_frames) { // For dropped frames, increase the pts. ++dropped_frames; - pts_int += TIMEBASE / 60; + pts_int += TIMEBASE / FPS; } } @@ -381,7 +386,8 @@ void Mixer::thread_func() double loudness_range_high = r128.range_max(); audio_level_callback(loudness_s, 20.0 * log10(peak), - loudness_i, loudness_range_low, loudness_range_high); + loudness_i, loudness_range_low, loudness_range_high, + last_gain_staging_db); } for (unsigned card_index = 1; card_index < num_cards; ++card_index) { @@ -398,7 +404,7 @@ void Mixer::thread_func() // just increase the pts (skipping over this frame) and don't try to compute anything new. if (card_copy[0].new_frame->len == 0) { ++dropped_frames; - pts_int += TIMEBASE / 60; + pts_int += TIMEBASE / FPS; continue; } @@ -460,10 +466,10 @@ void Mixer::thread_func() for (unsigned card_index = 0; card_index < num_cards; ++card_index) { input_frames.push_back(bmusb_current_rendering_frame[card_index]); } - const int64_t av_delay = TIMEBASE / 10; // Corresponds to the fixed delay in resampler.h. TODO: Make less hard-coded. + const int64_t av_delay = TIMEBASE / 10; // Corresponds to the fixed delay in resampling_queue.h. TODO: Make less hard-coded. h264_encoder->end_frame(fence, pts_int + av_delay, input_frames); ++frame; - pts_int += TIMEBASE / 60; + pts_int += TIMEBASE / FPS; // The live frame just shows the RGBA texture we just rendered. // It owns rgba_tex now. @@ -521,25 +527,88 @@ void Mixer::thread_func() void Mixer::process_audio_one_frame() { - // TODO: Allow using audio from the other card(s) as well. + vector samples_card; vector samples_out; for (unsigned card_index = 0; card_index < num_cards; ++card_index) { - samples_out.resize((48000 / 60) * 2); + samples_card.resize((OUTPUT_FREQUENCY / FPS) * 2); { unique_lock lock(cards[card_index].audio_mutex); - if (!cards[card_index].resampler->get_output_samples(pts(), &samples_out[0], 48000 / 60)) { + if (!cards[card_index].resampling_queue->get_output_samples(pts(), &samples_card[0], OUTPUT_FREQUENCY / FPS)) { 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 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); } } + + // Cut away everything under 150 Hz; we don't need it for voice, + // and it will reduce headroom and confuse the compressor. + // (In particular, any hums at 50 or 60 Hz should be dampened.) + locut.render(samples_out.data(), samples_out.size() / 2, locut_cutoff_hz * 2.0 * M_PI / OUTPUT_FREQUENCY, 0.5f); + + // 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 -14 dBFS. -14 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. + float ref_level_dbfs = -14.0f; + { + float threshold = 0.01f; // -40 dBFS. + float ratio = 20.0f; + float attack_time = 0.5f; + float release_time = 20.0f; + float makeup_gain = pow(10.0f, (ref_level_dbfs - (-40.0f)) / 20.0f); // +26 dB. + level_compressor.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain); + last_gain_staging_db = 20.0 * log10(level_compressor.get_attenuation() * makeup_gain); + } + +#if 0 + printf("level=%f (%+5.2f dBFS) attenuation=%f (%+5.2f dB) end_result=%+5.2f dB\n", + level_compressor.get_level(), 20.0 * log10(level_compressor.get_level()), + level_compressor.get_attenuation(), 20.0 * log10(level_compressor.get_attenuation()), + 20.0 * log10(level_compressor.get_level() * level_compressor.get_attenuation() * makeup_gain)); +#endif + +// float limiter_att, compressor_att; + + // Then a limiter at +0 dB (so, -14 dBFS) to take out the worst peaks only. + // Note that since ratio is not infinite, we could go slightly higher than this. + // Probably more tuning is warranted here. + { + float threshold = pow(10.0f, (ref_level_dbfs + 0.0f) / 20.0f); // +0 dB. + float ratio = 30.0f; + float attack_time = 0.0f; // Instant. + float release_time = 0.005f; + float makeup_gain = 1.0f; // 0 dB. + limiter.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain); +// limiter_att = limiter.get_attenuation(); + } + + // Finally, the real compressor. + { + float threshold = pow(10.0f, (ref_level_dbfs - 12.0f) / 20.0f); // -12 dB. + float ratio = 20.0f; + float attack_time = 0.005f; + float release_time = 0.040f; + float makeup_gain = 2.0f; // +6 dB. + compressor.process(samples_out.data(), samples_out.size() / 2, threshold, ratio, attack_time, release_time, makeup_gain); +// compressor_att = compressor.get_attenuation(); + } + +// printf("limiter=%+5.1f compressor=%+5.1f\n", 20.0*log10(limiter_att), 20.0*log10(compressor_att)); + + // Find peak and R128 levels. + peak = max(peak, find_peak(samples_out)); + vector 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) @@ -627,6 +696,13 @@ void Mixer::channel_clicked(int preview_num) theme->channel_clicked(preview_num); } +void Mixer::reset_meters() +{ + peak = 0.0f; + r128.reset(); + r128.integr_start(); +} + Mixer::OutputChannel::~OutputChannel() { if (has_current_frame) {