X-Git-Url: https://git.sesse.net/?p=nageru;a=blobdiff_plain;f=audio_mixer.cpp;h=e7d545f69c6320c25cddcde039f3b100d8332529;hp=46ae44c6f74e73a1f432b712e76008aafdb92394;hb=703e00da89118df9be0354dda621bed023e6030e;hpb=7c1bb8357495778076a47636c2c4192674034165 diff --git a/audio_mixer.cpp b/audio_mixer.cpp index 46ae44c..e7d545f 100644 --- a/audio_mixer.cpp +++ b/audio_mixer.cpp @@ -1,24 +1,32 @@ #include "audio_mixer.h" #include -#include #include -#include #include -#include -#include -#ifdef __SSE__ +#include +#ifdef __SSE2__ #include #endif +#include +#include +#include +#include +#include +#include +#include +#include +#include +#include #include "db.h" #include "flags.h" -#include "mixer.h" +#include "metrics.h" #include "state.pb.h" #include "timebase.h" using namespace bmusb; using namespace std; +using namespace std::chrono; using namespace std::placeholders; namespace { @@ -159,13 +167,13 @@ void deinterleave_samples(const vector &in, vector *out_l, vector< } // namespace -AudioMixer::AudioMixer(unsigned num_cards) - : num_cards(num_cards), +AudioMixer::AudioMixer(unsigned num_capture_cards, unsigned num_ffmpeg_inputs) + : num_capture_cards(num_capture_cards), + num_ffmpeg_inputs(num_ffmpeg_inputs), + ffmpeg_inputs(new AudioDevice[num_ffmpeg_inputs]), limiter(OUTPUT_FREQUENCY), correlation(OUTPUT_FREQUENCY) { - global_audio_mixer = this; - for (unsigned bus_index = 0; bus_index < MAX_BUSES; ++bus_index) { locut[bus_index].init(FILTER_HPF, 2); eq[bus_index][EQ_BAND_BASS].init(FILTER_LOW_SHELF, 1); @@ -178,9 +186,19 @@ AudioMixer::AudioMixer(unsigned num_cards) } set_limiter_enabled(global_flags.limiter_enabled); set_final_makeup_gain_auto(global_flags.final_makeup_gain_auto); + + r128.init(2, OUTPUT_FREQUENCY); + r128.integr_start(); + + // hlen=16 is pretty low quality, but we use quite a bit of CPU otherwise, + // and there's a limit to how important the peak meter is. + peak_resampler.setup(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY * 4, /*num_channels=*/2, /*hlen=*/16, /*frel=*/1.0); + + global_audio_mixer = this; alsa_pool.init(); if (!global_flags.input_mapping_filename.empty()) { + // Must happen after ALSAPool is initialized, as it needs to know the card list. current_mapping_mode = MappingMode::MULTICHANNEL; InputMapping new_input_mapping; if (!load_input_mapping_from_file(get_devices(), @@ -198,12 +216,13 @@ AudioMixer::AudioMixer(unsigned num_cards) } } - r128.init(2, OUTPUT_FREQUENCY); - r128.integr_start(); - - // hlen=16 is pretty low quality, but we use quite a bit of CPU otherwise, - // and there's a limit to how important the peak meter is. - peak_resampler.setup(OUTPUT_FREQUENCY, OUTPUT_FREQUENCY * 4, /*num_channels=*/2, /*hlen=*/16, /*frel=*/1.0); + global_metrics.add("audio_loudness_short_lufs", &metric_audio_loudness_short_lufs, Metrics::TYPE_GAUGE); + global_metrics.add("audio_loudness_integrated_lufs", &metric_audio_loudness_integrated_lufs, Metrics::TYPE_GAUGE); + global_metrics.add("audio_loudness_range_low_lufs", &metric_audio_loudness_range_low_lufs, Metrics::TYPE_GAUGE); + global_metrics.add("audio_loudness_range_high_lufs", &metric_audio_loudness_range_high_lufs, Metrics::TYPE_GAUGE); + global_metrics.add("audio_peak_dbfs", &metric_audio_peak_dbfs, Metrics::TYPE_GAUGE); + global_metrics.add("audio_final_makeup_gain_db", &metric_audio_final_makeup_gain_db, Metrics::TYPE_GAUGE); + global_metrics.add("audio_correlation", &metric_audio_correlation, Metrics::TYPE_GAUGE); } void AudioMixer::reset_resampler(DeviceSpec device_spec) @@ -219,14 +238,13 @@ void AudioMixer::reset_resampler_mutex_held(DeviceSpec device_spec) if (device->interesting_channels.empty()) { device->resampling_queue.reset(); } else { - // TODO: ResamplingQueue should probably take the full device spec. - // (It's only used for console output, though.) - device->resampling_queue.reset(new ResamplingQueue(device_spec.index, device->capture_frequency, OUTPUT_FREQUENCY, device->interesting_channels.size())); + device->resampling_queue.reset(new ResamplingQueue( + device_spec, device->capture_frequency, OUTPUT_FREQUENCY, device->interesting_channels.size(), + global_flags.audio_queue_length_ms * 0.001)); } - device->next_local_pts = 0; } -bool AudioMixer::add_audio(DeviceSpec device_spec, const uint8_t *data, unsigned num_samples, AudioFormat audio_format, int64_t frame_length) +bool AudioMixer::add_audio(DeviceSpec device_spec, const uint8_t *data, unsigned num_samples, AudioFormat audio_format, int64_t frame_length, steady_clock::time_point frame_time) { AudioDevice *device = find_audio_device(device_spec); @@ -265,10 +283,14 @@ bool AudioMixer::add_audio(DeviceSpec device_spec, const uint8_t *data, unsigned } } + // If we changed frequency since last frame, we'll need to reset the resampler. + if (audio_format.sample_rate != device->capture_frequency) { + device->capture_frequency = audio_format.sample_rate; + reset_resampler_mutex_held(device_spec); + } + // Now add it. - int64_t local_pts = device->next_local_pts; - device->resampling_queue->add_input_samples(local_pts / double(TIMEBASE), audio.get(), num_samples); - device->next_local_pts = local_pts + frame_length; + device->resampling_queue->add_input_samples(frame_time, audio.get(), num_samples, ResamplingQueue::ADJUST_RATE); return true; } @@ -290,11 +312,7 @@ bool AudioMixer::add_silence(DeviceSpec device_spec, unsigned samples_per_frame, vector silence(samples_per_frame * num_channels, 0.0f); for (unsigned i = 0; i < num_frames; ++i) { - device->resampling_queue->add_input_samples(device->next_local_pts / double(TIMEBASE), silence.data(), samples_per_frame); - // Note that if the format changed in the meantime, we have - // no way of detecting that; we just have to assume the frame length - // is always the same. - device->next_local_pts += frame_length; + device->resampling_queue->add_input_samples(steady_clock::now(), silence.data(), samples_per_frame, ResamplingQueue::DO_NOT_ADJUST_RATE); } return true; } @@ -371,6 +389,8 @@ AudioMixer::AudioDevice *AudioMixer::find_audio_device(DeviceSpec device) return &video_cards[device.index]; case InputSourceType::ALSA_INPUT: return &alsa_inputs[device.index]; + case InputSourceType::FFMPEG_VIDEO_INPUT: + return &ffmpeg_inputs[device.index]; case InputSourceType::SILENCE: default: assert(false); @@ -409,7 +429,8 @@ void AudioMixer::fill_audio_bus(const map> &samples_ca memset(output, 0, num_samples * 2 * sizeof(*output)); } else { assert(bus.device.type == InputSourceType::CAPTURE_CARD || - bus.device.type == InputSourceType::ALSA_INPUT); + bus.device.type == InputSourceType::ALSA_INPUT || + bus.device.type == InputSourceType::FFMPEG_VIDEO_INPUT); const float *lsrc, *rsrc; unsigned lstride, rstride; float *dptr = output; @@ -439,6 +460,12 @@ vector AudioMixer::get_active_devices() const ret.push_back(device_spec); } } + for (unsigned card_index = 0; card_index < num_ffmpeg_inputs; ++card_index) { + const DeviceSpec device_spec{InputSourceType::FFMPEG_VIDEO_INPUT, card_index}; + if (!find_audio_device(device_spec)->interesting_channels.empty()) { + ret.push_back(device_spec); + } + } return ret; } @@ -467,7 +494,7 @@ void apply_gain(float db, float last_db, vector *samples) } // namespace -vector AudioMixer::get_output(double pts, unsigned num_samples, ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy) +vector AudioMixer::get_output(steady_clock::time_point ts, unsigned num_samples, ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy) { map> samples_card; vector samples_bus; @@ -482,7 +509,7 @@ vector AudioMixer::get_output(double pts, unsigned num_samples, Resamplin memset(&samples_card[device_spec][0], 0, samples_card[device_spec].size() * sizeof(float)); } else { device->resampling_queue->get_output_samples( - pts, + ts, &samples_card[device_spec][0], num_samples, rate_adjustment_policy); @@ -577,13 +604,12 @@ vector AudioMixer::get_output(double pts, unsigned num_samples, Resamplin // (half-time of 30 seconds). double target_loudness_factor, alpha; double loudness_lu = r128.loudness_M() - ref_level_lufs; - double current_makeup_lu = to_db(final_makeup_gain); target_loudness_factor = final_makeup_gain * from_db(-loudness_lu); - // If we're outside +/- 5 LU uncorrected, we don't count it as + // If we're outside +/- 5 LU (after correction), we don't count it as // a normal signal (probably silence) and don't change the // correction factor; just apply what we already have. - if (fabs(loudness_lu - current_makeup_lu) >= 5.0 || !final_makeup_gain_auto) { + if (fabs(loudness_lu) >= 5.0 || !final_makeup_gain_auto) { alpha = 0.0; } else { // Formula adapted from @@ -819,23 +845,35 @@ void AudioMixer::send_audio_level_callback() double loudness_range_low = r128.range_min(); double loudness_range_high = r128.range_max(); + metric_audio_loudness_short_lufs = loudness_s; + metric_audio_loudness_integrated_lufs = loudness_i; + metric_audio_loudness_range_low_lufs = loudness_range_low; + metric_audio_loudness_range_high_lufs = loudness_range_high; + metric_audio_peak_dbfs = to_db(peak); + metric_audio_final_makeup_gain_db = to_db(final_makeup_gain); + metric_audio_correlation = correlation.get_correlation(); + vector bus_levels; bus_levels.resize(input_mapping.buses.size()); { lock_guard lock(compressor_mutex); for (unsigned bus_index = 0; bus_index < bus_levels.size(); ++bus_index) { - bus_levels[bus_index].current_level_dbfs[0] = to_db(peak_history[bus_index][0].current_level); - bus_levels[bus_index].current_level_dbfs[1] = to_db(peak_history[bus_index][1].current_level); - bus_levels[bus_index].peak_level_dbfs[0] = to_db(peak_history[bus_index][0].current_peak); - bus_levels[bus_index].peak_level_dbfs[1] = to_db(peak_history[bus_index][1].current_peak); - bus_levels[bus_index].historic_peak_dbfs = to_db( + BusLevel &levels = bus_levels[bus_index]; + BusMetrics &metrics = bus_metrics[bus_index]; + + levels.current_level_dbfs[0] = metrics.current_level_dbfs[0] = to_db(peak_history[bus_index][0].current_level); + levels.current_level_dbfs[1] = metrics.current_level_dbfs[1] = to_db(peak_history[bus_index][1].current_level); + levels.peak_level_dbfs[0] = metrics.peak_level_dbfs[0] = to_db(peak_history[bus_index][0].current_peak); + levels.peak_level_dbfs[1] = metrics.peak_level_dbfs[1] = to_db(peak_history[bus_index][1].current_peak); + levels.historic_peak_dbfs = metrics.historic_peak_dbfs = to_db( max(peak_history[bus_index][0].historic_peak, peak_history[bus_index][1].historic_peak)); - bus_levels[bus_index].gain_staging_db = gain_staging_db[bus_index]; + levels.gain_staging_db = metrics.gain_staging_db = gain_staging_db[bus_index]; if (compressor_enabled[bus_index]) { - bus_levels[bus_index].compressor_attenuation_db = -to_db(compressor[bus_index]->get_attenuation()); + levels.compressor_attenuation_db = metrics.compressor_attenuation_db = -to_db(compressor[bus_index]->get_attenuation()); } else { - bus_levels[bus_index].compressor_attenuation_db = 0.0; + levels.compressor_attenuation_db = 0.0; + metrics.compressor_attenuation_db = 0.0 / 0.0; } } } @@ -851,7 +889,7 @@ map AudioMixer::get_devices() lock_guard lock(audio_mutex); map devices; - for (unsigned card_index = 0; card_index < num_cards; ++card_index) { + for (unsigned card_index = 0; card_index < num_capture_cards; ++card_index) { const DeviceSpec spec{ InputSourceType::CAPTURE_CARD, card_index }; const AudioDevice *device = &video_cards[card_index]; DeviceInfo info; @@ -871,6 +909,14 @@ map AudioMixer::get_devices() info.alsa_address = device.address; devices.insert(make_pair(spec, info)); } + for (unsigned card_index = 0; card_index < num_ffmpeg_inputs; ++card_index) { + const DeviceSpec spec{ InputSourceType::FFMPEG_VIDEO_INPUT, card_index }; + const AudioDevice *device = &ffmpeg_inputs[card_index]; + DeviceInfo info; + info.display_name = device->display_name; + info.num_channels = 2; + devices.insert(make_pair(spec, info)); + } return devices; } @@ -897,16 +943,25 @@ void AudioMixer::serialize_device(DeviceSpec device_spec, DeviceSpecProto *devic case InputSourceType::ALSA_INPUT: alsa_pool.serialize_device(device_spec.index, device_spec_proto); break; + case InputSourceType::FFMPEG_VIDEO_INPUT: + device_spec_proto->set_type(DeviceSpecProto::FFMPEG_VIDEO_INPUT); + device_spec_proto->set_index(device_spec.index); + device_spec_proto->set_display_name(ffmpeg_inputs[device_spec.index].display_name); + break; } } void AudioMixer::set_simple_input(unsigned card_index) { + assert(card_index < num_capture_cards + num_ffmpeg_inputs); InputMapping new_input_mapping; InputMapping::Bus input; input.name = "Main"; - input.device.type = InputSourceType::CAPTURE_CARD; - input.device.index = card_index; + if (card_index >= num_capture_cards) { + input.device = DeviceSpec{InputSourceType::FFMPEG_VIDEO_INPUT, card_index - num_capture_cards}; + } else { + input.device = DeviceSpec{InputSourceType::CAPTURE_CARD, card_index}; + } input.source_channel[0] = 0; input.source_channel[1] = 1; @@ -926,6 +981,11 @@ unsigned AudioMixer::get_simple_input() const input_mapping.buses[0].source_channel[0] == 0 && input_mapping.buses[0].source_channel[1] == 1) { return input_mapping.buses[0].device.index; + } else if (input_mapping.buses.size() == 1 && + input_mapping.buses[0].device.type == InputSourceType::FFMPEG_VIDEO_INPUT && + input_mapping.buses[0].source_channel[0] == 0 && + input_mapping.buses[0].source_channel[1] == 1) { + return input_mapping.buses[0].device.index + num_capture_cards; } else { return numeric_limits::max(); } @@ -949,15 +1009,76 @@ void AudioMixer::set_input_mapping_lock_held(const InputMapping &new_input_mappi map> interesting_channels; for (const InputMapping::Bus &bus : new_input_mapping.buses) { if (bus.device.type == InputSourceType::CAPTURE_CARD || - bus.device.type == InputSourceType::ALSA_INPUT) { + bus.device.type == InputSourceType::ALSA_INPUT || + bus.device.type == InputSourceType::FFMPEG_VIDEO_INPUT) { for (unsigned channel = 0; channel < 2; ++channel) { if (bus.source_channel[channel] != -1) { interesting_channels[bus.device].insert(bus.source_channel[channel]); } } + } else { + assert(bus.device.type == InputSourceType::SILENCE); } } + // Kill all the old metrics, and set up new ones. + for (unsigned bus_index = 0; bus_index < input_mapping.buses.size(); ++bus_index) { + BusMetrics &metrics = bus_metrics[bus_index]; + + vector> labels_left = metrics.labels; + labels_left.emplace_back("channel", "left"); + vector> labels_right = metrics.labels; + labels_right.emplace_back("channel", "right"); + + global_metrics.remove("bus_current_level_dbfs", labels_left); + global_metrics.remove("bus_current_level_dbfs", labels_right); + global_metrics.remove("bus_peak_level_dbfs", labels_left); + global_metrics.remove("bus_peak_level_dbfs", labels_right); + global_metrics.remove("bus_historic_peak_dbfs", metrics.labels); + global_metrics.remove("bus_gain_staging_db", metrics.labels); + global_metrics.remove("bus_compressor_attenuation_db", metrics.labels); + } + bus_metrics.reset(new BusMetrics[new_input_mapping.buses.size()]); + for (unsigned bus_index = 0; bus_index < new_input_mapping.buses.size(); ++bus_index) { + const InputMapping::Bus &bus = new_input_mapping.buses[bus_index]; + BusMetrics &metrics = bus_metrics[bus_index]; + + char bus_index_str[16], source_index_str[16], source_channels_str[64]; + snprintf(bus_index_str, sizeof(bus_index_str), "%u", bus_index); + snprintf(source_index_str, sizeof(source_index_str), "%u", bus.device.index); + snprintf(source_channels_str, sizeof(source_channels_str), "%d:%d", bus.source_channel[0], bus.source_channel[1]); + + vector> labels; + metrics.labels.emplace_back("index", bus_index_str); + metrics.labels.emplace_back("name", bus.name); + if (bus.device.type == InputSourceType::SILENCE) { + metrics.labels.emplace_back("source_type", "silence"); + } else if (bus.device.type == InputSourceType::CAPTURE_CARD) { + metrics.labels.emplace_back("source_type", "capture_card"); + } else if (bus.device.type == InputSourceType::ALSA_INPUT) { + metrics.labels.emplace_back("source_type", "alsa_input"); + } else if (bus.device.type == InputSourceType::FFMPEG_VIDEO_INPUT) { + metrics.labels.emplace_back("source_type", "ffmpeg_video_input"); + } else { + assert(false); + } + metrics.labels.emplace_back("source_index", source_index_str); + metrics.labels.emplace_back("source_channels", source_channels_str); + + vector> labels_left = metrics.labels; + labels_left.emplace_back("channel", "left"); + vector> labels_right = metrics.labels; + labels_right.emplace_back("channel", "right"); + + global_metrics.add("bus_current_level_dbfs", labels_left, &metrics.current_level_dbfs[0], Metrics::TYPE_GAUGE); + global_metrics.add("bus_current_level_dbfs", labels_right, &metrics.current_level_dbfs[1], Metrics::TYPE_GAUGE); + global_metrics.add("bus_peak_level_dbfs", labels_left, &metrics.peak_level_dbfs[0], Metrics::TYPE_GAUGE); + global_metrics.add("bus_peak_level_dbfs", labels_right, &metrics.peak_level_dbfs[1], Metrics::TYPE_GAUGE); + global_metrics.add("bus_historic_peak_dbfs", metrics.labels, &metrics.historic_peak_dbfs, Metrics::TYPE_GAUGE); + global_metrics.add("bus_gain_staging_db", metrics.labels, &metrics.gain_staging_db, Metrics::TYPE_GAUGE); + global_metrics.add("bus_compressor_attenuation_db", metrics.labels, &metrics.compressor_attenuation_db, Metrics::TYPE_GAUGE); + } + // Reset resamplers for all cards that don't have the exact same state as before. for (unsigned card_index = 0; card_index < MAX_VIDEO_CARDS; ++card_index) { const DeviceSpec device_spec{InputSourceType::CAPTURE_CARD, card_index}; @@ -981,6 +1102,14 @@ void AudioMixer::set_input_mapping_lock_held(const InputMapping &new_input_mappi reset_resampler_mutex_held(device_spec); } } + for (unsigned card_index = 0; card_index < num_ffmpeg_inputs; ++card_index) { + const DeviceSpec device_spec{InputSourceType::FFMPEG_VIDEO_INPUT, card_index}; + AudioDevice *device = find_audio_device(device_spec); + if (device->interesting_channels != interesting_channels[device_spec]) { + device->interesting_channels = interesting_channels[device_spec]; + reset_resampler_mutex_held(device_spec); + } + } input_mapping = new_input_mapping; }