X-Git-Url: https://git.sesse.net/?a=blobdiff_plain;f=audio_mixer.cpp;h=769dabfa74a0831cb6678a52f15d245d2c4ca6f2;hb=22bb590d54107446f09d0cb60f32e610cc9ed7ad;hp=764d7a2217fbbdfa6a7122960beb29668b5e207d;hpb=2a30d1f3a428511dab64587bd7615bc528a2573b;p=nageru diff --git a/audio_mixer.cpp b/audio_mixer.cpp index 764d7a2..769dabf 100644 --- a/audio_mixer.cpp +++ b/audio_mixer.cpp @@ -4,6 +4,7 @@ #include #include #include +#include #include #include "db.h" @@ -12,34 +13,91 @@ using namespace bmusb; using namespace std; +using namespace std::placeholders; namespace { -void convert_fixed24_to_fp32(float *dst, size_t out_channels, const uint8_t *src, size_t in_channels, size_t num_samples) +// TODO: If these prove to be a bottleneck, they can be SSSE3-optimized +// (usually including multiple channels at a time). + +void convert_fixed16_to_fp32(float *dst, size_t out_channel, size_t out_num_channels, + const uint8_t *src, size_t in_channel, size_t in_num_channels, + size_t num_samples) { - assert(in_channels >= out_channels); + assert(in_channel < in_num_channels); + assert(out_channel < out_num_channels); + src += in_channel * 2; + dst += out_channel; + for (size_t i = 0; i < num_samples; ++i) { - for (size_t j = 0; j < out_channels; ++j) { - uint32_t s1 = *src++; - uint32_t s2 = *src++; - uint32_t s3 = *src++; - uint32_t s = s1 | (s1 << 8) | (s2 << 16) | (s3 << 24); - dst[i * out_channels + j] = int(s) * (1.0f / 2147483648.0f); - } - src += 3 * (in_channels - out_channels); + int16_t s = le16toh(*(int16_t *)src); + *dst = s * (1.0f / 32768.0f); + + src += 2 * in_num_channels; + dst += out_num_channels; } } -void convert_fixed32_to_fp32(float *dst, size_t out_channels, const uint8_t *src, size_t in_channels, size_t num_samples) +void convert_fixed24_to_fp32(float *dst, size_t out_channel, size_t out_num_channels, + const uint8_t *src, size_t in_channel, size_t in_num_channels, + size_t num_samples) { - assert(in_channels >= out_channels); + assert(in_channel < in_num_channels); + assert(out_channel < out_num_channels); + src += in_channel * 3; + dst += out_channel; + for (size_t i = 0; i < num_samples; ++i) { - for (size_t j = 0; j < out_channels; ++j) { - int32_t s = le32toh(*(int32_t *)src); - dst[i * out_channels + j] = s * (1.0f / 2147483648.0f); - src += 4; - } - src += 4 * (in_channels - out_channels); + uint32_t s1 = src[0]; + uint32_t s2 = src[1]; + uint32_t s3 = src[2]; + uint32_t s = s1 | (s1 << 8) | (s2 << 16) | (s3 << 24); + *dst = int(s) * (1.0f / 2147483648.0f); + + src += 3 * in_num_channels; + dst += out_num_channels; + } +} + +void convert_fixed32_to_fp32(float *dst, size_t out_channel, size_t out_num_channels, + const uint8_t *src, size_t in_channel, size_t in_num_channels, + size_t num_samples) +{ + assert(in_channel < in_num_channels); + assert(out_channel < out_num_channels); + src += in_channel * 4; + dst += out_channel; + + for (size_t i = 0; i < num_samples; ++i) { + int32_t s = le32toh(*(int32_t *)src); + *dst = s * (1.0f / 2147483648.0f); + + src += 4 * in_num_channels; + dst += out_num_channels; + } +} + +float find_peak(const float *samples, size_t num_samples) +{ + float m = fabs(samples[0]); + for (size_t i = 1; i < num_samples; ++i) { + m = max(m, fabs(samples[i])); + } + return m; +} + +void deinterleave_samples(const vector &in, vector *out_l, vector *out_r) +{ + size_t num_samples = in.size() / 2; + out_l->resize(num_samples); + out_r->resize(num_samples); + + const float *inptr = in.data(); + float *lptr = &(*out_l)[0]; + float *rptr = &(*out_r)[0]; + for (size_t i = 0; i < num_samples; ++i) { + *lptr++ = *inptr++; + *rptr++ = *inptr++; } } @@ -49,7 +107,8 @@ AudioMixer::AudioMixer(unsigned num_cards) : num_cards(num_cards), level_compressor(OUTPUT_FREQUENCY), limiter(OUTPUT_FREQUENCY), - compressor(OUTPUT_FREQUENCY) + compressor(OUTPUT_FREQUENCY), + correlation(OUTPUT_FREQUENCY) { locut.init(FILTER_HPF, 2); @@ -59,88 +118,257 @@ AudioMixer::AudioMixer(unsigned num_cards) set_compressor_enabled(global_flags.compressor_enabled); set_limiter_enabled(global_flags.limiter_enabled); set_final_makeup_gain_auto(global_flags.final_makeup_gain_auto); + + // Generate a very simple, default input mapping. + InputMapping::Bus input; + input.name = "Main"; + input.device.type = InputSourceType::CAPTURE_CARD; + input.device.index = 0; + input.source_channel[0] = 0; + input.source_channel[1] = 1; + + InputMapping new_input_mapping; + new_input_mapping.buses.push_back(input); + set_input_mapping(new_input_mapping); + + // Look for ALSA cards. + available_alsa_cards = ALSAInput::enumerate_devices(); + + 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); } -void AudioMixer::reset_card(unsigned card_index) +AudioMixer::~AudioMixer() { - CaptureCard *card = &cards[card_index]; + for (unsigned card_index = 0; card_index < available_alsa_cards.size(); ++card_index) { + const AudioDevice &device = alsa_inputs[card_index]; + if (device.alsa_device != nullptr) { + device.alsa_device->stop_capture_thread(); + } + } +} + - unique_lock lock(card->audio_mutex); - card->resampling_queue.reset(new ResamplingQueue(card_index, OUTPUT_FREQUENCY, OUTPUT_FREQUENCY, 2)); - card->next_local_pts = 0; +void AudioMixer::reset_resampler(DeviceSpec device_spec) +{ + lock_guard lock(audio_mutex); + reset_resampler_mutex_held(device_spec); } -void AudioMixer::add_audio(unsigned card_index, const uint8_t *data, unsigned num_samples, AudioFormat audio_format, int64_t frame_length) +void AudioMixer::reset_resampler_mutex_held(DeviceSpec device_spec) { - CaptureCard *card = &cards[card_index]; + AudioDevice *device = find_audio_device(device_spec); - // Convert the audio to stereo fp32. - vector audio; - audio.resize(num_samples * 2); - switch (audio_format.bits_per_sample) { - case 0: - assert(num_samples == 0); - break; - case 24: - convert_fixed24_to_fp32(&audio[0], 2, data, audio_format.num_channels, num_samples); - break; - case 32: - convert_fixed32_to_fp32(&audio[0], 2, data, audio_format.num_channels, num_samples); - break; - default: - fprintf(stderr, "Cannot handle audio with %u bits per sample\n", audio_format.bits_per_sample); - assert(false); + 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->next_local_pts = 0; +} - // Now add it. - { - unique_lock lock(card->audio_mutex); +void AudioMixer::reset_alsa_mutex_held(DeviceSpec device_spec) +{ + assert(device_spec.type == InputSourceType::ALSA_INPUT); + unsigned card_index = device_spec.index; + AudioDevice *device = find_audio_device(device_spec); - int64_t local_pts = card->next_local_pts; - card->resampling_queue->add_input_samples(local_pts / double(TIMEBASE), audio.data(), num_samples); - card->next_local_pts = local_pts + frame_length; + if (device->alsa_device != nullptr) { + device->alsa_device->stop_capture_thread(); } + if (device->interesting_channels.empty()) { + device->alsa_device.reset(); + } else { + const ALSAInput::Device &alsa_dev = available_alsa_cards[card_index]; + device->alsa_device.reset(new ALSAInput(alsa_dev.address.c_str(), OUTPUT_FREQUENCY, alsa_dev.num_channels, bind(&AudioMixer::add_audio, this, device_spec, _1, _2, _3, _4))); + device->capture_frequency = device->alsa_device->get_sample_rate(); + device->alsa_device->start_capture_thread(); + } +} + +bool AudioMixer::add_audio(DeviceSpec device_spec, const uint8_t *data, unsigned num_samples, AudioFormat audio_format, int64_t frame_length) +{ + AudioDevice *device = find_audio_device(device_spec); + + unique_lock lock(audio_mutex, defer_lock); + if (!lock.try_lock_for(chrono::milliseconds(10))) { + return false; + } + if (device->resampling_queue == nullptr) { + // No buses use this device; throw it away. + return true; + } + + unsigned num_channels = device->interesting_channels.size(); + assert(num_channels > 0); + + // Convert the audio to fp32. + vector audio; + audio.resize(num_samples * num_channels); + unsigned channel_index = 0; + for (auto channel_it = device->interesting_channels.cbegin(); channel_it != device->interesting_channels.end(); ++channel_it, ++channel_index) { + switch (audio_format.bits_per_sample) { + case 0: + assert(num_samples == 0); + break; + case 16: + convert_fixed16_to_fp32(&audio[0], channel_index, num_channels, data, *channel_it, audio_format.num_channels, num_samples); + break; + case 24: + convert_fixed24_to_fp32(&audio[0], channel_index, num_channels, data, *channel_it, audio_format.num_channels, num_samples); + break; + case 32: + convert_fixed32_to_fp32(&audio[0], channel_index, num_channels, data, *channel_it, audio_format.num_channels, num_samples); + break; + default: + fprintf(stderr, "Cannot handle audio with %u bits per sample\n", audio_format.bits_per_sample); + assert(false); + } + } + + // Now add it. + int64_t local_pts = device->next_local_pts; + device->resampling_queue->add_input_samples(local_pts / double(TIMEBASE), audio.data(), num_samples); + device->next_local_pts = local_pts + frame_length; + return true; } -void AudioMixer::add_silence(unsigned card_index, unsigned samples_per_frame, unsigned num_frames, int64_t frame_length) +bool AudioMixer::add_silence(DeviceSpec device_spec, unsigned samples_per_frame, unsigned num_frames, int64_t frame_length) { - CaptureCard *card = &cards[card_index]; - unique_lock lock(card->audio_mutex); + AudioDevice *device = find_audio_device(device_spec); + + unique_lock lock(audio_mutex, defer_lock); + if (!lock.try_lock_for(chrono::milliseconds(10))) { + return false; + } + if (device->resampling_queue == nullptr) { + // No buses use this device; throw it away. + return true; + } + + unsigned num_channels = device->interesting_channels.size(); + assert(num_channels > 0); - vector silence(samples_per_frame * 2, 0.0f); + vector silence(samples_per_frame * num_channels, 0.0f); for (unsigned i = 0; i < num_frames; ++i) { - card->resampling_queue->add_input_samples(card->next_local_pts / double(TIMEBASE), silence.data(), samples_per_frame); + 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. - card->next_local_pts += frame_length; + device->next_local_pts += frame_length; + } + return true; +} + +AudioMixer::AudioDevice *AudioMixer::find_audio_device(DeviceSpec device) +{ + switch (device.type) { + case InputSourceType::CAPTURE_CARD: + return &video_cards[device.index]; + case InputSourceType::ALSA_INPUT: + return &alsa_inputs[device.index]; + case InputSourceType::SILENCE: + default: + assert(false); + } + return nullptr; +} + +// Get a pointer to the given channel from the given device. +// The channel must be picked out earlier and resampled. +void AudioMixer::find_sample_src_from_device(const map> &samples_card, DeviceSpec device_spec, int source_channel, const float **srcptr, unsigned *stride) +{ + static float zero = 0.0f; + if (source_channel == -1 || device_spec.type == InputSourceType::SILENCE) { + *srcptr = &zero; + *stride = 0; + return; + } + AudioDevice *device = find_audio_device(device_spec); + assert(device->interesting_channels.count(source_channel) != 0); + unsigned channel_index = 0; + for (int channel : device->interesting_channels) { + if (channel == source_channel) break; + ++channel_index; + } + assert(channel_index < device->interesting_channels.size()); + const auto it = samples_card.find(device_spec); + assert(it != samples_card.end()); + *srcptr = &(it->second)[channel_index]; + *stride = device->interesting_channels.size(); +} + +// TODO: Can be SSSE3-optimized if need be. +void AudioMixer::fill_audio_bus(const map> &samples_card, const InputMapping::Bus &bus, unsigned num_samples, float *output) +{ + if (bus.device.type == InputSourceType::SILENCE) { + memset(output, 0, num_samples * sizeof(*output)); + } else { + assert(bus.device.type == InputSourceType::CAPTURE_CARD || + bus.device.type == InputSourceType::ALSA_INPUT); + const float *lsrc, *rsrc; + unsigned lstride, rstride; + float *dptr = output; + find_sample_src_from_device(samples_card, bus.device, bus.source_channel[0], &lsrc, &lstride); + find_sample_src_from_device(samples_card, bus.device, bus.source_channel[1], &rsrc, &rstride); + for (unsigned i = 0; i < num_samples; ++i) { + *dptr++ = *lsrc; + *dptr++ = *rsrc; + lsrc += lstride; + rsrc += rstride; + } } } vector AudioMixer::get_output(double pts, unsigned num_samples, ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy) { - vector samples_card; - vector samples_out; - samples_out.resize(num_samples * 2); + map> samples_card; + vector samples_bus; - // TODO: Allow more flexible input mapping. - for (unsigned card_index = 0; card_index < num_cards; ++card_index) { - samples_card.resize(num_samples * 2); - { - unique_lock lock(cards[card_index].audio_mutex); - cards[card_index].resampling_queue->get_output_samples( + lock_guard lock(audio_mutex); + + // Pick out all the interesting channels from all the cards. + // TODO: If the card has been hotswapped, the number of channels + // might have changed; if so, we need to do some sort of remapping + // to silence. + for (const auto &spec_and_info : get_devices_mutex_held()) { + const DeviceSpec &device_spec = spec_and_info.first; + AudioDevice *device = find_audio_device(device_spec); + if (!device->interesting_channels.empty()) { + samples_card[device_spec].resize(num_samples * device->interesting_channels.size()); + device->resampling_queue->get_output_samples( pts, - &samples_card[0], + &samples_card[device_spec][0], num_samples, rate_adjustment_policy); } - if (card_index == 0) { + } + + // TODO: Move lo-cut etc. into each bus. + vector samples_out, left, right; + samples_out.resize(num_samples * 2); + samples_bus.resize(num_samples * 2); + for (unsigned bus_index = 0; bus_index < input_mapping.buses.size(); ++bus_index) { + fill_audio_bus(samples_card, input_mapping.buses[bus_index], num_samples, &samples_bus[0]); + + // TODO: We should measure post-fader. + deinterleave_samples(samples_bus, &left, &right); + measure_bus_levels(bus_index, left, right); + + float volume = from_db(fader_volume_db[bus_index]); + if (bus_index == 0) { for (unsigned i = 0; i < num_samples * 2; ++i) { - samples_out[i] = samples_card[i]; + samples_out[i] = samples_bus[i] * volume; } } else { for (unsigned i = 0; i < num_samples * 2; ++i) { - samples_out[i] += samples_card[i]; + samples_out[i] += samples_bus[i] * volume; } } } @@ -154,7 +382,7 @@ vector AudioMixer::get_output(double pts, unsigned num_samples, Resamplin } { - unique_lock lock(compressor_mutex); + lock_guard lock(compressor_mutex); // 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 @@ -225,11 +453,11 @@ vector AudioMixer::get_output(double pts, unsigned num_samples, Resamplin // something we get out per-sample. // // Note that there's a feedback loop here, so we choose a very slow filter - // (half-time of 100 seconds). + // (half-time of 30 seconds). double target_loudness_factor, alpha; - double loudness_lu = loudness_lufs - ref_level_lufs; + double loudness_lu = r128.loudness_M() - ref_level_lufs; double current_makeup_lu = to_db(final_makeup_gain); - target_loudness_factor = from_db(-loudness_lu); + target_loudness_factor = final_makeup_gain * from_db(-loudness_lu); // If we're outside +/- 5 LU uncorrected, we don't count it as // a normal signal (probably silence) and don't change the @@ -239,13 +467,13 @@ vector AudioMixer::get_output(double pts, unsigned num_samples, Resamplin } else { // Formula adapted from // https://en.wikipedia.org/wiki/Low-pass_filter#Simple_infinite_impulse_response_filter. - const double half_time_s = 100.0; + const double half_time_s = 30.0; const double fc_mul_2pi_delta_t = 1.0 / (half_time_s * OUTPUT_FREQUENCY); alpha = fc_mul_2pi_delta_t / (fc_mul_2pi_delta_t + 1.0); } { - unique_lock lock(compressor_mutex); + lock_guard lock(compressor_mutex); double m = final_makeup_gain; for (size_t i = 0; i < samples_out.size(); i += 2) { samples_out[i + 0] *= m; @@ -255,5 +483,170 @@ vector AudioMixer::get_output(double pts, unsigned num_samples, Resamplin final_makeup_gain = m; } + update_meters(samples_out); + return samples_out; } + +void AudioMixer::measure_bus_levels(unsigned bus_index, const vector &left, const vector &right) +{ + const float *ptrs[] = { left.data(), right.data() }; + { + lock_guard lock(audio_measure_mutex); + bus_r128[bus_index]->process(left.size(), const_cast(ptrs)); + } +} + +void AudioMixer::update_meters(const vector &samples) +{ + // Upsample 4x to find interpolated peak. + peak_resampler.inp_data = const_cast(samples.data()); + peak_resampler.inp_count = samples.size() / 2; + + vector interpolated_samples; + interpolated_samples.resize(samples.size()); + { + lock_guard lock(audio_measure_mutex); + + while (peak_resampler.inp_count > 0) { // About four iterations. + peak_resampler.out_data = &interpolated_samples[0]; + peak_resampler.out_count = interpolated_samples.size() / 2; + peak_resampler.process(); + size_t out_stereo_samples = interpolated_samples.size() / 2 - peak_resampler.out_count; + peak = max(peak, find_peak(interpolated_samples.data(), out_stereo_samples * 2)); + peak_resampler.out_data = nullptr; + } + } + + // Find R128 levels and L/R correlation. + vector left, right; + deinterleave_samples(samples, &left, &right); + float *ptrs[] = { left.data(), right.data() }; + { + lock_guard lock(audio_measure_mutex); + r128.process(left.size(), ptrs); + correlation.process_samples(samples); + } + + send_audio_level_callback(); +} + +void AudioMixer::reset_meters() +{ + lock_guard lock(audio_measure_mutex); + peak_resampler.reset(); + peak = 0.0f; + r128.reset(); + r128.integr_start(); + correlation.reset(); +} + +void AudioMixer::send_audio_level_callback() +{ + if (audio_level_callback == nullptr) { + return; + } + + lock_guard lock(audio_measure_mutex); + double loudness_s = r128.loudness_S(); + double loudness_i = r128.integrated(); + double loudness_range_low = r128.range_min(); + double loudness_range_high = r128.range_max(); + + vector bus_loudness; + bus_loudness.resize(input_mapping.buses.size()); + for (unsigned bus_index = 0; bus_index < bus_r128.size(); ++bus_index) { + bus_loudness[bus_index] = bus_r128[bus_index]->loudness_S(); + } + + audio_level_callback(loudness_s, to_db(peak), bus_loudness, + loudness_i, loudness_range_low, loudness_range_high, + gain_staging_db, + to_db(final_makeup_gain), + correlation.get_correlation()); +} + +map AudioMixer::get_devices() const +{ + lock_guard lock(audio_mutex); + return get_devices_mutex_held(); +} + +map AudioMixer::get_devices_mutex_held() const +{ + map devices; + for (unsigned card_index = 0; card_index < num_cards; ++card_index) { + const DeviceSpec spec{ InputSourceType::CAPTURE_CARD, card_index }; + const AudioDevice *device = &video_cards[card_index]; + DeviceInfo info; + info.name = device->name; + info.num_channels = 8; // FIXME: This is wrong for fake cards. + devices.insert(make_pair(spec, info)); + } + for (unsigned card_index = 0; card_index < available_alsa_cards.size(); ++card_index) { + const DeviceSpec spec{ InputSourceType::ALSA_INPUT, card_index }; + const ALSAInput::Device &device = available_alsa_cards[card_index]; + DeviceInfo info; + info.name = device.name + " (" + device.info + ")"; + info.num_channels = device.num_channels; + devices.insert(make_pair(spec, info)); + } + return devices; +} + +void AudioMixer::set_name(DeviceSpec device_spec, const string &name) +{ + AudioDevice *device = find_audio_device(device_spec); + + lock_guard lock(audio_mutex); + device->name = name; +} + +void AudioMixer::set_input_mapping(const InputMapping &new_input_mapping) +{ + lock_guard lock(audio_mutex); + + 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) { + for (unsigned channel = 0; channel < 2; ++channel) { + if (bus.source_channel[channel] != -1) { + interesting_channels[bus.device].insert(bus.source_channel[channel]); + } + } + } + } + + // Reset resamplers for all cards that don't have the exact same state as before. + for (const auto &spec_and_info : get_devices_mutex_held()) { + const DeviceSpec &device_spec = spec_and_info.first; + AudioDevice *device = find_audio_device(device_spec); + if (device->interesting_channels != interesting_channels[device_spec]) { + device->interesting_channels = interesting_channels[device_spec]; + if (device_spec.type == InputSourceType::ALSA_INPUT) { + reset_alsa_mutex_held(device_spec); + } + reset_resampler_mutex_held(device_spec); + } + } + + { + lock_guard lock(audio_measure_mutex); + bus_r128.resize(new_input_mapping.buses.size()); + for (unsigned bus_index = 0; bus_index < bus_r128.size(); ++bus_index) { + if (bus_r128[bus_index] == nullptr) { + bus_r128[bus_index].reset(new Ebu_r128_proc); + } + bus_r128[bus_index]->init(2, OUTPUT_FREQUENCY); + } + } + + input_mapping = new_input_mapping; +} + +InputMapping AudioMixer::get_input_mapping() const +{ + lock_guard lock(audio_mutex); + return input_mapping; +}