#include "audio_mixer.h"
#include <assert.h>
-#include <endian.h>
#include <bmusb/bmusb.h>
-#include <stdio.h>
#include <endian.h>
-#include <cmath>
-#ifdef __SSE__
+#include <math.h>
+#ifdef __SSE2__
#include <immintrin.h>
#endif
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <algorithm>
+#include <chrono>
+#include <cmath>
+#include <cstddef>
+#include <limits>
+#include <utility>
#include "db.h"
#include "flags.h"
-#include "mixer.h"
#include "state.pb.h"
#include "timebase.h"
using namespace bmusb;
using namespace std;
+using namespace std::chrono;
using namespace std::placeholders;
namespace {
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);
}
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();
- InputMapping new_input_mapping;
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(),
global_flags.input_mapping_filename,
&new_input_mapping)) {
global_flags.input_mapping_filename.c_str());
exit(1);
}
+ set_input_mapping(new_input_mapping);
} else {
- // 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;
-
- new_input_mapping.buses.push_back(input);
+ set_simple_input(/*card_index=*/0);
+ if (global_flags.multichannel_mapping_mode) {
+ current_mapping_mode = MappingMode::MULTICHANNEL;
+ }
}
- set_input_mapping(new_input_mapping);
-
- 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_resampler(DeviceSpec device_spec)
} 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.index, 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);
}
}
+ // 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;
}
vector<float> 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;
}
{
BusSettings settings;
settings.fader_volume_db = 0.0f;
+ settings.muted = false;
settings.locut_enabled = global_flags.locut_enabled;
for (unsigned band_index = 0; band_index < NUM_EQ_BANDS; ++band_index) {
settings.eq_level_db[band_index] = 0.0f;
lock_guard<timed_mutex> lock(audio_mutex);
BusSettings settings;
settings.fader_volume_db = fader_volume_db[bus_index];
+ settings.muted = mute[bus_index];
settings.locut_enabled = locut_enabled[bus_index];
for (unsigned band_index = 0; band_index < NUM_EQ_BANDS; ++band_index) {
settings.eq_level_db[band_index] = eq_level_db[bus_index][band_index];
{
lock_guard<timed_mutex> lock(audio_mutex);
fader_volume_db[bus_index] = settings.fader_volume_db;
+ mute[bus_index] = settings.muted;
locut_enabled[bus_index] = settings.locut_enabled;
for (unsigned band_index = 0; band_index < NUM_EQ_BANDS; ++band_index) {
eq_level_db[bus_index][band_index] = settings.eq_level_db[band_index];
}
gain_staging_db[bus_index] = settings.gain_staging_db;
+ last_gain_staging_db[bus_index] = gain_staging_db[bus_index];
level_compressor_enabled[bus_index] = settings.level_compressor_enabled;
compressor_threshold_dbfs[bus_index] = settings.compressor_threshold_dbfs;
compressor_enabled[bus_index] = settings.compressor_enabled;
void AudioMixer::fill_audio_bus(const map<DeviceSpec, vector<float>> &samples_card, const InputMapping::Bus &bus, unsigned num_samples, float *output)
{
if (bus.device.type == InputSourceType::SILENCE) {
- memset(output, 0, num_samples * sizeof(*output));
+ memset(output, 0, num_samples * 2 * sizeof(*output));
} else {
assert(bus.device.type == InputSourceType::CAPTURE_CARD ||
bus.device.type == InputSourceType::ALSA_INPUT);
return ret;
}
-vector<float> AudioMixer::get_output(double pts, unsigned num_samples, ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy)
+namespace {
+
+void apply_gain(float db, float last_db, vector<float> *samples)
+{
+ if (fabs(db - last_db) < 1e-3) {
+ // Constant over this frame.
+ const float gain = from_db(db);
+ for (size_t i = 0; i < samples->size(); ++i) {
+ (*samples)[i] *= gain;
+ }
+ } else {
+ // We need to do a fade.
+ unsigned num_samples = samples->size() / 2;
+ float gain = from_db(last_db);
+ const float gain_inc = pow(from_db(db - last_db), 1.0 / num_samples);
+ for (size_t i = 0; i < num_samples; ++i) {
+ (*samples)[i * 2 + 0] *= gain;
+ (*samples)[i * 2 + 1] *= gain;
+ gain *= gain_inc;
+ }
+ }
+}
+
+} // namespace
+
+vector<float> AudioMixer::get_output(steady_clock::time_point ts, unsigned num_samples, ResamplingQueue::RateAdjustmentPolicy rate_adjustment_policy)
{
map<DeviceSpec, vector<float>> samples_card;
vector<float> samples_bus;
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);
gain_staging_db[bus_index] = to_db(level_compressor[bus_index]->get_attenuation() * makeup_gain);
} else {
// Just apply the gain we already had.
- float g = from_db(gain_staging_db[bus_index]);
- for (size_t i = 0; i < samples_bus.size(); ++i) {
- samples_bus[i] *= g;
- }
+ float db = gain_staging_db[bus_index];
+ float last_db = last_gain_staging_db[bus_index];
+ apply_gain(db, last_db, &samples_bus);
}
+ last_gain_staging_db[bus_index] = gain_staging_db[bus_index];
#if 0
printf("level=%f (%+5.2f dBFS) attenuation=%f (%+5.2f dB) end_result=%+5.2f dB\n",
// (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
return samples_out;
}
+namespace {
+
+void apply_filter_fade(StereoFilter *filter, float *data, unsigned num_samples, float cutoff_hz, float db, float last_db)
+{
+ // A granularity of 32 samples is an okay tradeoff between speed and
+ // smoothness; recalculating the filters is pretty expensive, so it's
+ // good that we don't do this all the time.
+ static constexpr unsigned filter_granularity_samples = 32;
+
+ const float cutoff_linear = cutoff_hz * 2.0 * M_PI / OUTPUT_FREQUENCY;
+ if (fabs(db - last_db) < 1e-3) {
+ // Constant over this frame.
+ if (fabs(db) > 0.01f) {
+ filter->render(data, num_samples, cutoff_linear, 0.5f, db / 40.0f);
+ }
+ } else {
+ // We need to do a fade. (Rounding up avoids division by zero.)
+ unsigned num_blocks = (num_samples + filter_granularity_samples - 1) / filter_granularity_samples;
+ const float inc_db_norm = (db - last_db) / 40.0f / num_blocks;
+ float db_norm = db / 40.0f;
+ for (size_t i = 0; i < num_samples; i += filter_granularity_samples) {
+ size_t samples_this_block = std::min<size_t>(num_samples - i, filter_granularity_samples);
+ filter->render(data + i * 2, samples_this_block, cutoff_linear, 0.5f, db_norm);
+ db_norm += inc_db_norm;
+ }
+ }
+}
+
+} // namespace
+
void AudioMixer::apply_eq(unsigned bus_index, vector<float> *samples_bus)
{
constexpr float bass_freq_hz = 200.0f;
// set the mid-level filter, and then offset the low and high bands
// from that if we need to. (We could perhaps have folded the gain into
// the next part, but it's so cheap that the trouble isn't worth it.)
- if (fabs(eq_level_db[bus_index][EQ_BAND_MID]) > 0.01f) {
- float g = from_db(eq_level_db[bus_index][EQ_BAND_MID]);
- for (size_t i = 0; i < samples_bus->size(); ++i) {
- (*samples_bus)[i] *= g;
- }
- }
+ //
+ // If any part of the EQ has changed appreciably since last frame,
+ // we fade smoothly during the course of this frame.
+ const float bass_db = eq_level_db[bus_index][EQ_BAND_BASS];
+ const float mid_db = eq_level_db[bus_index][EQ_BAND_MID];
+ const float treble_db = eq_level_db[bus_index][EQ_BAND_TREBLE];
- float bass_adj_db = eq_level_db[bus_index][EQ_BAND_BASS] - eq_level_db[bus_index][EQ_BAND_MID];
- if (fabs(bass_adj_db) > 0.01f) {
- eq[bus_index][EQ_BAND_BASS].render(samples_bus->data(), samples_bus->size() / 2,
- bass_freq_hz * 2.0 * M_PI / OUTPUT_FREQUENCY, 0.5f, bass_adj_db / 40.0f);
- }
+ const float last_bass_db = last_eq_level_db[bus_index][EQ_BAND_BASS];
+ const float last_mid_db = last_eq_level_db[bus_index][EQ_BAND_MID];
+ const float last_treble_db = last_eq_level_db[bus_index][EQ_BAND_TREBLE];
- float treble_adj_db = eq_level_db[bus_index][EQ_BAND_TREBLE] - eq_level_db[bus_index][EQ_BAND_MID];
- if (fabs(treble_adj_db) > 0.01f) {
- eq[bus_index][EQ_BAND_TREBLE].render(samples_bus->data(), samples_bus->size() / 2,
- treble_freq_hz * 2.0 * M_PI / OUTPUT_FREQUENCY, 0.5f, treble_adj_db / 40.0f);
- }
+ assert(samples_bus->size() % 2 == 0);
+ const unsigned num_samples = samples_bus->size() / 2;
+
+ apply_gain(mid_db, last_mid_db, samples_bus);
+
+ apply_filter_fade(&eq[bus_index][EQ_BAND_BASS], samples_bus->data(), num_samples, bass_freq_hz, bass_db - mid_db, last_bass_db - last_mid_db);
+ apply_filter_fade(&eq[bus_index][EQ_BAND_TREBLE], samples_bus->data(), num_samples, treble_freq_hz, treble_db - mid_db, last_treble_db - last_mid_db);
+
+ last_eq_level_db[bus_index][EQ_BAND_BASS] = bass_db;
+ last_eq_level_db[bus_index][EQ_BAND_MID] = mid_db;
+ last_eq_level_db[bus_index][EQ_BAND_TREBLE] = treble_db;
}
void AudioMixer::add_bus_to_master(unsigned bus_index, const vector<float> &samples_bus, vector<float> *samples_out)
assert(samples_bus.size() == samples_out->size());
assert(samples_bus.size() % 2 == 0);
unsigned num_samples = samples_bus.size() / 2;
- if (fabs(fader_volume_db[bus_index] - last_fader_volume_db[bus_index]) > 1e-3) {
+ const float new_volume_db = mute[bus_index] ? -90.0f : fader_volume_db[bus_index].load();
+ if (fabs(new_volume_db - last_fader_volume_db[bus_index]) > 1e-3) {
// The volume has changed; do a fade over the course of this frame.
// (We might have some numerical issues here, but it seems to sound OK.)
// For the purpose of fading here, the silence floor is set to -90 dB
// (the fader only goes to -84).
float old_volume = from_db(max<float>(last_fader_volume_db[bus_index], -90.0f));
- float volume = from_db(max<float>(fader_volume_db[bus_index], -90.0f));
+ float volume = from_db(max<float>(new_volume_db, -90.0f));
float volume_inc = pow(volume / old_volume, 1.0 / num_samples);
volume = old_volume;
volume *= volume_inc;
}
}
- } else {
- float volume = from_db(fader_volume_db[bus_index]);
+ } else if (new_volume_db > -90.0f) {
+ float volume = from_db(new_volume_db);
if (bus_index == 0) {
for (unsigned i = 0; i < num_samples; ++i) {
(*samples_out)[i * 2 + 0] = samples_bus[i * 2 + 0] * volume;
}
}
- last_fader_volume_db[bus_index] = fader_volume_db[bus_index];
+ last_fader_volume_db[bus_index] = new_volume_db;
}
void AudioMixer::measure_bus_levels(unsigned bus_index, const vector<float> &left, const vector<float> &right)
{
assert(left.size() == right.size());
- const float volume = from_db(fader_volume_db[bus_index]);
+ const float volume = mute[bus_index] ? 0.0f : from_db(fader_volume_db[bus_index]);
const float peak_levels[2] = {
find_peak(left.data(), left.size()) * volume,
find_peak(right.data(), right.size()) * volume
}
}
+void AudioMixer::set_simple_input(unsigned card_index)
+{
+ InputMapping new_input_mapping;
+ InputMapping::Bus input;
+ input.name = "Main";
+ input.device.type = InputSourceType::CAPTURE_CARD;
+ input.device.index = card_index;
+ input.source_channel[0] = 0;
+ input.source_channel[1] = 1;
+
+ new_input_mapping.buses.push_back(input);
+
+ lock_guard<timed_mutex> lock(audio_mutex);
+ current_mapping_mode = MappingMode::SIMPLE;
+ set_input_mapping_lock_held(new_input_mapping);
+ fader_volume_db[0] = 0.0f;
+}
+
+unsigned AudioMixer::get_simple_input() const
+{
+ lock_guard<timed_mutex> lock(audio_mutex);
+ if (input_mapping.buses.size() == 1 &&
+ input_mapping.buses[0].device.type == InputSourceType::CAPTURE_CARD &&
+ input_mapping.buses[0].source_channel[0] == 0 &&
+ input_mapping.buses[0].source_channel[1] == 1) {
+ return input_mapping.buses[0].device.index;
+ } else {
+ return numeric_limits<unsigned>::max();
+ }
+}
+
void AudioMixer::set_input_mapping(const InputMapping &new_input_mapping)
{
lock_guard<timed_mutex> lock(audio_mutex);
+ set_input_mapping_lock_held(new_input_mapping);
+ current_mapping_mode = MappingMode::MULTICHANNEL;
+}
+
+AudioMixer::MappingMode AudioMixer::get_mapping_mode() const
+{
+ lock_guard<timed_mutex> lock(audio_mutex);
+ return current_mapping_mode;
+}
+void AudioMixer::set_input_mapping_lock_held(const InputMapping &new_input_mapping)
+{
map<DeviceSpec, set<unsigned>> interesting_channels;
for (const InputMapping::Bus &bus : new_input_mapping.buses) {
if (bus.device.type == InputSourceType::CAPTURE_CARD ||
return input_mapping;
}
+unsigned AudioMixer::num_buses() const
+{
+ lock_guard<timed_mutex> lock(audio_mutex);
+ return input_mapping.buses.size();
+}
+
void AudioMixer::reset_peak(unsigned bus_index)
{
lock_guard<timed_mutex> lock(audio_mutex);