New module for the adaptive level controlling functionality in the audio processing module

webrtc/modules/audio_processing/level_controller/level_controller.cc

Index: webrtc/modules/audio_processing/level_controller/level_controller.cc
diff --git a/webrtc/modules/audio_processing/level_controller/level_controller.cc b/webrtc/modules/audio_processing/level_controller/level_controller.cc
new file mode 100644
index 0000000000000000000000000000000000000000..22114a912b349b97439e303171273b6609304fc0
--- /dev/null
+++ b/webrtc/modules/audio_processing/level_controller/level_controller.cc
@@ -0,0 +1,151 @@
+/*
+ *  Copyright (c) 2016 The WebRTC project authors. All Rights Reserved.
+ *
+ *  Use of this source code is governed by a BSD-style license
+ *  that can be found in the LICENSE file in the root of the source
+ *  tree. An additional intellectual property rights grant can be found
+ *  in the file PATENTS.  All contributing project authors may
+ *  be found in the AUTHORS file in the root of the source tree.
+ */
+
+#include "webrtc/modules/audio_processing/level_controller/level_controller.h"
+
+#include <algorithm>
+#include <numeric>
+
+#include "webrtc/base/array_view.h"
+#include "webrtc/base/checks.h"
+#include "webrtc/modules/audio_processing/audio_buffer.h"
+#include "webrtc/modules/audio_processing/level_controller/gain_applier.h"
+#include "webrtc/modules/audio_processing/level_controller/gain_selector.h"
+#include "webrtc/modules/audio_processing/level_controller/noise_level_estimator.h"
+#include "webrtc/modules/audio_processing/level_controller/peak_level_estimator.h"
+#include "webrtc/modules/audio_processing/level_controller/saturating_gain_estimator.h"
+#include "webrtc/modules/audio_processing/level_controller/signal_classifier.h"
+#include "webrtc/modules/audio_processing/logging/apm_data_dumper.h"
+
+namespace webrtc {
+namespace {
+
+void UpdateAndRemoveDcLevel(float* dc_level, rtc::ArrayView<float> x) {
+  RTC_DCHECK_LT(0u, x.size());

[hlundin-webrtc, 2016/06/27 11:21:15]

I find it easier to read RTC_DCHECK(!x.empty());

[peah-webrtc, 2016/06/27 22:51:48]

Done.

+  float mean = 0;
+  for (float v : x) {
+    mean += v;
+  }
+  mean /= x.size();
+  *dc_level += 0.01f * (mean - *dc_level);

[hlundin-webrtc, 2016/06/27 11:21:15]

Is the size of x always constant in milliseconds? Otherwise, shouldn't the
forgetting factor be variable with the size? This is probably a nit in
practice...

[peah-webrtc, 2016/06/27 22:51:48]

Good point!
Done.

+
+  for (float& v : x) {
+    v -= *dc_level;
+  }
+}
+
+float FrameEnergy(const AudioBuffer& audio) {
+  auto sample_power = [](float a, float b) { return a + b * b; };

[hlundin-webrtc, 2016/06/27 11:21:15]

Nit: "Specify the return type of the lambda explicitly if that will make it more
obvious to readers"
float is better than auto, imo.

or

Define the lambda in-line in the accumulate call where it is used. It is short
enough.

[peah-webrtc, 2016/06/27 22:51:48]

Good point!
Done.

+
+  float energy = 0.f;
+  for (size_t k = 0; k < audio.num_channels(); ++k) {
+    float channel_energy = 0;

[hlundin-webrtc, 2016/06/27 11:21:15]

No need to initialize this to zero. Just add float to the line below.

[peah-webrtc, 2016/06/27 22:51:48]

Done.

+    channel_energy = std::accumulate(
+        audio.channels_const_f()[k],
+        audio.channels_const_f()[k] + audio.num_frames(), 0, sample_power);
+    energy = std::max(channel_energy, energy);
+  }
+  return energy;
+}
+
+float PeakLevel(const AudioBuffer& audio) {
+  float peak_level = 0.f;
+  auto compare_abs = [](float a, float b) { return std::abs(a) < std::abs(b); };

[hlundin-webrtc, 2016/06/27 11:21:15]

Move this lambda to in-line below.

[peah-webrtc, 2016/06/27 22:51:48]

Done.

+
+  for (size_t k = 0; k < audio.num_channels(); ++k) {
+    auto channel_peak_level = std::max_element(
+        audio.channels_const_f()[k],
+        audio.channels_const_f()[k] + audio.num_frames(), compare_abs);
+    peak_level = std::max(*channel_peak_level, peak_level);
+  }
+  return peak_level;
+}
+
+}  // namespace
+
+int LevelController::instance_count_ = 0;
+
+LevelController::LevelController() {
+  ++instance_count_;
+  data_dumper_.reset(new ApmDataDumper(instance_count_));
+  gain_selector_.reset(new GainSelector());
+  gain_applier_.reset(new GainApplier(data_dumper_.get()));
+  signal_classifier_.reset(new SignalClassifier(data_dumper_.get()));
+  noise_level_estimator_.reset(new NoiseLevelEstimator());
+  peak_level_estimator_.reset(new PeakLevelEstimator());
+  saturating_gain_estimator_.reset(new SaturatingGainEstimator());
+  Initialize(AudioProcessing::kSampleRate48kHz, 1u);
+}
+
+// TODO(peah): See if the destructors can be removed.
+LevelController::~LevelController() {}
+
+void LevelController::Initialize(int sample_rate_hz, size_t num_channels_) {
+  data_dumper_->InitiateNewSetOfRecordings();
+  gain_selector_->Initialize(sample_rate_hz);
+  gain_applier_->Initialize(sample_rate_hz);
+  signal_classifier_->Initialize(sample_rate_hz);
+  noise_level_estimator_->Initialize(sample_rate_hz);
+  peak_level_estimator_->Initialize();
+  saturating_gain_estimator_->Initialize();
+
+  sample_rate_hz_ = sample_rate_hz;
+}
+
+void LevelController::Process(AudioBuffer* audio) {
+  RTC_DCHECK_LT(0u, audio->num_channels());
+  RTC_DCHECK_GE(2u, audio->num_channels());
+  data_dumper_->DumpWav("lc_input", audio->num_frames(),
+                        audio->channels_const_f()[0], sample_rate_hz_, 1);
+
+  // Remove DC level.
+  for (size_t k = 0; k < audio->num_channels(); ++k) {
+    UpdateAndRemoveDcLevel(
+        &dc_level_[k],
+        rtc::ArrayView<float>(audio->channels_f()[k], audio->num_frames()));
+  }
+
+  SignalClassifier::SignalType signal_type;
+  signal_classifier_->Analyze(*audio, &signal_type);
+  int tmp = static_cast<int>(signal_type);
+  data_dumper_->DumpRaw("lc_signal_type", 1, &tmp);
+
+  // Estimate the noise energy.
+  float frame_energy = FrameEnergy(*audio);
+  float noise_energy =
+      noise_level_estimator_->Analyze(signal_type, frame_energy);

[hlundin-webrtc, 2016/06/27 11:21:15]

You can just as well call FrameEnergy(*audio) directly here.

[peah-webrtc, 2016/06/27 22:51:48]

Done.

+
+  // Estimate the overall signal peak level.
+  float frame_peak_level = PeakLevel(*audio);
+  float peak_level =
+      peak_level_estimator_->Analyze(signal_type, frame_peak_level);

[hlundin-webrtc, 2016/06/27 11:21:15]

Same here. Call PeakLevel(*audio) directly as a parameter.

[peah-webrtc, 2016/06/27 22:51:48]

Done.

+
+  float saturating_gain = saturating_gain_estimator_->GetGain();
+
+  // Compute the new gain to apply.
+  float new_gain =
+      gain_selector_->GetNewGain(peak_level, noise_energy, saturating_gain);
+
+  // Apply the gain to the signal.
+  int num_saturations = gain_applier_->Process(new_gain, audio);
+
+  // Estimate the gain that saturates the overall signal.
+  saturating_gain_estimator_->Update(new_gain, num_saturations);
+
+  data_dumper_->DumpRaw("lc_selected_gain", 1, &new_gain);
+  data_dumper_->DumpRaw("lc_noise_energy", 1, &noise_energy);
+  data_dumper_->DumpRaw("lc_peak_level", 1, &peak_level);
+  data_dumper_->DumpRaw("lc_saturating_gain", 1, &saturating_gain);
+
+  data_dumper_->DumpWav("lc_output", audio->num_frames(),
+                        audio->channels_f()[0], sample_rate_hz_, 1);
+}
+
+}  // namespace webrtc