Add new variance update option and unittests for intelligibility

webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer_unittest.cc

Index: webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer_unittest.cc
diff --git a/webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer_unittest.cc b/webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer_unittest.cc
new file mode 100644
index 0000000000000000000000000000000000000000..845f4fc3f97c2ad3825c777540131b0743628afd
--- /dev/null
+++ b/webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer_unittest.cc
@@ -0,0 +1,226 @@
+/*
+ *  Copyright (c) 2015 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.
+ */
+
+//
+//  Unit tests for intelligibility enhancer.
+//
+
+#include <math.h>
+#include <algorithm>
+#include <vector>
+
+#include "testing/gtest/include/gtest/gtest.h"
+#include "webrtc/base/arraysize.h"
+#include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
+#include "webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer.h"
+
+namespace {
+
+// Generated with matlab code: normrnd(0,1000,64,1).
+const double kGaussianSamples[] = {
+    1689.1,  1437,    -2251.1, 356.49,  -850.24, -299.55, -634.25, 1624.5,
+    1241.1,  555.28,  703.42,  458.16,  683.98,  251.29,  -178.5,  507.73,
+    -309.9,  -394.37, -269.74, -88.13,  8.0293,  2531.8,  -1223.2, -1071.8,
+    246.06,  -50.611, -730.15, 326.99,  752.99,  -1153.7, -407.87, -1287.9,
+    83.578,  163.8,   682.57,  -1086.4, 297.49,  -143.31, 1392,    306.75,
+    -537.18, -228.93, -536.22, 1439,    -511.1,  -1606.8, -201.24, 1143.5,
+    663.29,  164.08,  1785.4,  -587.71, 259.04,  -871.83, -787.92, -344.34,
+    647.62,  2054.1,  798.94,  -1071.1, -205.16, -554.44, -292.94, 1180.2};
+static_assert(arraysize(kGaussianSamples) == 64, "Samples badly initialized.");
+

[hlundin-webrtc, 2015/07/02 10:53:13]

You have a blank line here that is not present between the rest of the
constants.

[ekm, 2015/07/07 21:57:02]

Done.

+// Target output for ERB create test. Generated with matlab.
+const double kTestCenterFreqs[] = {
+    13.169, 26.965, 41.423, 56.577, 72.461, 89.113, 106.57, 124.88,
+    144.08, 164.21, 185.34, 207.5,  230.75, 255.16, 280.77, 307.66,
+    335.9,  365.56, 396.71, 429.44, 463.84, 500};
+const double kTestFilterBank[][2] = {{0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0},
+                                     {0.055556, 0.2},
+                                     {0, 0.2},
+                                     {0, 0.2},
+                                     {0, 0.2},
+                                     {0, 0.2}};
+static_assert(arraysize(kTestCenterFreqs) == arraysize(kTestFilterBank),
+              "Test filterbank badly initialized.");
+// Target output for gain solving test. Generated with matlab.
+const int kTestStartFreq = 12;  // Lowest integral frequency for ERBs.
+const double kTestZeroVar[] = {1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
+                               0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0};
+static_assert(arraysize(kTestCenterFreqs) == arraysize(kTestZeroVar),
+              "Variance test data badly initialized.");
+const double kTestNonZeroVarLambdaTop[] = {
+    1,      1,      1,      1,      1,      1,      1,      1,
+    1,      1,      1,      0,      0,      0.0351, 0.0636, 0.0863,
+    0.1037, 0.1162, 0.1236, 0.1251, 0.1189, 0.0993};
+static_assert(arraysize(kTestCenterFreqs) ==
+                  arraysize(kTestNonZeroVarLambdaTop),
+              "Variance test data badly initialized.");
+const float kMaxTestError = 0.005f;
+
+// Enhancer initialization parameters.
+const int kSamples = 2000;
+const int kErbResolution = 2;
+const int kSampleRate = 1000;
+const int kFragmentSize = kSampleRate / 100;
+const int kNumChannels = 1;
+const float kDecayRate = 0.9f;
+const int kWindowSize = 800;
+const int kAnalyzeRate = 800;
+const int kVarianceRate = 2;
+const float kGainLimit = 0.1f;
+
+}  // namespace
+
+namespace webrtc {
+
+using std::vector;
+using intelligibility::VarianceArray;
+
+void GenerateConstantData(vector<float>* data, float constant) {

[Andrew MacDonald, 2015/07/02 02:46:48]

Replace this function with:
std::fill(data.begin(), data.end(), constant);

at all callsites.

[ekm, 2015/07/07 21:57:02]

Done.

+  for (size_t i = 0; i < data->size(); i++) {
+    (*data)[i] = constant;
+  }
+}
+
+void GenerateGaussianData(vector<float>* data, int* count) {

[Andrew MacDonald, 2015/07/02 02:46:48]

Does this need to be Gaussian distributed, or will uniform noise do? Consider
just std::rand() with a seed to make it deterministic.

[ekm, 2015/07/07 21:57:02]

Done.

+  for (size_t i = 0; i < data->size(); i++) {
+    (*data)[i] = kGaussianSamples[(*count) % 64];
+    (*count)++;
+  }
+}
+
+class IntelligibilityEnhancerTest : public ::testing::Test {
+ protected:
+  IntelligibilityEnhancerTest()
+      : enh_(kErbResolution,
+             kSampleRate,
+             kNumChannels,
+             VarianceArray::kStepInfinite,
+             kDecayRate,
+             kWindowSize,
+             kAnalyzeRate,
+             kVarianceRate,
+             kGainLimit),
+        clear_data_(kSamples),
+        noise_data_(kSamples),
+        orig_data_(kSamples) {}
+
+  bool CheckUpdate(VarianceArray::StepType step_type) {
+    IntelligibilityEnhancer enh(kErbResolution, kSampleRate, kNumChannels,
+                                step_type, kDecayRate, kWindowSize,
+                                kAnalyzeRate, kVarianceRate, kGainLimit);
+    float* clear_cursor = &clear_data_[0];
+    float* noise_cursor = &noise_data_[0];
+    for (int i = 0; i < kSamples; i += kFragmentSize) {
+      enh.ProcessCaptureAudio(&noise_cursor);
+      enh.ProcessRenderAudio(&clear_cursor);
+      clear_cursor += kFragmentSize;
+      noise_cursor += kFragmentSize;
+    }
+    for (int i = 0; i < kSamples; i++) {
+      if (std::fabs(clear_data_[i] - orig_data_[i]) > kMaxTestError) {
+        return true;
+      }
+    }
+    return false;
+  }
+
+  IntelligibilityEnhancer enh_;
+  vector<float> clear_data_;
+  vector<float> noise_data_;
+  vector<float> orig_data_;
+};
+
+// For each class of generated data, tests that render stream is
+// updated when it should be for each variance update method.
+TEST_F(IntelligibilityEnhancerTest, TestRenderUpdate) {
+  vector<VarianceArray::StepType> step_types = {
+      VarianceArray::kStepInfinite,
+      VarianceArray::kStepDecaying,
+      VarianceArray::kStepWindowed,
+      VarianceArray::kStepBlocked,
+      VarianceArray::kStepBlockBasedMovingAverage};
+  GenerateConstantData(&noise_data_, 0.0f);
+  GenerateConstantData(&orig_data_, 0.0f);
+  for (auto step_type : step_types) {
+    GenerateConstantData(&clear_data_, 0.0f);
+    EXPECT_FALSE(CheckUpdate(step_type));
+  }
+  int samples_grabbed = 0;
+  GenerateGaussianData(&noise_data_, &samples_grabbed);
+  for (auto step_type : step_types) {
+    EXPECT_FALSE(CheckUpdate(step_type));
+  }
+  for (auto step_type : step_types) {
+    GenerateGaussianData(&clear_data_, &samples_grabbed);
+    orig_data_ = clear_data_;
+    EXPECT_TRUE(CheckUpdate(step_type));
+  }
+}
+
+// Tests ERB bank creation, comparing against matlab output.
+TEST_F(IntelligibilityEnhancerTest, TestErbCreation) {
+  ASSERT_EQ(static_cast<int>(arraysize(kTestCenterFreqs)), enh_.bank_size_);
+  for (int i = 0; i < enh_.bank_size_; ++i) {
+    EXPECT_NEAR(enh_.center_freqs_[i], kTestCenterFreqs[i], kMaxTestError);
+    ASSERT_EQ(static_cast<int>(arraysize(kTestFilterBank[0])), enh_.freqs_);
+    for (int j = 0; j < enh_.freqs_; ++j) {
+      EXPECT_NEAR(enh_.filter_bank_[i][j], kTestFilterBank[i][j],
+                  kMaxTestError);
+    }
+  }
+}
+
+// Tests analytic solution for optimal gains, comparing
+// against matlab output.
+TEST_F(IntelligibilityEnhancerTest, TestSolveForGains) {
+  ASSERT_EQ(kTestStartFreq, enh_.start_freq_);
+  vector<float> sols(enh_.bank_size_);
+  float lambda = -0.001f;
+  for (int i = 0; i < enh_.bank_size_; i++) {
+    enh_.filtered_clear_var_[i] = 0.0;
+    enh_.filtered_noise_var_[i] = 0.0;
+    enh_.rho_[i] = 0.02;
+  }
+  enh_.SolveForGainsGivenLambda(lambda, enh_.start_freq_, &sols[0]);
+  for (int i = 0; i < enh_.bank_size_; i++) {
+    EXPECT_NEAR(sols[i], kTestZeroVar[i], kMaxTestError);
+  }
+  for (int i = 0; i < enh_.bank_size_; i++) {
+    enh_.filtered_clear_var_[i] = static_cast<float>(i + 1);
+    enh_.filtered_noise_var_[i] = static_cast<float>(enh_.bank_size_ - i);
+  }
+  enh_.SolveForGainsGivenLambda(lambda, enh_.start_freq_, &sols[0]);
+  for (int i = 0; i < enh_.bank_size_; i++) {
+    EXPECT_NEAR(sols[i], kTestNonZeroVarLambdaTop[i], kMaxTestError);
+  }
+  lambda = -1.0;
+  enh_.SolveForGainsGivenLambda(lambda, enh_.start_freq_, &sols[0]);
+  for (int i = 0; i < enh_.bank_size_; i++) {
+    EXPECT_NEAR(sols[i], kTestZeroVar[i], kMaxTestError);
+  }
+}
+
+}  // namespace webrtc