Add new variance update option and unittests for intelligibility

webrtc/modules/audio_processing/intelligibility/intelligibility_enhancer_unittest.cc

+/*
+ *  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 <stdlib.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 webrtc {
+
+namespace {
+
+// 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
+
+using std::vector;
+using intelligibility::VarianceArray;
+
+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 = {

[Andrew MacDonald, 2015/07/09 03:22:47]

I don't think you should bother now, but just so you know there's a gtest
facility for doing this kind of thing called a value-parameterized test.

Example here:
https://code.google.com/p/chromium/codesearch#chromium/src/third_party/webrtc...

[ekm, 2015/07/09 18:19:22]

Acknowledged. Will use in future tests.

+      VarianceArray::kStepInfinite,
+      VarianceArray::kStepDecaying,
+      VarianceArray::kStepWindowed,
+      VarianceArray::kStepBlocked,
+      VarianceArray::kStepBlockBasedMovingAverage};
+  std::fill(noise_data_.begin(), noise_data_.end(), 0.0f);
+  std::fill(orig_data_.begin(), orig_data_.end(), 0.0f);
+  for (auto step_type : step_types) {
+    std::fill(clear_data_.begin(), clear_data_.end(), 0.0f);
+    EXPECT_FALSE(CheckUpdate(step_type));
+  }
+  std::srand(1);
+  std::fill(noise_data_.begin(), noise_data_.end(),

[Andrew MacDonald, 2015/07/09 03:22:47]

This isn't doing what you want. It's filling noise_data_ with a single randomly
generated value. Instead:

auto float_rand = []() { return std::rand() * 2.f / RAND_MAX - 1; }
std::generate(noise_data_.begin(), noise_data_.end(), float_rand);

(usually floating point audio values are given in the [-1, 1] range.)

(and below with clear_data_)

[ekm, 2015/07/09 18:19:22]

Whoops. This is a very nice snippet, thanks. Done.

+            static_cast<float>(std::rand()));
+  for (auto step_type : step_types) {
+    EXPECT_FALSE(CheckUpdate(step_type));
+  }
+  for (auto step_type : step_types) {
+    std::fill(clear_data_.begin(), clear_data_.end(),
+              static_cast<float>(std::rand()));
+    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(kTestCenterFreqs[i], enh_.center_freqs_[i], kMaxTestError);
+    ASSERT_EQ(static_cast<int>(arraysize(kTestFilterBank[0])), enh_.freqs_);
+    for (int j = 0; j < enh_.freqs_; ++j) {
+      EXPECT_NEAR(kTestFilterBank[i][j], enh_.filter_bank_[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(kTestZeroVar[i], sols[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(kTestNonZeroVarLambdaTop[i], sols[i], kMaxTestError);
+  }
+  lambda = -1.0;
+  enh_.SolveForGainsGivenLambda(lambda, enh_.start_freq_, &sols[0]);
+  for (int i = 0; i < enh_.bank_size_; i++) {
+    EXPECT_NEAR(kTestZeroVar[i], sols[i], kMaxTestError);
+  }
+}
+
+}  // namespace webrtc