Chromium Code Reviews| OLD | NEW |
|---|---|
| (Empty) | |
| 1 /* | |
| 2 * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved. | |
|
hlundin-webrtc
2015/07/01 09:09:27
2015
ekm
2015/07/01 23:48:26
Done.
| |
| 3 * | |
| 4 * Use of this source code is governed by a BSD-style license | |
| 5 * that can be found in the LICENSE file in the root of the source | |
| 6 * tree. An additional intellectual property rights grant can be found | |
| 7 * in the file PATENTS. All contributing project authors may | |
| 8 * be found in the AUTHORS file in the root of the source tree. | |
| 9 */ | |
| 10 | |
| 11 // | |
| 12 // Unit tests for intelligibility utils. | |
| 13 // | |
| 14 | |
| 15 | |
| 16 #include <cmath> | |
|
hlundin-webrtc
2015/07/01 09:09:27
math.h
ekm
2015/07/01 23:48:27
Done.
| |
| 17 #include <iostream> | |
| 18 #include <vector> | |
| 19 | |
| 20 #include "testing/gtest/include/gtest/gtest.h" | |
| 21 #include "webrtc/modules/audio_processing/intelligibility/intelligibility_utils. h" | |
| 22 | |
| 23 using std::complex; | |
| 24 using std::vector; | |
| 25 | |
| 26 namespace webrtc { | |
| 27 | |
| 28 namespace intelligibility { | |
| 29 | |
| 30 vector<vector<complex<float>>> GenerateTestData(int freqs, int samples) { | |
| 31 vector<vector<complex<float>>> data(samples); | |
| 32 for (int i = 0; i < samples; i++) { | |
| 33 data[i].resize(freqs); | |
| 34 for (int j = 0; j < freqs; j++) { | |
| 35 data[i][j].real(0.99f / ((i+1)*(j+1))); | |
| 36 data[i][j].imag(0.99f / ((i+1)*(j+1))); | |
| 37 } | |
| 38 } | |
| 39 return data; | |
| 40 } | |
| 41 | |
| 42 // Tests UpdateFactor. | |
| 43 TEST(IntelligibilityUtilsTest, TestUpdateFactor) { | |
| 44 EXPECT_EQ(intelligibility::UpdateFactor(0, 0, 0), 0); | |
|
hlundin-webrtc
2015/07/01 09:09:27
Switch the order of the parameters to EXPECT_EQ. S
ekm
2015/07/01 23:48:26
Done. It looks like for _NEAR order doesn't matter
Andrew MacDonald
2015/07/02 02:46:47
For consistency, I'd still use EXPECT_NEAR(expecte
ekm
2015/07/07 21:57:02
Done.
| |
| 45 EXPECT_EQ(intelligibility::UpdateFactor(4, 2, 3), 4); | |
| 46 EXPECT_EQ(intelligibility::UpdateFactor(4, 2, 1), 3); | |
| 47 EXPECT_EQ(intelligibility::UpdateFactor(2, 4, 3), 2); | |
| 48 EXPECT_EQ(intelligibility::UpdateFactor(2, 4, 1), 3); | |
| 49 } | |
| 50 | |
| 51 // Tests cplxfinite, cplxnormal, and zerofudge. | |
| 52 TEST(IntelligibilityUtilsTest, TestCplx) { | |
| 53 complex<float> t; | |
| 54 t.real(1.f); | |
| 55 t.imag(0.f); | |
| 56 EXPECT_TRUE(intelligibility::cplxfinite(t)); | |
| 57 EXPECT_FALSE(intelligibility::cplxnormal(t)); | |
| 58 t = intelligibility::zerofudge(t); | |
| 59 EXPECT_NE(t.imag(), 0.f); | |
| 60 EXPECT_NE(t.real(), 0.f); | |
| 61 t.imag(1.f/0.f); | |
| 62 EXPECT_FALSE(intelligibility::cplxfinite(t)); | |
| 63 EXPECT_FALSE(intelligibility::cplxnormal(t)); | |
| 64 t.imag(sqrt(-1.f)); | |
| 65 EXPECT_FALSE(intelligibility::cplxfinite(t)); | |
| 66 EXPECT_FALSE(intelligibility::cplxnormal(t)); | |
| 67 t.imag(1.f); | |
| 68 EXPECT_TRUE(intelligibility::cplxfinite(t)); | |
| 69 EXPECT_TRUE(intelligibility::cplxnormal(t)); | |
| 70 } | |
| 71 | |
| 72 // Tests NewMean and AddToMean. | |
| 73 /*TEST(IntelligibilityUtilsTest, TestMeanUpdate) { | |
|
hlundin-webrtc
2015/07/01 09:09:27
Don't comment out tests. Instead, disable them so
ekm
2015/07/01 23:48:26
Thanks for telling me about DISABLED_, that'll def
hlundin-webrtc
2015/07/02 10:53:13
Acknowledged.
| |
| 74 vector<complex<float>> data = {{3, 8}, {7, 6}, {2, 1}, {8, 9}, {0, 6}}; | |
| 75 vector<complex<float>> means = {{3, 8}, {5, 7}, {4, 5}, {5, 6}, {4, 6}}; | |
| 76 complex<float> mean(3, 8); | |
| 77 for (vector<int>::size_type i = 0; i < data.size(); i++) { | |
| 78 EXPECT_EQ(NewMean(mean, data[i], i+1), means[i]); | |
| 79 AddToMean(data[i], i+1, &mean); | |
| 80 EXPECT_EQ(mean, means[i]); | |
| 81 } | |
| 82 }*/ | |
| 83 | |
| 84 // Tests VarianceArray, for all variance step types. | |
| 85 TEST(IntelligibilityUtilsTest, TestVarianceArray) { | |
| 86 const int kFreqs = 10; | |
| 87 const int kSamples = 100; | |
| 88 const int kWindowSize = 10; // Should pass for all kWindowSize > 1. | |
| 89 const float kDecay = 0.5; | |
| 90 vector<VarianceArray::StepType> step_types = { | |
|
hlundin-webrtc
2015/07/01 09:09:27
const vector
ekm
2015/07/01 23:48:27
Done.
| |
| 91 VarianceArray::kStepInfinite, VarianceArray::kStepDecaying, | |
| 92 VarianceArray::kStepWindowed, VarianceArray::kStepBlocked, | |
| 93 VarianceArray::kStepBlockBasedMovingAverage}; | |
| 94 const vector<vector<complex<float>>> test_data( | |
| 95 GenerateTestData(kFreqs, kSamples)); | |
| 96 for (VarianceArray::StepType step_type : step_types) { | |
|
hlundin-webrtc
2015/07/01 09:09:27
for (auto ...)
ekm
2015/07/01 23:48:27
Done.
| |
| 97 VarianceArray variance_array(kFreqs, step_type, kWindowSize, kDecay); | |
| 98 EXPECT_EQ(variance_array.variance()[0], 0); | |
| 99 EXPECT_EQ(variance_array.array_mean(), 0); | |
| 100 variance_array.ApplyScale(2.0f); | |
| 101 EXPECT_EQ(variance_array.variance()[0], 0); | |
| 102 EXPECT_EQ(variance_array.array_mean(), 0); | |
| 103 | |
| 104 // Makes sure Step is doing something. | |
| 105 variance_array.Step(&test_data[0][0]); | |
| 106 for (int i = 1; i < kSamples; i++) { | |
| 107 variance_array.Step(&test_data[i][0]); | |
| 108 EXPECT_GE(variance_array.array_mean(), 0.0f); | |
| 109 EXPECT_LE(variance_array.array_mean(), 1.0f); | |
| 110 for (int j = 0; j < kFreqs; j++) { | |
| 111 EXPECT_GE(variance_array.variance()[j], 0.0f); | |
| 112 EXPECT_LE(variance_array.variance()[j], 1.0f); | |
| 113 } | |
| 114 } | |
| 115 variance_array.Clear(); | |
| 116 EXPECT_EQ(variance_array.variance()[0], 0); | |
| 117 EXPECT_EQ(variance_array.array_mean(), 0); | |
| 118 } | |
| 119 } | |
| 120 | |
| 121 // Tests exact computation on synthetic data. | |
| 122 TEST(IntelligibilityUtilsTest, TestMovingBlockAverage) { | |
| 123 // Exact, not unbiased estimates. | |
| 124 const float kTestVarianceBufferNotFull = 16.5f; | |
| 125 const float kTestVarianceBufferFull1 = 66.5f; | |
| 126 const float kTestVarianceBufferFull2 = 333.375f; | |
| 127 const int kFreqs = 2; | |
| 128 const int kSamples = 50; | |
| 129 const int kWindowSize = 2; | |
| 130 const float kDecay = 0.5f; | |
| 131 const float kMaxError = 0.0001f; | |
| 132 | |
| 133 VarianceArray variance_array( | |
| 134 kFreqs, VarianceArray::kStepBlockBasedMovingAverage, | |
| 135 kWindowSize, kDecay); | |
| 136 | |
| 137 vector<vector<complex<float>>> test_data(kSamples); | |
| 138 for (int i = 0; i < kSamples; i++) { | |
| 139 test_data[i].resize(kFreqs); | |
| 140 for (int j = 0; j < kFreqs; j++) { | |
| 141 if (i < 30) { | |
| 142 test_data[i][j].real(static_cast<float>(kSamples - i)); | |
| 143 test_data[i][j].imag(static_cast<float>(i + 1)); | |
| 144 } else { | |
| 145 test_data[i][j].real(0.f); | |
| 146 test_data[i][j].imag(0.f); | |
| 147 } | |
| 148 } | |
| 149 } | |
| 150 | |
| 151 for (int i = 0; i < kSamples; i++) { | |
| 152 variance_array.Step(&test_data[i][0]); | |
| 153 for (int j = 0; j < kFreqs; j++) { | |
| 154 if (i < 9) { // In utils, kWindowBlockSize = 10. | |
| 155 EXPECT_EQ(variance_array.variance()[j], 0); | |
| 156 } else if (i < 19) { | |
| 157 EXPECT_NEAR(variance_array.variance()[j], kTestVarianceBufferNotFull, | |
| 158 kMaxError); | |
|
hlundin-webrtc
2015/07/01 09:09:27
Wrong indentation.
ekm
2015/07/01 23:48:27
Done.
Andrew MacDonald
2015/07/02 02:46:47
Strongly suggest running "git cl format" on all yo
ekm
2015/07/07 21:57:02
Acknowledged.
| |
| 159 } else if (i < 39) { | |
| 160 EXPECT_NEAR(variance_array.variance()[j], kTestVarianceBufferFull1, | |
| 161 kMaxError); | |
| 162 } else if (i < 49) { | |
| 163 EXPECT_NEAR(variance_array.variance()[j], kTestVarianceBufferFull2, | |
| 164 kMaxError); | |
| 165 } else { | |
| 166 EXPECT_EQ(variance_array.variance()[j], 0); | |
| 167 } | |
| 168 } | |
| 169 } | |
| 170 } | |
| 171 | |
| 172 // Tests gain applier. | |
| 173 TEST(IntelligibilityUtilsTest, TestGainApplier) { | |
| 174 const int kFreqs = 10; | |
| 175 const int kSamples = 100; | |
| 176 const float kChangeLimit = 0.1f; | |
| 177 GainApplier gain_applier(kFreqs, kChangeLimit); | |
| 178 const vector<vector<complex<float>>> in_data( | |
| 179 GenerateTestData(kFreqs, kSamples)); | |
| 180 vector<vector<complex<float>>> out_data( | |
| 181 GenerateTestData(kFreqs, kSamples)); | |
| 182 for (int i = 0; i < kSamples; i++) { | |
| 183 gain_applier.Apply(&in_data[i][0], &out_data[i][0]); | |
| 184 for (int j = 0; j < kFreqs; j++) { | |
| 185 EXPECT_GT(out_data[i][j].real(), 0.0f); | |
| 186 EXPECT_LT(out_data[i][j].real(), 1.0f); | |
| 187 EXPECT_GT(out_data[i][j].imag(), 0.0f); | |
| 188 EXPECT_LT(out_data[i][j].imag(), 1.0f); | |
| 189 } | |
| 190 } | |
| 191 } | |
| 192 | |
| 193 } // namespace intelligibility | |
| 194 | |
| 195 } // namespace webrtc | |
| OLD | NEW |