| OLD | NEW |
| 1 /* | 1 /* |
| 2 * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved. | 2 * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved. |
| 3 * | 3 * |
| 4 * Use of this source code is governed by a BSD-style license | 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 | 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 | 6 * tree. An additional intellectual property rights grant can be found |
| 7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
| 8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
| 9 */ | 9 */ |
| 10 | 10 |
| 11 #include <cmath> | 11 #include <cmath> |
| 12 #include <algorithm> | 12 #include <algorithm> |
| 13 #include <vector> | 13 #include <vector> |
| 14 | 14 |
| 15 #include "testing/gtest/include/gtest/gtest.h" | 15 #include "testing/gtest/include/gtest/gtest.h" |
| 16 #include "webrtc/base/arraysize.h" |
| 16 #include "webrtc/base/format_macros.h" | 17 #include "webrtc/base/format_macros.h" |
| 17 #include "webrtc/base/scoped_ptr.h" | 18 #include "webrtc/base/scoped_ptr.h" |
| 18 #include "webrtc/common_audio/audio_converter.h" | 19 #include "webrtc/common_audio/audio_converter.h" |
| 19 #include "webrtc/common_audio/channel_buffer.h" | 20 #include "webrtc/common_audio/channel_buffer.h" |
| 20 #include "webrtc/common_audio/resampler/push_sinc_resampler.h" | 21 #include "webrtc/common_audio/resampler/push_sinc_resampler.h" |
| 21 | 22 |
| 22 namespace webrtc { | 23 namespace webrtc { |
| 23 | 24 |
| 24 typedef rtc::scoped_ptr<ChannelBuffer<float>> ScopedBuffer; | 25 typedef rtc::scoped_ptr<ChannelBuffer<float>> ScopedBuffer; |
| 25 | 26 |
| 26 // Sets the signal value to increase by |data| with every sample. | 27 // Sets the signal value to increase by |data| with every sample. |
| 27 ScopedBuffer CreateBuffer(const std::vector<float>& data, int frames) { | 28 ScopedBuffer CreateBuffer(const std::vector<float>& data, size_t frames) { |
| 28 const int num_channels = static_cast<int>(data.size()); | 29 const size_t num_channels = data.size(); |
| 29 ScopedBuffer sb(new ChannelBuffer<float>(frames, num_channels)); | 30 ScopedBuffer sb(new ChannelBuffer<float>(frames, num_channels)); |
| 30 for (int i = 0; i < num_channels; ++i) | 31 for (size_t i = 0; i < num_channels; ++i) |
| 31 for (int j = 0; j < frames; ++j) | 32 for (size_t j = 0; j < frames; ++j) |
| 32 sb->channels()[i][j] = data[i] * j; | 33 sb->channels()[i][j] = data[i] * j; |
| 33 return sb; | 34 return sb; |
| 34 } | 35 } |
| 35 | 36 |
| 36 void VerifyParams(const ChannelBuffer<float>& ref, | 37 void VerifyParams(const ChannelBuffer<float>& ref, |
| 37 const ChannelBuffer<float>& test) { | 38 const ChannelBuffer<float>& test) { |
| 38 EXPECT_EQ(ref.num_channels(), test.num_channels()); | 39 EXPECT_EQ(ref.num_channels(), test.num_channels()); |
| 39 EXPECT_EQ(ref.num_frames(), test.num_frames()); | 40 EXPECT_EQ(ref.num_frames(), test.num_frames()); |
| 40 } | 41 } |
| 41 | 42 |
| 42 // Computes the best SNR based on the error between |ref_frame| and | 43 // Computes the best SNR based on the error between |ref_frame| and |
| 43 // |test_frame|. It searches around |expected_delay| in samples between the | 44 // |test_frame|. It searches around |expected_delay| in samples between the |
| 44 // signals to compensate for the resampling delay. | 45 // signals to compensate for the resampling delay. |
| 45 float ComputeSNR(const ChannelBuffer<float>& ref, | 46 float ComputeSNR(const ChannelBuffer<float>& ref, |
| 46 const ChannelBuffer<float>& test, | 47 const ChannelBuffer<float>& test, |
| 47 size_t expected_delay) { | 48 size_t expected_delay) { |
| 48 VerifyParams(ref, test); | 49 VerifyParams(ref, test); |
| 49 float best_snr = 0; | 50 float best_snr = 0; |
| 50 size_t best_delay = 0; | 51 size_t best_delay = 0; |
| 51 | 52 |
| 52 // Search within one sample of the expected delay. | 53 // Search within one sample of the expected delay. |
| 53 for (size_t delay = std::max(expected_delay, static_cast<size_t>(1)) - 1; | 54 for (size_t delay = std::max(expected_delay, static_cast<size_t>(1)) - 1; |
| 54 delay <= std::min(expected_delay + 1, ref.num_frames()); | 55 delay <= std::min(expected_delay + 1, ref.num_frames()); |
| 55 ++delay) { | 56 ++delay) { |
| 56 float mse = 0; | 57 float mse = 0; |
| 57 float variance = 0; | 58 float variance = 0; |
| 58 float mean = 0; | 59 float mean = 0; |
| 59 for (int i = 0; i < ref.num_channels(); ++i) { | 60 for (size_t i = 0; i < ref.num_channels(); ++i) { |
| 60 for (size_t j = 0; j < ref.num_frames() - delay; ++j) { | 61 for (size_t j = 0; j < ref.num_frames() - delay; ++j) { |
| 61 float error = ref.channels()[i][j] - test.channels()[i][j + delay]; | 62 float error = ref.channels()[i][j] - test.channels()[i][j + delay]; |
| 62 mse += error * error; | 63 mse += error * error; |
| 63 variance += ref.channels()[i][j] * ref.channels()[i][j]; | 64 variance += ref.channels()[i][j] * ref.channels()[i][j]; |
| 64 mean += ref.channels()[i][j]; | 65 mean += ref.channels()[i][j]; |
| 65 } | 66 } |
| 66 } | 67 } |
| 67 | 68 |
| 68 const size_t length = ref.num_channels() * (ref.num_frames() - delay); | 69 const size_t length = ref.num_channels() * (ref.num_frames() - delay); |
| 69 mse /= length; | 70 mse /= length; |
| 70 variance /= length; | 71 variance /= length; |
| 71 mean /= length; | 72 mean /= length; |
| 72 variance -= mean * mean; | 73 variance -= mean * mean; |
| 73 float snr = 100; // We assign 100 dB to the zero-error case. | 74 float snr = 100; // We assign 100 dB to the zero-error case. |
| 74 if (mse > 0) | 75 if (mse > 0) |
| 75 snr = 10 * std::log10(variance / mse); | 76 snr = 10 * std::log10(variance / mse); |
| 76 if (snr > best_snr) { | 77 if (snr > best_snr) { |
| 77 best_snr = snr; | 78 best_snr = snr; |
| 78 best_delay = delay; | 79 best_delay = delay; |
| 79 } | 80 } |
| 80 } | 81 } |
| 81 printf("SNR=%.1f dB at delay=%" PRIuS "\n", best_snr, best_delay); | 82 printf("SNR=%.1f dB at delay=%" PRIuS "\n", best_snr, best_delay); |
| 82 return best_snr; | 83 return best_snr; |
| 83 } | 84 } |
| 84 | 85 |
| 85 // Sets the source to a linearly increasing signal for which we can easily | 86 // Sets the source to a linearly increasing signal for which we can easily |
| 86 // generate a reference. Runs the AudioConverter and ensures the output has | 87 // generate a reference. Runs the AudioConverter and ensures the output has |
| 87 // sufficiently high SNR relative to the reference. | 88 // sufficiently high SNR relative to the reference. |
| 88 void RunAudioConverterTest(int src_channels, | 89 void RunAudioConverterTest(size_t src_channels, |
| 89 int src_sample_rate_hz, | 90 int src_sample_rate_hz, |
| 90 int dst_channels, | 91 size_t dst_channels, |
| 91 int dst_sample_rate_hz) { | 92 int dst_sample_rate_hz) { |
| 92 const float kSrcLeft = 0.0002f; | 93 const float kSrcLeft = 0.0002f; |
| 93 const float kSrcRight = 0.0001f; | 94 const float kSrcRight = 0.0001f; |
| 94 const float resampling_factor = (1.f * src_sample_rate_hz) / | 95 const float resampling_factor = (1.f * src_sample_rate_hz) / |
| 95 dst_sample_rate_hz; | 96 dst_sample_rate_hz; |
| 96 const float dst_left = resampling_factor * kSrcLeft; | 97 const float dst_left = resampling_factor * kSrcLeft; |
| 97 const float dst_right = resampling_factor * kSrcRight; | 98 const float dst_right = resampling_factor * kSrcRight; |
| 98 const float dst_mono = (dst_left + dst_right) / 2; | 99 const float dst_mono = (dst_left + dst_right) / 2; |
| 99 const int src_frames = src_sample_rate_hz / 100; | 100 const size_t src_frames = static_cast<size_t>(src_sample_rate_hz / 100); |
| 100 const int dst_frames = dst_sample_rate_hz / 100; | 101 const size_t dst_frames = static_cast<size_t>(dst_sample_rate_hz / 100); |
| 101 | 102 |
| 102 std::vector<float> src_data(1, kSrcLeft); | 103 std::vector<float> src_data(1, kSrcLeft); |
| 103 if (src_channels == 2) | 104 if (src_channels == 2) |
| 104 src_data.push_back(kSrcRight); | 105 src_data.push_back(kSrcRight); |
| 105 ScopedBuffer src_buffer = CreateBuffer(src_data, src_frames); | 106 ScopedBuffer src_buffer = CreateBuffer(src_data, src_frames); |
| 106 | 107 |
| 107 std::vector<float> dst_data(1, 0); | 108 std::vector<float> dst_data(1, 0); |
| 108 std::vector<float> ref_data; | 109 std::vector<float> ref_data; |
| 109 if (dst_channels == 1) { | 110 if (dst_channels == 1) { |
| 110 if (src_channels == 1) | 111 if (src_channels == 1) |
| 111 ref_data.push_back(dst_left); | 112 ref_data.push_back(dst_left); |
| 112 else | 113 else |
| 113 ref_data.push_back(dst_mono); | 114 ref_data.push_back(dst_mono); |
| 114 } else { | 115 } else { |
| 115 dst_data.push_back(0); | 116 dst_data.push_back(0); |
| 116 ref_data.push_back(dst_left); | 117 ref_data.push_back(dst_left); |
| 117 if (src_channels == 1) | 118 if (src_channels == 1) |
| 118 ref_data.push_back(dst_left); | 119 ref_data.push_back(dst_left); |
| 119 else | 120 else |
| 120 ref_data.push_back(dst_right); | 121 ref_data.push_back(dst_right); |
| 121 } | 122 } |
| 122 ScopedBuffer dst_buffer = CreateBuffer(dst_data, dst_frames); | 123 ScopedBuffer dst_buffer = CreateBuffer(dst_data, dst_frames); |
| 123 ScopedBuffer ref_buffer = CreateBuffer(ref_data, dst_frames); | 124 ScopedBuffer ref_buffer = CreateBuffer(ref_data, dst_frames); |
| 124 | 125 |
| 125 // The sinc resampler has a known delay, which we compute here. | 126 // The sinc resampler has a known delay, which we compute here. |
| 126 const size_t delay_frames = src_sample_rate_hz == dst_sample_rate_hz ? 0 : | 127 const size_t delay_frames = src_sample_rate_hz == dst_sample_rate_hz ? 0 : |
| 127 static_cast<size_t>( | 128 static_cast<size_t>( |
| 128 PushSincResampler::AlgorithmicDelaySeconds(src_sample_rate_hz) * | 129 PushSincResampler::AlgorithmicDelaySeconds(src_sample_rate_hz) * |
| 129 dst_sample_rate_hz); | 130 dst_sample_rate_hz); |
| 130 printf("(%d, %d Hz) -> (%d, %d Hz) ", // SNR reported on the same line later. | 131 // SNR reported on the same line later. |
| 131 src_channels, src_sample_rate_hz, dst_channels, dst_sample_rate_hz); | 132 printf("(%" PRIuS ", %d Hz) -> (%" PRIuS ", %d Hz) ", |
| 133 src_channels, src_sample_rate_hz, dst_channels, dst_sample_rate_hz); |
| 132 | 134 |
| 133 rtc::scoped_ptr<AudioConverter> converter = AudioConverter::Create( | 135 rtc::scoped_ptr<AudioConverter> converter = AudioConverter::Create( |
| 134 src_channels, src_frames, dst_channels, dst_frames); | 136 src_channels, src_frames, dst_channels, dst_frames); |
| 135 converter->Convert(src_buffer->channels(), src_buffer->size(), | 137 converter->Convert(src_buffer->channels(), src_buffer->size(), |
| 136 dst_buffer->channels(), dst_buffer->size()); | 138 dst_buffer->channels(), dst_buffer->size()); |
| 137 | 139 |
| 138 EXPECT_LT(43.f, | 140 EXPECT_LT(43.f, |
| 139 ComputeSNR(*ref_buffer.get(), *dst_buffer.get(), delay_frames)); | 141 ComputeSNR(*ref_buffer.get(), *dst_buffer.get(), delay_frames)); |
| 140 } | 142 } |
| 141 | 143 |
| 142 TEST(AudioConverterTest, ConversionsPassSNRThreshold) { | 144 TEST(AudioConverterTest, ConversionsPassSNRThreshold) { |
| 143 const int kSampleRates[] = {8000, 16000, 32000, 44100, 48000}; | 145 const int kSampleRates[] = {8000, 16000, 32000, 44100, 48000}; |
| 144 const int kSampleRatesSize = sizeof(kSampleRates) / sizeof(*kSampleRates); | 146 const size_t kChannels[] = {1, 2}; |
| 145 const int kChannels[] = {1, 2}; | 147 for (size_t src_rate = 0; src_rate < arraysize(kSampleRates); ++src_rate) { |
| 146 const int kChannelsSize = sizeof(kChannels) / sizeof(*kChannels); | 148 for (size_t dst_rate = 0; dst_rate < arraysize(kSampleRates); ++dst_rate) { |
| 147 for (int src_rate = 0; src_rate < kSampleRatesSize; ++src_rate) { | 149 for (size_t src_channel = 0; src_channel < arraysize(kChannels); |
| 148 for (int dst_rate = 0; dst_rate < kSampleRatesSize; ++dst_rate) { | 150 ++src_channel) { |
| 149 for (int src_channel = 0; src_channel < kChannelsSize; ++src_channel) { | 151 for (size_t dst_channel = 0; dst_channel < arraysize(kChannels); |
| 150 for (int dst_channel = 0; dst_channel < kChannelsSize; ++dst_channel) { | 152 ++dst_channel) { |
| 151 RunAudioConverterTest(kChannels[src_channel], kSampleRates[src_rate], | 153 RunAudioConverterTest(kChannels[src_channel], kSampleRates[src_rate], |
| 152 kChannels[dst_channel], kSampleRates[dst_rate]); | 154 kChannels[dst_channel], kSampleRates[dst_rate]); |
| 153 } | 155 } |
| 154 } | 156 } |
| 155 } | 157 } |
| 156 } | 158 } |
| 157 } | 159 } |
| 158 | 160 |
| 159 } // namespace webrtc | 161 } // namespace webrtc |
| OLD | NEW |