2 * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved.
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.
15 #include "testing/gtest/include/gtest/gtest.h"
16 #include "webrtc/common_audio/audio_converter.h"
17 #include "webrtc/common_audio/resampler/push_sinc_resampler.h"
18 #include "webrtc/modules/audio_processing/common.h"
19 #include "webrtc/system_wrappers/interface/scoped_ptr.h"
23 typedef scoped_ptr<ChannelBuffer<float>> ScopedBuffer;
25 // Sets the signal value to increase by |data| with every sample.
26 ScopedBuffer CreateBuffer(const std::vector<float>& data, int frames) {
27 const int num_channels = static_cast<int>(data.size());
28 ScopedBuffer sb(new ChannelBuffer<float>(frames, num_channels));
29 for (int i = 0; i < num_channels; ++i)
30 for (int j = 0; j < frames; ++j)
31 sb->channel(i)[j] = data[i] * j;
35 void VerifyParams(const ChannelBuffer<float>& ref,
36 const ChannelBuffer<float>& test) {
37 EXPECT_EQ(ref.num_channels(), test.num_channels());
38 EXPECT_EQ(ref.samples_per_channel(), test.samples_per_channel());
41 // Computes the best SNR based on the error between |ref_frame| and
42 // |test_frame|. It searches around |expected_delay| in samples between the
43 // signals to compensate for the resampling delay.
44 float ComputeSNR(const ChannelBuffer<float>& ref,
45 const ChannelBuffer<float>& test,
47 VerifyParams(ref, test);
51 // Search within one sample of the expected delay.
52 for (int delay = std::max(expected_delay - 1, 0);
53 delay <= std::min(expected_delay + 1, ref.samples_per_channel());
58 for (int i = 0; i < ref.num_channels(); ++i) {
59 for (int j = 0; j < ref.samples_per_channel() - delay; ++j) {
60 float error = ref.channel(i)[j] - test.channel(i)[j + delay];
62 variance += ref.channel(i)[j] * ref.channel(i)[j];
63 mean += ref.channel(i)[j];
66 const int length = ref.num_channels() * (ref.samples_per_channel() - delay);
70 variance -= mean * mean;
71 float snr = 100; // We assign 100 dB to the zero-error case.
73 snr = 10 * log10(variance / mse);
79 printf("SNR=%.1f dB at delay=%d\n", best_snr, best_delay);
83 // Sets the source to a linearly increasing signal for which we can easily
84 // generate a reference. Runs the AudioConverter and ensures the output has
85 // sufficiently high SNR relative to the reference.
86 void RunAudioConverterTest(int src_channels,
87 int src_sample_rate_hz,
89 int dst_sample_rate_hz) {
90 const float kSrcLeft = 0.0002f;
91 const float kSrcRight = 0.0001f;
92 const float resampling_factor = (1.f * src_sample_rate_hz) /
94 const float dst_left = resampling_factor * kSrcLeft;
95 const float dst_right = resampling_factor * kSrcRight;
96 const float dst_mono = (dst_left + dst_right) / 2;
97 const int src_frames = src_sample_rate_hz / 100;
98 const int dst_frames = dst_sample_rate_hz / 100;
100 std::vector<float> src_data(1, kSrcLeft);
101 if (src_channels == 2)
102 src_data.push_back(kSrcRight);
103 ScopedBuffer src_buffer = CreateBuffer(src_data, src_frames);
105 std::vector<float> dst_data(1, 0);
106 std::vector<float> ref_data;
107 if (dst_channels == 1) {
108 if (src_channels == 1)
109 ref_data.push_back(dst_left);
111 ref_data.push_back(dst_mono);
113 dst_data.push_back(0);
114 ref_data.push_back(dst_left);
115 if (src_channels == 1)
116 ref_data.push_back(dst_left);
118 ref_data.push_back(dst_right);
120 ScopedBuffer dst_buffer = CreateBuffer(dst_data, dst_frames);
121 ScopedBuffer ref_buffer = CreateBuffer(ref_data, dst_frames);
123 // The sinc resampler has a known delay, which we compute here.
124 const int delay_frames = src_sample_rate_hz == dst_sample_rate_hz ? 0 :
125 PushSincResampler::AlgorithmicDelaySeconds(src_sample_rate_hz) *
127 printf("(%d, %d Hz) -> (%d, %d Hz) ", // SNR reported on the same line later.
128 src_channels, src_sample_rate_hz, dst_channels, dst_sample_rate_hz);
130 AudioConverter converter(src_channels, src_frames, dst_channels, dst_frames);
131 converter.Convert(src_buffer->channels(), src_channels, src_frames,
132 dst_channels, dst_frames, dst_buffer->channels());
135 ComputeSNR(*ref_buffer.get(), *dst_buffer.get(), delay_frames));
138 TEST(AudioConverterTest, ConversionsPassSNRThreshold) {
139 const int kSampleRates[] = {8000, 16000, 32000, 44100, 48000};
140 const int kSampleRatesSize = sizeof(kSampleRates) / sizeof(*kSampleRates);
141 const int kChannels[] = {1, 2};
142 const int kChannelsSize = sizeof(kChannels) / sizeof(*kChannels);
143 for (int src_rate = 0; src_rate < kSampleRatesSize; ++src_rate) {
144 for (int dst_rate = 0; dst_rate < kSampleRatesSize; ++dst_rate) {
145 for (int src_channel = 0; src_channel < kChannelsSize; ++src_channel) {
146 for (int dst_channel = 0; dst_channel < kChannelsSize; ++dst_channel) {
147 RunAudioConverterTest(kChannels[src_channel], kSampleRates[src_rate],
148 kChannels[dst_channel], kSampleRates[dst_rate]);
155 } // namespace webrtc