Line data Source code
1 : /*
2 : * Copyright (c) 2016 The WebRTC project authors. All Rights Reserved.
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 : #include "webrtc/modules/audio_processing/residual_echo_detector.h"
12 :
13 : #include <algorithm>
14 : #include <numeric>
15 :
16 : #include "webrtc/modules/audio_processing/audio_buffer.h"
17 : #include "webrtc/system_wrappers/include/metrics.h"
18 :
19 : namespace {
20 :
21 0 : float Power(rtc::ArrayView<const float> input) {
22 0 : return std::inner_product(input.begin(), input.end(), input.begin(), 0.f);
23 : }
24 :
25 : constexpr size_t kLookbackFrames = 650;
26 : // TODO(ivoc): Verify the size of this buffer.
27 : constexpr size_t kRenderBufferSize = 30;
28 : constexpr float kAlpha = 0.001f;
29 : // 10 seconds of data, updated every 10 ms.
30 : constexpr size_t kAggregationBufferSize = 10 * 100;
31 :
32 : } // namespace
33 :
34 : namespace webrtc {
35 :
36 0 : ResidualEchoDetector::ResidualEchoDetector()
37 : : render_buffer_(kRenderBufferSize),
38 : render_power_(kLookbackFrames),
39 : render_power_mean_(kLookbackFrames),
40 : render_power_std_dev_(kLookbackFrames),
41 : covariances_(kLookbackFrames),
42 0 : recent_likelihood_max_(kAggregationBufferSize) {}
43 :
44 : ResidualEchoDetector::~ResidualEchoDetector() = default;
45 :
46 0 : void ResidualEchoDetector::AnalyzeRenderAudio(
47 : rtc::ArrayView<const float> render_audio) {
48 0 : if (render_buffer_.Size() == 0) {
49 0 : frames_since_zero_buffer_size_ = 0;
50 0 : } else if (frames_since_zero_buffer_size_ >= kRenderBufferSize) {
51 : // This can happen in a few cases: at the start of a call, due to a glitch
52 : // or due to clock drift. The excess capture value will be ignored.
53 : // TODO(ivoc): Include how often this happens in APM stats.
54 0 : render_buffer_.Pop();
55 0 : frames_since_zero_buffer_size_ = 0;
56 : }
57 0 : ++frames_since_zero_buffer_size_;
58 0 : float power = Power(render_audio);
59 0 : render_buffer_.Push(power);
60 0 : }
61 :
62 0 : void ResidualEchoDetector::AnalyzeCaptureAudio(
63 : rtc::ArrayView<const float> capture_audio) {
64 0 : if (first_process_call_) {
65 : // On the first process call (so the start of a call), we must flush the
66 : // render buffer, otherwise the render data will be delayed.
67 0 : render_buffer_.Clear();
68 0 : first_process_call_ = false;
69 : }
70 :
71 : // Get the next render value.
72 0 : const rtc::Optional<float> buffered_render_power = render_buffer_.Pop();
73 0 : if (!buffered_render_power) {
74 : // This can happen in a few cases: at the start of a call, due to a glitch
75 : // or due to clock drift. The excess capture value will be ignored.
76 : // TODO(ivoc): Include how often this happens in APM stats.
77 0 : return;
78 : }
79 : // Update the render statistics, and store the statistics in circular buffers.
80 0 : render_statistics_.Update(*buffered_render_power);
81 0 : RTC_DCHECK_LT(next_insertion_index_, kLookbackFrames);
82 0 : render_power_[next_insertion_index_] = *buffered_render_power;
83 0 : render_power_mean_[next_insertion_index_] = render_statistics_.mean();
84 0 : render_power_std_dev_[next_insertion_index_] =
85 0 : render_statistics_.std_deviation();
86 :
87 : // Get the next capture value, update capture statistics and add the relevant
88 : // values to the buffers.
89 0 : const float capture_power = Power(capture_audio);
90 0 : capture_statistics_.Update(capture_power);
91 0 : const float capture_mean = capture_statistics_.mean();
92 0 : const float capture_std_deviation = capture_statistics_.std_deviation();
93 :
94 : // Update the covariance values and determine the new echo likelihood.
95 0 : echo_likelihood_ = 0.f;
96 0 : for (size_t delay = 0; delay < covariances_.size(); ++delay) {
97 : const size_t read_index =
98 0 : (kLookbackFrames + next_insertion_index_ - delay) % kLookbackFrames;
99 0 : RTC_DCHECK_LT(read_index, render_power_.size());
100 0 : covariances_[delay].Update(capture_power, capture_mean,
101 0 : capture_std_deviation, render_power_[read_index],
102 0 : render_power_mean_[read_index],
103 0 : render_power_std_dev_[read_index]);
104 0 : echo_likelihood_ = std::max(
105 0 : echo_likelihood_, covariances_[delay].normalized_cross_correlation());
106 : }
107 0 : reliability_ = (1.0f - kAlpha) * reliability_ + kAlpha * 1.0f;
108 0 : echo_likelihood_ *= reliability_;
109 0 : int echo_percentage = static_cast<int>(echo_likelihood_ * 100);
110 0 : RTC_HISTOGRAM_COUNTS("WebRTC.Audio.ResidualEchoDetector.EchoLikelihood",
111 : echo_percentage, 0, 100, 100 /* number of bins */);
112 :
113 : // Update the buffer of recent likelihood values.
114 0 : recent_likelihood_max_.Update(echo_likelihood_);
115 :
116 : // Update the next insertion index.
117 0 : ++next_insertion_index_;
118 0 : next_insertion_index_ %= kLookbackFrames;
119 : }
120 :
121 0 : void ResidualEchoDetector::Initialize() {
122 0 : render_buffer_.Clear();
123 0 : std::fill(render_power_.begin(), render_power_.end(), 0.f);
124 0 : std::fill(render_power_mean_.begin(), render_power_mean_.end(), 0.f);
125 0 : std::fill(render_power_std_dev_.begin(), render_power_std_dev_.end(), 0.f);
126 0 : render_statistics_.Clear();
127 0 : capture_statistics_.Clear();
128 0 : recent_likelihood_max_.Clear();
129 0 : for (auto& cov : covariances_) {
130 0 : cov.Clear();
131 : }
132 0 : echo_likelihood_ = 0.f;
133 0 : next_insertion_index_ = 0;
134 0 : reliability_ = 0.f;
135 0 : }
136 :
137 0 : void ResidualEchoDetector::PackRenderAudioBuffer(
138 : AudioBuffer* audio,
139 : std::vector<float>* packed_buffer) {
140 0 : RTC_DCHECK_GE(160, audio->num_frames_per_band());
141 :
142 0 : packed_buffer->clear();
143 0 : packed_buffer->insert(packed_buffer->end(),
144 0 : audio->split_bands_const_f(0)[kBand0To8kHz],
145 0 : (audio->split_bands_const_f(0)[kBand0To8kHz] +
146 0 : audio->num_frames_per_band()));
147 0 : }
148 :
149 : } // namespace webrtc
|