Chromium Code Reviews
chromiumcodereview-hr@appspot.gserviceaccount.com (chromiumcodereview-hr) | Please choose your nickname with Settings | Help | Chromium Project | Gerrit Changes | Sign out
(546)

Side by Side Diff: webrtc/modules/audio_processing/aec3/comfort_noise_generator.cc

Issue 2678423005: Finalization of the first version of EchoCanceller 3 (Closed)
Patch Set: Fixed compilation error Created 3 years, 10 months ago
Use n/p to move between diff chunks; N/P to move between comments. Draft comments are only viewable by you.
Jump to:
View unified diff | Download patch
OLDNEW
(Empty)
1 /*
2 * Copyright (c) 2017 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/aec3/comfort_noise_generator.h"
12
13 #include "webrtc/typedefs.h"
14 #if defined(WEBRTC_ARCH_X86_FAMILY)
15 #include <emmintrin.h>
16 #endif
17 #include <math.h>
18 #include <algorithm>
19 #include <array>
20 #include <functional>
21 #include <numeric>
22
23 #include "webrtc/common_audio/signal_processing/include/signal_processing_librar y.h"
24
25 namespace webrtc {
26
27 namespace {
28
29 // Creates an array of uniformly distributed variables.
30 void TableRandomValue(int16_t* vector, int16_t vector_length, uint32_t* seed) {
31 for (int i = 0; i < vector_length; i++) {
32 seed[0] = (seed[0] * ((int32_t)69069) + 1) & (0x80000000 - 1);
33 vector[i] = (int16_t)(seed[0] >> 16);
34 }
35 }
36
37 } // namespace
38
39 namespace aec3 {
40
41 #if defined(WEBRTC_ARCH_X86_FAMILY)
42
43 void EstimateComfortNoise_SSE2(const std::array<float, kFftLengthBy2Plus1>& N2,
44 uint32_t* seed,
45 FftData* lower_band_noise,
46 FftData* upper_band_noise) {
47 FftData* N_low = lower_band_noise;
48 FftData* N_high = upper_band_noise;
49
50 // Compute square root spectrum.
51 std::array<float, kFftLengthBy2Plus1> N;
52 for (size_t k = 0; k < kFftLengthBy2; k += 4) {
53 __m128 v = _mm_loadu_ps(&N2[k]);
54 v = _mm_sqrt_ps(v);
55 _mm_storeu_ps(&N[k], v);
56 }
57
58 N[kFftLengthBy2] = sqrtf(N2[kFftLengthBy2]);
59
60 // Compute the noise level for the upper bands.
61 constexpr float kOneByNumBands = 1.f / (kFftLengthBy2Plus1 / 2 + 1);
62 constexpr int kFftLengthBy2Plus1By2 = kFftLengthBy2Plus1 / 2;
63 const float high_band_noise_level =
64 std::accumulate(N.begin() + kFftLengthBy2Plus1By2, N.end(), 0.f) *
65 kOneByNumBands;
66
67 // Generate complex noise.
68 std::array<int16_t, kFftLengthBy2 - 1> random_values_int;
69 TableRandomValue(random_values_int.data(), random_values_int.size(), seed);
70
71 std::array<float, kFftLengthBy2 - 1> sin;
72 std::array<float, kFftLengthBy2 - 1> cos;
73 constexpr float kScale = 6.28318530717959f / 32768.0f;
74 std::transform(random_values_int.begin(), random_values_int.end(),
75 sin.begin(), [&](int16_t a) { return -sinf(kScale * a); });
76 std::transform(random_values_int.begin(), random_values_int.end(),
77 cos.begin(), [&](int16_t a) { return cosf(kScale * a); });
78
79 // Form low-frequency noise via spectral shaping.
80 N_low->re[0] = N_low->re[kFftLengthBy2] = N_high->re[0] =
81 N_high->re[kFftLengthBy2] = 0.f;
82 std::transform(cos.begin(), cos.end(), N.begin() + 1, N_low->re.begin() + 1,
83 std::multiplies<float>());
84 std::transform(sin.begin(), sin.end(), N.begin() + 1, N_low->im.begin() + 1,
85 std::multiplies<float>());
86
87 // Form the high-frequency noise via simple levelling.
88 std::transform(cos.begin(), cos.end(), N_high->re.begin() + 1,
89 [&](float a) { return high_band_noise_level * a; });
90 std::transform(sin.begin(), sin.end(), N_high->im.begin() + 1,
91 [&](float a) { return high_band_noise_level * a; });
92 }
93
94 #endif
95
96 void EstimateComfortNoise(const std::array<float, kFftLengthBy2Plus1>& N2,
97 uint32_t* seed,
98 FftData* lower_band_noise,
99 FftData* upper_band_noise) {
100 FftData* N_low = lower_band_noise;
101 FftData* N_high = upper_band_noise;
102
103 // Compute square root spectrum.
104 std::array<float, kFftLengthBy2Plus1> N;
105 std::transform(N2.begin(), N2.end(), N.begin(),
106 [](float a) { return sqrtf(a); });
107
108 // Compute the noise level for the upper bands.
109 constexpr float kOneByNumBands = 1.f / (kFftLengthBy2Plus1 / 2 + 1);
110 constexpr int kFftLengthBy2Plus1By2 = kFftLengthBy2Plus1 / 2;
111 const float high_band_noise_level =
112 std::accumulate(N.begin() + kFftLengthBy2Plus1By2, N.end(), 0.f) *
113 kOneByNumBands;
114
115 // Generate complex noise.
116 std::array<int16_t, kFftLengthBy2 - 1> random_values_int;
117 TableRandomValue(random_values_int.data(), random_values_int.size(), seed);
118
119 std::array<float, kFftLengthBy2 - 1> sin;
120 std::array<float, kFftLengthBy2 - 1> cos;
121 constexpr float kScale = 6.28318530717959f / 32768.0f;
122 std::transform(random_values_int.begin(), random_values_int.end(),
123 sin.begin(), [&](int16_t a) { return -sinf(kScale * a); });
124 std::transform(random_values_int.begin(), random_values_int.end(),
125 cos.begin(), [&](int16_t a) { return cosf(kScale * a); });
126
127 // Form low-frequency noise via spectral shaping.
128 N_low->re[0] = N_low->re[kFftLengthBy2] = N_high->re[0] =
129 N_high->re[kFftLengthBy2] = 0.f;
130 std::transform(cos.begin(), cos.end(), N.begin() + 1, N_low->re.begin() + 1,
131 std::multiplies<float>());
132 std::transform(sin.begin(), sin.end(), N.begin() + 1, N_low->im.begin() + 1,
133 std::multiplies<float>());
134
135 // Form the high-frequency noise via simple levelling.
136 std::transform(cos.begin(), cos.end(), N_high->re.begin() + 1,
137 [&](float a) { return high_band_noise_level * a; });
138 std::transform(sin.begin(), sin.end(), N_high->im.begin() + 1,
139 [&](float a) { return high_band_noise_level * a; });
140 }
141
142 } // namespace aec3
143
144 ComfortNoiseGenerator::ComfortNoiseGenerator(Aec3Optimization optimization)
145 : optimization_(optimization),
146 seed_(42),
147 N2_initial_(new std::array<float, kFftLengthBy2Plus1>()) {
148 N2_initial_->fill(0.f);
149 Y2_smoothed_.fill(0.f);
150 N2_.fill(1.0e6f);
151 }
152
153 ComfortNoiseGenerator::~ComfortNoiseGenerator() = default;
154
155 void ComfortNoiseGenerator::Compute(
156 const AecState& aec_state,
157 const std::array<float, kFftLengthBy2Plus1>& capture_spectrum,
158 FftData* lower_band_noise,
159 FftData* upper_band_noise) {
160 RTC_DCHECK(lower_band_noise);
161 RTC_DCHECK(upper_band_noise);
162 const auto& Y2 = capture_spectrum;
163
164 if (!aec_state.SaturatedCapture()) {
165 // Smooth Y2.
166 std::transform(Y2_smoothed_.begin(), Y2_smoothed_.end(), Y2.begin(),
167 Y2_smoothed_.begin(),
168 [](float a, float b) { return a + 0.1f * (b - a); });
169
170 if (N2_counter_ > 50) {
171 // Update N2 from Y2_smoothed.
172 std::transform(N2_.begin(), N2_.end(), Y2_smoothed_.begin(), N2_.begin(),
173 [](float a, float b) {
174 return b < a ? (0.9f * b + 0.1f * a) * 1.0002f
175 : a * 1.0002f;
176 });
177 }
178
179 if (N2_initial_) {
180 if (++N2_counter_ == 1000) {
181 N2_initial_.reset();
182 } else {
183 // Compute the N2_initial from N2.
184 std::transform(
185 N2_.begin(), N2_.end(), N2_initial_->begin(), N2_initial_->begin(),
186 [](float a, float b) { return a > b ? b + 0.001f * (a - b) : a; });
187 }
188 }
189 }
190
191 // Choose N2 estimate to use.
192 const std::array<float, kFftLengthBy2Plus1>& N2 =
193 N2_initial_ ? *N2_initial_ : N2_;
194
195 switch (optimization_) {
196 #if defined(WEBRTC_ARCH_X86_FAMILY)
197 case Aec3Optimization::kSse2:
198 aec3::EstimateComfortNoise_SSE2(N2, &seed_, lower_band_noise,
199 upper_band_noise);
200 break;
201 #endif
202 default:
203 aec3::EstimateComfortNoise(N2, &seed_, lower_band_noise,
204 upper_band_noise);
205 }
206 }
207
208 } // namespace webrtc
OLDNEW

Powered by Google App Engine
This is Rietveld 408576698