1 /* 2 * Copyright (c) 2018 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 #ifndef MODULES_AUDIO_PROCESSING_AGC2_RNN_VAD_COMMON_H_ 12 #define MODULES_AUDIO_PROCESSING_AGC2_RNN_VAD_COMMON_H_ 13 14 #include <stddef.h> 15 16 namespace webrtc { 17 namespace rnn_vad { 18 19 constexpr double kPi = 3.14159265358979323846; 20 21 constexpr size_t kSampleRate24kHz = 24000; 22 constexpr size_t kFrameSize10ms24kHz = kSampleRate24kHz / 100; 23 constexpr size_t kFrameSize20ms24kHz = kFrameSize10ms24kHz * 2; 24 25 // Pitch buffer. 26 constexpr size_t kMinPitch24kHz = kSampleRate24kHz / 800; // 0.00125 s. 27 constexpr size_t kMaxPitch24kHz = kSampleRate24kHz / 62.5; // 0.016 s. 28 constexpr size_t kBufSize24kHz = kMaxPitch24kHz + kFrameSize20ms24kHz; 29 static_assert((kBufSize24kHz & 1) == 0, "The buffer size must be even."); 30 31 // 24 kHz analysis. 32 // Define a higher minimum pitch period for the initial search. This is used to 33 // avoid searching for very short periods, for which a refinement step is 34 // responsible. 35 constexpr size_t kInitialMinPitch24kHz = 3 * kMinPitch24kHz; 36 static_assert(kMinPitch24kHz < kInitialMinPitch24kHz, ""); 37 static_assert(kInitialMinPitch24kHz < kMaxPitch24kHz, ""); 38 static_assert(kMaxPitch24kHz > kInitialMinPitch24kHz, ""); 39 constexpr size_t kNumInvertedLags24kHz = kMaxPitch24kHz - kInitialMinPitch24kHz; 40 41 // 12 kHz analysis. 42 constexpr size_t kSampleRate12kHz = 12000; 43 constexpr size_t kFrameSize10ms12kHz = kSampleRate12kHz / 100; 44 constexpr size_t kFrameSize20ms12kHz = kFrameSize10ms12kHz * 2; 45 constexpr size_t kBufSize12kHz = kBufSize24kHz / 2; 46 constexpr size_t kInitialMinPitch12kHz = kInitialMinPitch24kHz / 2; 47 constexpr size_t kMaxPitch12kHz = kMaxPitch24kHz / 2; 48 static_assert(kMaxPitch12kHz > kInitialMinPitch12kHz, ""); 49 // The inverted lags for the pitch interval [|kInitialMinPitch12kHz|, 50 // |kMaxPitch12kHz|] are in the range [0, |kNumInvertedLags12kHz|]. 51 constexpr size_t kNumInvertedLags12kHz = kMaxPitch12kHz - kInitialMinPitch12kHz; 52 53 // 48 kHz constants. 54 constexpr size_t kMinPitch48kHz = kMinPitch24kHz * 2; 55 constexpr size_t kMaxPitch48kHz = kMaxPitch24kHz * 2; 56 57 // Spectral features. 58 constexpr size_t kNumBands = 22; 59 constexpr size_t kNumLowerBands = 6; 60 static_assert((0 < kNumLowerBands) && (kNumLowerBands < kNumBands), ""); 61 constexpr size_t kCepstralCoeffsHistorySize = 8; 62 static_assert(kCepstralCoeffsHistorySize > 2, 63 "The history size must at least be 3 to compute first and second " 64 "derivatives."); 65 66 constexpr size_t kFeatureVectorSize = 42; 67 68 enum class Optimization { kNone, kSse2, kNeon }; 69 70 // Detects what kind of optimizations to use for the code. 71 Optimization DetectOptimization(); 72 73 } // namespace rnn_vad 74 } // namespace webrtc 75 76 #endif // MODULES_AUDIO_PROCESSING_AGC2_RNN_VAD_COMMON_H_ 77