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 int kSampleRate24kHz = 24000; 22 constexpr int kFrameSize10ms24kHz = kSampleRate24kHz / 100; 23 constexpr int kFrameSize20ms24kHz = kFrameSize10ms24kHz * 2; 24 25 // Pitch buffer. 26 constexpr int kMinPitch24kHz = kSampleRate24kHz / 800; // 0.00125 s. 27 constexpr int kMaxPitch24kHz = kSampleRate24kHz / 62.5; // 0.016 s. 28 constexpr int 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 int kInitialMinPitch24kHz = 3 * kMinPitch24kHz; 36 static_assert(kMinPitch24kHz < kInitialMinPitch24kHz, ""); 37 static_assert(kInitialMinPitch24kHz < kMaxPitch24kHz, ""); 38 static_assert(kMaxPitch24kHz > kInitialMinPitch24kHz, ""); 39 // Number of (inverted) lags during the initial pitch search phase at 24 kHz. 40 constexpr int kInitialNumLags24kHz = kMaxPitch24kHz - kInitialMinPitch24kHz; 41 // Number of (inverted) lags during the pitch search refinement phase at 24 kHz. 42 constexpr int kRefineNumLags24kHz = kMaxPitch24kHz + 1; 43 static_assert( 44 kRefineNumLags24kHz > kInitialNumLags24kHz, 45 "The refinement step must search the pitch in an extended pitch range."); 46 47 // 12 kHz analysis. 48 constexpr int kSampleRate12kHz = 12000; 49 constexpr int kFrameSize10ms12kHz = kSampleRate12kHz / 100; 50 constexpr int kFrameSize20ms12kHz = kFrameSize10ms12kHz * 2; 51 constexpr int kBufSize12kHz = kBufSize24kHz / 2; 52 constexpr int kInitialMinPitch12kHz = kInitialMinPitch24kHz / 2; 53 constexpr int kMaxPitch12kHz = kMaxPitch24kHz / 2; 54 static_assert(kMaxPitch12kHz > kInitialMinPitch12kHz, ""); 55 // The inverted lags for the pitch interval [`kInitialMinPitch12kHz`, 56 // `kMaxPitch12kHz`] are in the range [0, `kNumLags12kHz`]. 57 constexpr int kNumLags12kHz = kMaxPitch12kHz - kInitialMinPitch12kHz; 58 59 // 48 kHz constants. 60 constexpr int kMinPitch48kHz = kMinPitch24kHz * 2; 61 constexpr int kMaxPitch48kHz = kMaxPitch24kHz * 2; 62 63 // Spectral features. 64 constexpr int kNumBands = 22; 65 constexpr int kNumLowerBands = 6; 66 static_assert((0 < kNumLowerBands) && (kNumLowerBands < kNumBands), ""); 67 constexpr int kCepstralCoeffsHistorySize = 8; 68 static_assert(kCepstralCoeffsHistorySize > 2, 69 "The history size must at least be 3 to compute first and second " 70 "derivatives."); 71 72 constexpr int kFeatureVectorSize = 42; 73 74 } // namespace rnn_vad 75 } // namespace webrtc 76 77 #endif // MODULES_AUDIO_PROCESSING_AGC2_RNN_VAD_COMMON_H_ 78