1 /* 2 * Copyright (c) 2012 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_CODING_NETEQ_TIME_STRETCH_H_ 12 #define MODULES_AUDIO_CODING_NETEQ_TIME_STRETCH_H_ 13 14 #include <assert.h> 15 #include <string.h> // memset, size_t 16 17 #include "modules/audio_coding/neteq/audio_multi_vector.h" 18 #include "rtc_base/constructor_magic.h" 19 20 namespace webrtc { 21 22 // Forward declarations. 23 class BackgroundNoise; 24 25 // This is the base class for Accelerate and PreemptiveExpand. This class 26 // cannot be instantiated, but must be used through either of the derived 27 // classes. 28 class TimeStretch { 29 public: 30 enum ReturnCodes { 31 kSuccess = 0, 32 kSuccessLowEnergy = 1, 33 kNoStretch = 2, 34 kError = -1 35 }; 36 TimeStretch(int sample_rate_hz,size_t num_channels,const BackgroundNoise & background_noise)37 TimeStretch(int sample_rate_hz, 38 size_t num_channels, 39 const BackgroundNoise& background_noise) 40 : sample_rate_hz_(sample_rate_hz), 41 fs_mult_(sample_rate_hz / 8000), 42 num_channels_(num_channels), 43 background_noise_(background_noise), 44 max_input_value_(0) { 45 assert(sample_rate_hz_ == 8000 || sample_rate_hz_ == 16000 || 46 sample_rate_hz_ == 32000 || sample_rate_hz_ == 48000); 47 assert(num_channels_ > 0); 48 memset(auto_correlation_, 0, sizeof(auto_correlation_)); 49 } 50 ~TimeStretch()51 virtual ~TimeStretch() {} 52 53 // This method performs the processing common to both Accelerate and 54 // PreemptiveExpand. 55 ReturnCodes Process(const int16_t* input, 56 size_t input_len, 57 bool fast_mode, 58 AudioMultiVector* output, 59 size_t* length_change_samples); 60 61 protected: 62 // Sets the parameters |best_correlation| and |peak_index| to suitable 63 // values when the signal contains no active speech. This method must be 64 // implemented by the sub-classes. 65 virtual void SetParametersForPassiveSpeech(size_t input_length, 66 int16_t* best_correlation, 67 size_t* peak_index) const = 0; 68 69 // Checks the criteria for performing the time-stretching operation and, 70 // if possible, performs the time-stretching. This method must be implemented 71 // by the sub-classes. 72 virtual ReturnCodes CheckCriteriaAndStretch( 73 const int16_t* input, 74 size_t input_length, 75 size_t peak_index, 76 int16_t best_correlation, 77 bool active_speech, 78 bool fast_mode, 79 AudioMultiVector* output) const = 0; 80 81 static const size_t kCorrelationLen = 50; 82 static const size_t kLogCorrelationLen = 6; // >= log2(kCorrelationLen). 83 static const size_t kMinLag = 10; 84 static const size_t kMaxLag = 60; 85 static const size_t kDownsampledLen = kCorrelationLen + kMaxLag; 86 static const int kCorrelationThreshold = 14746; // 0.9 in Q14. 87 static constexpr size_t kRefChannel = 0; // First channel is reference. 88 89 const int sample_rate_hz_; 90 const int fs_mult_; // Sample rate multiplier = sample_rate_hz_ / 8000. 91 const size_t num_channels_; 92 const BackgroundNoise& background_noise_; 93 int16_t max_input_value_; 94 int16_t downsampled_input_[kDownsampledLen]; 95 // Adding 1 to the size of |auto_correlation_| because of how it is used 96 // by the peak-detection algorithm. 97 int16_t auto_correlation_[kCorrelationLen + 1]; 98 99 private: 100 // Calculates the auto-correlation of |downsampled_input_| and writes the 101 // result to |auto_correlation_|. 102 void AutoCorrelation(); 103 104 // Performs a simple voice-activity detection based on the input parameters. 105 bool SpeechDetection(int32_t vec1_energy, 106 int32_t vec2_energy, 107 size_t peak_index, 108 int scaling) const; 109 110 RTC_DISALLOW_COPY_AND_ASSIGN(TimeStretch); 111 }; 112 113 } // namespace webrtc 114 #endif // MODULES_AUDIO_CODING_NETEQ_TIME_STRETCH_H_ 115