henrik.lundin@webrtc.org | 9a40081 | 2013-01-29 12:09:21 +0000 | [diff] [blame] | 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 WEBRTC_MODULES_AUDIO_CODING_NETEQ4_TIME_STRETCH_H_ |
| 12 | #define WEBRTC_MODULES_AUDIO_CODING_NETEQ4_TIME_STRETCH_H_ |
| 13 | |
| 14 | #include <assert.h> |
pbos@webrtc.org | 3f45c2e | 2013-08-05 16:22:53 +0000 | [diff] [blame] | 15 | #include <string.h> // memset, size_t |
henrik.lundin@webrtc.org | 9a40081 | 2013-01-29 12:09:21 +0000 | [diff] [blame] | 16 | |
| 17 | #include "webrtc/modules/audio_coding/neteq4/audio_multi_vector.h" |
| 18 | #include "webrtc/system_wrappers/interface/constructor_magic.h" |
| 19 | #include "webrtc/typedefs.h" |
| 20 | |
| 21 | namespace webrtc { |
| 22 | |
| 23 | // Forward declarations. |
| 24 | class BackgroundNoise; |
| 25 | |
| 26 | // This is the base class for Accelerate and PreemptiveExpand. This class |
| 27 | // cannot be instantiated, but must be used through either of the derived |
| 28 | // classes. |
| 29 | class TimeStretch { |
| 30 | public: |
| 31 | enum ReturnCodes { |
| 32 | kSuccess = 0, |
| 33 | kSuccessLowEnergy = 1, |
| 34 | kNoStretch = 2, |
| 35 | kError = -1 |
| 36 | }; |
| 37 | |
| 38 | TimeStretch(int sample_rate_hz, 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 | master_channel_(0), // First channel is master. |
| 44 | background_noise_(background_noise), |
| 45 | max_input_value_(0) { |
| 46 | assert(sample_rate_hz_ == 8000 || |
| 47 | sample_rate_hz_ == 16000 || |
| 48 | sample_rate_hz_ == 32000 || |
| 49 | sample_rate_hz_ == 48000); |
| 50 | assert(num_channels_ > 0); |
| 51 | assert(static_cast<int>(master_channel_) < num_channels_); |
| 52 | memset(auto_correlation_, 0, sizeof(auto_correlation_)); |
| 53 | } |
| 54 | |
| 55 | virtual ~TimeStretch() {} |
| 56 | |
| 57 | // This method performs the processing common to both Accelerate and |
| 58 | // PreemptiveExpand. |
| 59 | ReturnCodes Process(const int16_t* input, |
| 60 | size_t input_len, |
| 61 | AudioMultiVector<int16_t>* output, |
| 62 | int16_t* length_change_samples); |
| 63 | |
| 64 | protected: |
| 65 | // Sets the parameters |best_correlation| and |peak_index| to suitable |
| 66 | // values when the signal contains no active speech. This method must be |
| 67 | // implemented by the sub-classes. |
| 68 | virtual void SetParametersForPassiveSpeech(int input_length, |
| 69 | int16_t* best_correlation, |
| 70 | int* peak_index) const = 0; |
| 71 | |
| 72 | // Checks the criteria for performing the time-stretching operation and, |
| 73 | // if possible, performs the time-stretching. This method must be implemented |
| 74 | // by the sub-classes. |
| 75 | virtual ReturnCodes CheckCriteriaAndStretch( |
| 76 | const int16_t* input, int input_length, size_t peak_index, |
| 77 | int16_t best_correlation, bool active_speech, |
| 78 | AudioMultiVector<int16_t>* output) const = 0; |
| 79 | |
| 80 | static const int kCorrelationLen = 50; |
| 81 | static const int kLogCorrelationLen = 6; // >= log2(kCorrelationLen). |
| 82 | static const int kMinLag = 10; |
| 83 | static const int kMaxLag = 60; |
| 84 | static const int kDownsampledLen = kCorrelationLen + kMaxLag; |
| 85 | static const int kCorrelationThreshold = 14746; // 0.9 in Q14. |
| 86 | |
| 87 | const int sample_rate_hz_; |
| 88 | const int fs_mult_; // Sample rate multiplier = sample_rate_hz_ / 8000. |
| 89 | const int num_channels_; |
| 90 | const size_t master_channel_; |
| 91 | const BackgroundNoise& background_noise_; |
| 92 | int16_t max_input_value_; |
| 93 | int16_t downsampled_input_[kDownsampledLen]; |
| 94 | // Adding 1 to the size of |auto_correlation_| because of how it is used |
| 95 | // by the peak-detection algorithm. |
| 96 | int16_t auto_correlation_[kCorrelationLen + 1]; |
| 97 | |
| 98 | private: |
| 99 | // Calculates the auto-correlation of |downsampled_input_| and writes the |
| 100 | // result to |auto_correlation_|. |
| 101 | void AutoCorrelation(); |
| 102 | |
| 103 | // Performs a simple voice-activity detection based on the input parameters. |
| 104 | bool SpeechDetection(int32_t vec1_energy, int32_t vec2_energy, |
| 105 | int peak_index, int scaling) const; |
| 106 | |
| 107 | DISALLOW_COPY_AND_ASSIGN(TimeStretch); |
| 108 | }; |
| 109 | |
| 110 | } // namespace webrtc |
| 111 | #endif // WEBRTC_MODULES_AUDIO_CODING_NETEQ4_TIME_STRETCH_H_ |