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 #include "modules/audio_coding/neteq/accelerate.h"
12 
13 #include "common_audio/signal_processing/include/signal_processing_library.h"
14 
15 namespace webrtc {
16 
Process(const int16_t * input,size_t input_length,bool fast_accelerate,AudioMultiVector * output,size_t * length_change_samples)17 Accelerate::ReturnCodes Accelerate::Process(const int16_t* input,
18                                             size_t input_length,
19                                             bool fast_accelerate,
20                                             AudioMultiVector* output,
21                                             size_t* length_change_samples) {
22   // Input length must be (almost) 30 ms.
23   static const size_t k15ms = 120;  // 15 ms = 120 samples at 8 kHz sample rate.
24   if (num_channels_ == 0 ||
25       input_length / num_channels_ < (2 * k15ms - 1) * fs_mult_) {
26     // Length of input data too short to do accelerate. Simply move all data
27     // from input to output.
28     output->PushBackInterleaved(input, input_length);
29     return kError;
30   }
31   return TimeStretch::Process(input, input_length, fast_accelerate, output,
32                               length_change_samples);
33 }
34 
SetParametersForPassiveSpeech(size_t,int16_t * best_correlation,size_t *) const35 void Accelerate::SetParametersForPassiveSpeech(size_t /*len*/,
36                                                int16_t* best_correlation,
37                                                size_t* /*peak_index*/) const {
38   // When the signal does not contain any active speech, the correlation does
39   // not matter. Simply set it to zero.
40   *best_correlation = 0;
41 }
42 
CheckCriteriaAndStretch(const int16_t * input,size_t input_length,size_t peak_index,int16_t best_correlation,bool active_speech,bool fast_mode,AudioMultiVector * output) const43 Accelerate::ReturnCodes Accelerate::CheckCriteriaAndStretch(
44     const int16_t* input,
45     size_t input_length,
46     size_t peak_index,
47     int16_t best_correlation,
48     bool active_speech,
49     bool fast_mode,
50     AudioMultiVector* output) const {
51   // Check for strong correlation or passive speech.
52   // Use 8192 (0.5 in Q14) in fast mode.
53   const int correlation_threshold = fast_mode ? 8192 : kCorrelationThreshold;
54   if ((best_correlation > correlation_threshold) || !active_speech) {
55     // Do accelerate operation by overlap add.
56 
57     // Pre-calculate common multiplication with |fs_mult_|.
58     // 120 corresponds to 15 ms.
59     size_t fs_mult_120 = fs_mult_ * 120;
60 
61     if (fast_mode) {
62       // Fit as many multiples of |peak_index| as possible in fs_mult_120.
63       // TODO(henrik.lundin) Consider finding multiple correlation peaks and
64       // pick the one with the longest correlation lag in this case.
65       peak_index = (fs_mult_120 / peak_index) * peak_index;
66     }
67 
68     assert(fs_mult_120 >= peak_index);  // Should be handled in Process().
69     // Copy first part; 0 to 15 ms.
70     output->PushBackInterleaved(input, fs_mult_120 * num_channels_);
71     // Copy the |peak_index| starting at 15 ms to |temp_vector|.
72     AudioMultiVector temp_vector(num_channels_);
73     temp_vector.PushBackInterleaved(&input[fs_mult_120 * num_channels_],
74                                     peak_index * num_channels_);
75     // Cross-fade |temp_vector| onto the end of |output|.
76     output->CrossFade(temp_vector, peak_index);
77     // Copy the last unmodified part, 15 ms + pitch period until the end.
78     output->PushBackInterleaved(
79         &input[(fs_mult_120 + peak_index) * num_channels_],
80         input_length - (fs_mult_120 + peak_index) * num_channels_);
81 
82     if (active_speech) {
83       return kSuccess;
84     } else {
85       return kSuccessLowEnergy;
86     }
87   } else {
88     // Accelerate not allowed. Simply move all data from decoded to outData.
89     output->PushBackInterleaved(input, input_length);
90     return kNoStretch;
91   }
92 }
93 
Create(int sample_rate_hz,size_t num_channels,const BackgroundNoise & background_noise) const94 Accelerate* AccelerateFactory::Create(
95     int sample_rate_hz,
96     size_t num_channels,
97     const BackgroundNoise& background_noise) const {
98   return new Accelerate(sample_rate_hz, num_channels, background_noise);
99 }
100 
101 }  // namespace webrtc
102