Line data Source code
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 "webrtc/modules/audio_coding/neteq/accelerate.h"
12 :
13 : #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
14 :
15 : namespace webrtc {
16 :
17 0 : 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 0 : if (num_channels_ == 0 ||
25 0 : 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 0 : output->PushBackInterleaved(input, input_length);
29 0 : return kError;
30 : }
31 0 : return TimeStretch::Process(input, input_length, fast_accelerate, output,
32 0 : length_change_samples);
33 : }
34 :
35 0 : 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 0 : *best_correlation = 0;
41 0 : }
42 :
43 0 : 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 0 : const int correlation_threshold = fast_mode ? 8192 : kCorrelationThreshold;
54 0 : 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 0 : size_t fs_mult_120 = fs_mult_ * 120;
60 :
61 0 : 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 0 : peak_index = (fs_mult_120 / peak_index) * peak_index;
66 : }
67 :
68 0 : assert(fs_mult_120 >= peak_index); // Should be handled in Process().
69 : // Copy first part; 0 to 15 ms.
70 0 : output->PushBackInterleaved(input, fs_mult_120 * num_channels_);
71 : // Copy the |peak_index| starting at 15 ms to |temp_vector|.
72 0 : AudioMultiVector temp_vector(num_channels_);
73 0 : temp_vector.PushBackInterleaved(&input[fs_mult_120 * num_channels_],
74 0 : peak_index * num_channels_);
75 : // Cross-fade |temp_vector| onto the end of |output|.
76 0 : output->CrossFade(temp_vector, peak_index);
77 : // Copy the last unmodified part, 15 ms + pitch period until the end.
78 0 : output->PushBackInterleaved(
79 0 : &input[(fs_mult_120 + peak_index) * num_channels_],
80 0 : input_length - (fs_mult_120 + peak_index) * num_channels_);
81 :
82 0 : if (active_speech) {
83 0 : return kSuccess;
84 : } else {
85 0 : return kSuccessLowEnergy;
86 : }
87 : } else {
88 : // Accelerate not allowed. Simply move all data from decoded to outData.
89 0 : output->PushBackInterleaved(input, input_length);
90 0 : return kNoStretch;
91 : }
92 : }
93 :
94 0 : Accelerate* AccelerateFactory::Create(
95 : int sample_rate_hz,
96 : size_t num_channels,
97 : const BackgroundNoise& background_noise) const {
98 0 : return new Accelerate(sample_rate_hz, num_channels, background_noise);
99 : }
100 :
101 : } // namespace webrtc
|