Home | History | Annotate | Download | only in neteq
      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/preemptive_expand.h"
     12 
     13 #include <algorithm>  // min, max
     14 
     15 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
     16 
     17 namespace webrtc {
     18 
     19 PreemptiveExpand::ReturnCodes PreemptiveExpand::Process(
     20     const int16_t* input,
     21     int input_length,
     22     int old_data_length,
     23     AudioMultiVector* output,
     24     int16_t* length_change_samples) {
     25   old_data_length_per_channel_ = old_data_length;
     26   // Input length must be (almost) 30 ms.
     27   // Also, the new part must be at least |overlap_samples_| elements.
     28   static const int k15ms = 120;  // 15 ms = 120 samples at 8 kHz sample rate.
     29   if (num_channels_ == 0 ||
     30       input_length / num_channels_ < (2 * k15ms - 1) * fs_mult_ ||
     31       old_data_length >= input_length / num_channels_ - overlap_samples_) {
     32     // Length of input data too short to do preemptive expand. Simply move all
     33     // data from input to output.
     34     output->PushBackInterleaved(input, input_length);
     35     return kError;
     36   }
     37   return TimeStretch::Process(input, input_length, output,
     38                               length_change_samples);
     39 }
     40 
     41 void PreemptiveExpand::SetParametersForPassiveSpeech(size_t len,
     42                                                      int16_t* best_correlation,
     43                                                      int* peak_index) const {
     44   // When the signal does not contain any active speech, the correlation does
     45   // not matter. Simply set it to zero.
     46   *best_correlation = 0;
     47 
     48   // For low energy expansion, the new data can be less than 15 ms,
     49   // but we must ensure that best_correlation is not larger than the length of
     50   // the new data.
     51   // but we must ensure that best_correlation is not larger than the new data.
     52   *peak_index = std::min(*peak_index,
     53                          static_cast<int>(len - old_data_length_per_channel_));
     54 }
     55 
     56 PreemptiveExpand::ReturnCodes PreemptiveExpand::CheckCriteriaAndStretch(
     57     const int16_t *input, size_t input_length, size_t peak_index,
     58     int16_t best_correlation, bool active_speech,
     59     AudioMultiVector* output) const {
     60   // Pre-calculate common multiplication with |fs_mult_|.
     61   // 120 corresponds to 15 ms.
     62   int fs_mult_120 = fs_mult_ * 120;
     63   assert(old_data_length_per_channel_ >= 0);  // Make sure it's been set.
     64   // Check for strong correlation (>0.9 in Q14) and at least 15 ms new data,
     65   // or passive speech.
     66   if (((best_correlation > kCorrelationThreshold) &&
     67       (old_data_length_per_channel_ <= fs_mult_120)) ||
     68       !active_speech) {
     69     // Do accelerate operation by overlap add.
     70 
     71     // Set length of the first part, not to be modified.
     72     size_t unmodified_length = std::max(old_data_length_per_channel_,
     73                                         fs_mult_120);
     74     // Copy first part, including cross-fade region.
     75     output->PushBackInterleaved(
     76         input, (unmodified_length + peak_index) * num_channels_);
     77     // Copy the last |peak_index| samples up to 15 ms to |temp_vector|.
     78     AudioMultiVector temp_vector(num_channels_);
     79     temp_vector.PushBackInterleaved(
     80         &input[(unmodified_length - peak_index) * num_channels_],
     81         peak_index * num_channels_);
     82     // Cross-fade |temp_vector| onto the end of |output|.
     83     output->CrossFade(temp_vector, peak_index);
     84     // Copy the last unmodified part, 15 ms + pitch period until the end.
     85     output->PushBackInterleaved(
     86         &input[unmodified_length * num_channels_],
     87         input_length - unmodified_length * num_channels_);
     88 
     89     if (active_speech) {
     90       return kSuccess;
     91     } else {
     92       return kSuccessLowEnergy;
     93     }
     94   } else {
     95     // Accelerate not allowed. Simply move all data from decoded to outData.
     96     output->PushBackInterleaved(input, input_length);
     97     return kNoStretch;
     98   }
     99 }
    100 
    101 PreemptiveExpand* PreemptiveExpandFactory::Create(
    102     int sample_rate_hz,
    103     size_t num_channels,
    104     const BackgroundNoise& background_noise,
    105     int overlap_samples) const {
    106   return new PreemptiveExpand(
    107       sample_rate_hz, num_channels, background_noise, overlap_samples);
    108 }
    109 
    110 }  // namespace webrtc
    111