tor-browser

The Tor Browser
git clone https://git.dasho.dev/tor-browser.git
Log | Files | Refs | README | LICENSE

accelerate.cc (4318B)


      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 <cstddef>
     14 #include <cstdint>
     15 
     16 #include "api/array_view.h"
     17 #include "modules/audio_coding/neteq/audio_multi_vector.h"
     18 #include "modules/audio_coding/neteq/time_stretch.h"
     19 #include "rtc_base/checks.h"
     20 
     21 namespace webrtc {
     22 
     23 Accelerate::ReturnCodes Accelerate::Process(const int16_t* input,
     24                                            size_t input_length,
     25                                            bool fast_accelerate,
     26                                            AudioMultiVector* output,
     27                                            size_t* length_change_samples) {
     28  // Input length must be (almost) 30 ms.
     29  static const size_t k15ms = 120;  // 15 ms = 120 samples at 8 kHz sample rate.
     30  if (num_channels_ == 0 ||
     31      input_length / num_channels_ < (2 * k15ms - 1) * fs_mult_) {
     32    // Length of input data too short to do accelerate. Simply move all data
     33    // from input to output.
     34    output->PushBackInterleaved(ArrayView<const int16_t>(input, input_length));
     35    return kError;
     36  }
     37  return TimeStretch::Process(input, input_length, fast_accelerate, output,
     38                              length_change_samples);
     39 }
     40 
     41 void Accelerate::SetParametersForPassiveSpeech(size_t /*len*/,
     42                                               int16_t* best_correlation,
     43                                               size_t* /*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 
     49 Accelerate::ReturnCodes Accelerate::CheckCriteriaAndStretch(
     50    const int16_t* input,
     51    size_t input_length,
     52    size_t peak_index,
     53    int16_t best_correlation,
     54    bool active_speech,
     55    bool fast_mode,
     56    AudioMultiVector* output) const {
     57  // Check for strong correlation or passive speech.
     58  // Use 8192 (0.5 in Q14) in fast mode.
     59  const int correlation_threshold = fast_mode ? 8192 : kCorrelationThreshold;
     60  if ((best_correlation > correlation_threshold) || !active_speech) {
     61    // Do accelerate operation by overlap add.
     62 
     63    // Pre-calculate common multiplication with `fs_mult_`.
     64    // 120 corresponds to 15 ms.
     65    size_t fs_mult_120 = fs_mult_ * 120;
     66 
     67    if (fast_mode) {
     68      // Fit as many multiples of `peak_index` as possible in fs_mult_120.
     69      // TODO(henrik.lundin) Consider finding multiple correlation peaks and
     70      // pick the one with the longest correlation lag in this case.
     71      peak_index = (fs_mult_120 / peak_index) * peak_index;
     72    }
     73 
     74    RTC_DCHECK_GE(fs_mult_120, peak_index);  // Should be handled in Process().
     75    // Copy first part; 0 to 15 ms.
     76    output->PushBackInterleaved(
     77        ArrayView<const int16_t>(input, fs_mult_120 * num_channels_));
     78    // Copy the `peak_index` starting at 15 ms to `temp_vector`.
     79    AudioMultiVector temp_vector(num_channels_);
     80    temp_vector.PushBackInterleaved(ArrayView<const int16_t>(
     81        &input[fs_mult_120 * num_channels_], 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(ArrayView<const int16_t>(
     86        &input[(fs_mult_120 + peak_index) * num_channels_],
     87        input_length - (fs_mult_120 + peak_index) * 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(ArrayView<const int16_t>(input, input_length));
     97    return kNoStretch;
     98  }
     99 }
    100 
    101 Accelerate* AccelerateFactory::Create(
    102    int sample_rate_hz,
    103    size_t num_channels,
    104    const BackgroundNoise& background_noise) const {
    105  return new Accelerate(sample_rate_hz, num_channels, background_noise);
    106 }
    107 
    108 }  // namespace webrtc