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 #ifndef WEBRTC_MODULES_AUDIO_CODING_NETEQ_MERGE_H_
     12 #define WEBRTC_MODULES_AUDIO_CODING_NETEQ_MERGE_H_
     13 
     14 #include <assert.h>
     15 
     16 #include "webrtc/base/constructormagic.h"
     17 #include "webrtc/modules/audio_coding/neteq/audio_multi_vector.h"
     18 #include "webrtc/typedefs.h"
     19 
     20 namespace webrtc {
     21 
     22 // Forward declarations.
     23 class Expand;
     24 class SyncBuffer;
     25 
     26 // This class handles the transition from expansion to normal operation.
     27 // When a packet is not available for decoding when needed, the expand operation
     28 // is called to generate extrapolation data. If the missing packet arrives,
     29 // i.e., it was just delayed, it can be decoded and appended directly to the
     30 // end of the expanded data (thanks to how the Expand class operates). However,
     31 // if a later packet arrives instead, the loss is a fact, and the new data must
     32 // be stitched together with the end of the expanded data. This stitching is
     33 // what the Merge class does.
     34 class Merge {
     35  public:
     36   Merge(int fs_hz, size_t num_channels, Expand* expand, SyncBuffer* sync_buffer)
     37       : fs_hz_(fs_hz),
     38         num_channels_(num_channels),
     39         fs_mult_(fs_hz_ / 8000),
     40         timestamps_per_call_(fs_hz_ / 100),
     41         expand_(expand),
     42         sync_buffer_(sync_buffer),
     43         expanded_(num_channels_) {
     44     assert(num_channels_ > 0);
     45   }
     46 
     47   virtual ~Merge() {}
     48 
     49   // The main method to produce the audio data. The decoded data is supplied in
     50   // |input|, having |input_length| samples in total for all channels
     51   // (interleaved). The result is written to |output|. The number of channels
     52   // allocated in |output| defines the number of channels that will be used when
     53   // de-interleaving |input|. The values in |external_mute_factor_array| (Q14)
     54   // will be used to scale the audio, and is updated in the process. The array
     55   // must have |num_channels_| elements.
     56   virtual int Process(int16_t* input, size_t input_length,
     57                       int16_t* external_mute_factor_array,
     58                       AudioMultiVector* output);
     59 
     60   virtual int RequiredFutureSamples();
     61 
     62  protected:
     63   const int fs_hz_;
     64   const size_t num_channels_;
     65 
     66  private:
     67   static const int kMaxSampleRate = 48000;
     68   static const int kExpandDownsampLength = 100;
     69   static const int kInputDownsampLength = 40;
     70   static const int kMaxCorrelationLength = 60;
     71 
     72   // Calls |expand_| to get more expansion data to merge with. The data is
     73   // written to |expanded_signal_|. Returns the length of the expanded data,
     74   // while |expand_period| will be the number of samples in one expansion period
     75   // (typically one pitch period). The value of |old_length| will be the number
     76   // of samples that were taken from the |sync_buffer_|.
     77   int GetExpandedSignal(int* old_length, int* expand_period);
     78 
     79   // Analyzes |input| and |expanded_signal| to find maximum values. Returns
     80   // a muting factor (Q14) to be used on the new data.
     81   int16_t SignalScaling(const int16_t* input, int input_length,
     82                         const int16_t* expanded_signal,
     83                         int16_t* expanded_max, int16_t* input_max) const;
     84 
     85   // Downsamples |input| (|input_length| samples) and |expanded_signal| to
     86   // 4 kHz sample rate. The downsampled signals are written to
     87   // |input_downsampled_| and |expanded_downsampled_|, respectively.
     88   void Downsample(const int16_t* input, int input_length,
     89                   const int16_t* expanded_signal, int expanded_length);
     90 
     91   // Calculates cross-correlation between |input_downsampled_| and
     92   // |expanded_downsampled_|, and finds the correlation maximum. The maximizing
     93   // lag is returned.
     94   int16_t CorrelateAndPeakSearch(int16_t expanded_max, int16_t input_max,
     95                                  int start_position, int input_length,
     96                                  int expand_period) const;
     97 
     98   const int fs_mult_;  // fs_hz_ / 8000.
     99   const int timestamps_per_call_;
    100   Expand* expand_;
    101   SyncBuffer* sync_buffer_;
    102   int16_t expanded_downsampled_[kExpandDownsampLength];
    103   int16_t input_downsampled_[kInputDownsampLength];
    104   AudioMultiVector expanded_;
    105 
    106   DISALLOW_COPY_AND_ASSIGN(Merge);
    107 };
    108 
    109 }  // namespace webrtc
    110 #endif  // WEBRTC_MODULES_AUDIO_CODING_NETEQ_MERGE_H_
    111