1 /*
2  *  Copyright (c) 2014 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 COMMON_AUDIO_LAPPED_TRANSFORM_H_
12 #define COMMON_AUDIO_LAPPED_TRANSFORM_H_
13 
14 #include <complex>
15 #include <memory>
16 
17 #include "common_audio/blocker.h"
18 #include "common_audio/real_fourier.h"
19 #include "system_wrappers/include/aligned_array.h"
20 
21 namespace webrtc {
22 
23 // Helper class for audio processing modules which operate on frequency domain
24 // input derived from the windowed time domain audio stream.
25 //
26 // The input audio chunk is sliced into possibly overlapping blocks, multiplied
27 // by a window and transformed with an FFT implementation. The transformed data
28 // is supplied to the given callback for processing. The processed output is
29 // then inverse transformed into the time domain and spliced back into a chunk
30 // which constitutes the final output of this processing module.
31 class LappedTransform {
32  public:
33   class Callback {
34    public:
~Callback()35     virtual ~Callback() {}
36 
37     virtual void ProcessAudioBlock(const std::complex<float>* const* in_block,
38                                    size_t num_in_channels, size_t frames,
39                                    size_t num_out_channels,
40                                    std::complex<float>* const* out_block) = 0;
41   };
42 
43   // Construct a transform instance. |chunk_length| is the number of samples in
44   // each channel. |window| defines the window, owned by the caller (a copy is
45   // made internally); |window| should have length equal to |block_length|.
46   // |block_length| defines the length of a block, in samples.
47   // |shift_amount| is in samples. |callback| is the caller-owned audio
48   // processing function called for each block of the input chunk.
49   LappedTransform(size_t num_in_channels,
50                   size_t num_out_channels,
51                   size_t chunk_length,
52                   const float* window,
53                   size_t block_length,
54                   size_t shift_amount,
55                   Callback* callback);
56   ~LappedTransform();
57 
58   // Main audio processing helper method. Internally slices |in_chunk| into
59   // blocks, transforms them to frequency domain, calls the callback for each
60   // block and returns a de-blocked time domain chunk of audio through
61   // |out_chunk|. Both buffers are caller-owned.
62   void ProcessChunk(const float* const* in_chunk, float* const* out_chunk);
63 
64   // Get the chunk length.
65   //
66   // The chunk length is the number of samples per channel that must be passed
67   // to ProcessChunk via the parameter in_chunk.
68   //
69   // Returns the same chunk_length passed to the LappedTransform constructor.
chunk_length()70   size_t chunk_length() const { return chunk_length_; }
71 
72   // Get the number of input channels.
73   //
74   // This is the number of arrays that must be passed to ProcessChunk via
75   // in_chunk.
76   //
77   // Returns the same num_in_channels passed to the LappedTransform constructor.
num_in_channels()78   size_t num_in_channels() const { return num_in_channels_; }
79 
80   // Get the number of output channels.
81   //
82   // This is the number of arrays that must be passed to ProcessChunk via
83   // out_chunk.
84   //
85   // Returns the same num_out_channels passed to the LappedTransform
86   // constructor.
num_out_channels()87   size_t num_out_channels() const { return num_out_channels_; }
88 
89   // Returns the initial delay.
90   //
91   // This is the delay introduced by the |blocker_| to be able to get and return
92   // chunks of |chunk_length|, but process blocks of |block_length|.
initial_delay()93   size_t initial_delay() const { return blocker_.initial_delay(); }
94 
95  private:
96   // Internal middleware callback, given to the blocker. Transforms each block
97   // and hands it over to the processing method given at construction time.
98   class BlockThunk : public BlockerCallback {
99    public:
BlockThunk(LappedTransform * parent)100     explicit BlockThunk(LappedTransform* parent) : parent_(parent) {}
101 
102     void ProcessBlock(const float* const* input,
103                       size_t num_frames,
104                       size_t num_input_channels,
105                       size_t num_output_channels,
106                       float* const* output) override;
107 
108    private:
109     LappedTransform* const parent_;
110   } blocker_callback_;
111 
112   const size_t num_in_channels_;
113   const size_t num_out_channels_;
114 
115   const size_t block_length_;
116   const size_t chunk_length_;
117 
118   Callback* const block_processor_;
119   Blocker blocker_;
120 
121   std::unique_ptr<RealFourier> fft_;
122   const size_t cplx_length_;
123   AlignedArray<float> real_buf_;
124   AlignedArray<std::complex<float> > cplx_pre_;
125   AlignedArray<std::complex<float> > cplx_post_;
126 };
127 
128 }  // namespace webrtc
129 
130 #endif  // COMMON_AUDIO_LAPPED_TRANSFORM_H_
131 
132