OLD | NEW |
1 /* | 1 /* |
2 * Copyright (c) 2016 The WebRTC project authors. All Rights Reserved. | 2 * Copyright (c) 2014 The WebRTC project authors. All Rights Reserved. |
3 * | 3 * |
4 * Use of this source code is governed by a BSD-style license | 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 | 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 | 6 * tree. An additional intellectual property rights grant can be found |
7 * in the file PATENTS. All contributing project authors may | 7 * in the file PATENTS. All contributing project authors may |
8 * be found in the AUTHORS file in the root of the source tree. | 8 * be found in the AUTHORS file in the root of the source tree. |
9 */ | 9 */ |
10 | 10 |
11 #ifndef WEBRTC_COMMON_AUDIO_BLOCKER_H_ | 11 #ifndef WEBRTC_INTERNAL_BEAMFORMER_BLOCKER_H_ |
12 #define WEBRTC_COMMON_AUDIO_BLOCKER_H_ | 12 #define WEBRTC_INTERNAL_BEAMFORMER_BLOCKER_H_ |
13 | 13 |
14 // TODO(peah): Remove as soon as all downstream dependencies are resolved. | 14 #include <memory> |
15 #include "webrtc/modules/audio_processing/utility/blocker.h" | |
16 | 15 |
17 #endif // WEBRTC_COMMON_AUDIO_BLOCKER_H_ | 16 #include "webrtc/common_audio/audio_ring_buffer.h" |
| 17 #include "webrtc/common_audio/channel_buffer.h" |
| 18 |
| 19 namespace webrtc { |
| 20 |
| 21 // The callback function to process audio in the time domain. Input has already |
| 22 // been windowed, and output will be windowed. The number of input channels |
| 23 // must be >= the number of output channels. |
| 24 class BlockerCallback { |
| 25 public: |
| 26 virtual ~BlockerCallback() {} |
| 27 |
| 28 virtual void ProcessBlock(const float* const* input, |
| 29 size_t num_frames, |
| 30 size_t num_input_channels, |
| 31 size_t num_output_channels, |
| 32 float* const* output) = 0; |
| 33 }; |
| 34 |
| 35 // The main purpose of Blocker is to abstract away the fact that often we |
| 36 // receive a different number of audio frames than our transform takes. For |
| 37 // example, most FFTs work best when the fft-size is a power of 2, but suppose |
| 38 // we receive 20ms of audio at a sample rate of 48000. That comes to 960 frames |
| 39 // of audio, which is not a power of 2. Blocker allows us to specify the |
| 40 // transform and all other necessary processing via the Process() callback |
| 41 // function without any constraints on the transform-size |
| 42 // (read: |block_size_|) or received-audio-size (read: |chunk_size_|). |
| 43 // We handle this for the multichannel audio case, allowing for different |
| 44 // numbers of input and output channels (for example, beamforming takes 2 or |
| 45 // more input channels and returns 1 output channel). Audio signals are |
| 46 // represented as deinterleaved floats in the range [-1, 1]. |
| 47 // |
| 48 // Blocker is responsible for: |
| 49 // - blocking audio while handling potential discontinuities on the edges |
| 50 // of chunks |
| 51 // - windowing blocks before sending them to Process() |
| 52 // - windowing processed blocks, and overlap-adding them together before |
| 53 // sending back a processed chunk |
| 54 // |
| 55 // To use blocker: |
| 56 // 1. Impelment a BlockerCallback object |bc|. |
| 57 // 2. Instantiate a Blocker object |b|, passing in |bc|. |
| 58 // 3. As you receive audio, call b.ProcessChunk() to get processed audio. |
| 59 // |
| 60 // A small amount of delay is added to the first received chunk to deal with |
| 61 // the difference in chunk/block sizes. This delay is <= chunk_size. |
| 62 // |
| 63 // Ownership of window is retained by the caller. That is, Blocker makes a |
| 64 // copy of window and does not attempt to delete it. |
| 65 class Blocker { |
| 66 public: |
| 67 Blocker(size_t chunk_size, |
| 68 size_t block_size, |
| 69 size_t num_input_channels, |
| 70 size_t num_output_channels, |
| 71 const float* window, |
| 72 size_t shift_amount, |
| 73 BlockerCallback* callback); |
| 74 |
| 75 void ProcessChunk(const float* const* input, |
| 76 size_t chunk_size, |
| 77 size_t num_input_channels, |
| 78 size_t num_output_channels, |
| 79 float* const* output); |
| 80 |
| 81 private: |
| 82 const size_t chunk_size_; |
| 83 const size_t block_size_; |
| 84 const size_t num_input_channels_; |
| 85 const size_t num_output_channels_; |
| 86 |
| 87 // The number of frames of delay to add at the beginning of the first chunk. |
| 88 const size_t initial_delay_; |
| 89 |
| 90 // The frame index into the input buffer where the first block should be read |
| 91 // from. This is necessary because shift_amount_ is not necessarily a |
| 92 // multiple of chunk_size_, so blocks won't line up at the start of the |
| 93 // buffer. |
| 94 size_t frame_offset_; |
| 95 |
| 96 // Since blocks nearly always overlap, there are certain blocks that require |
| 97 // frames from the end of one chunk and the beginning of the next chunk. The |
| 98 // input and output buffers are responsible for saving those frames between |
| 99 // calls to ProcessChunk(). |
| 100 // |
| 101 // Both contain |initial delay| + |chunk_size| frames. The input is a fairly |
| 102 // standard FIFO, but due to the overlap-add it's harder to use an |
| 103 // AudioRingBuffer for the output. |
| 104 AudioRingBuffer input_buffer_; |
| 105 ChannelBuffer<float> output_buffer_; |
| 106 |
| 107 // Space for the input block (can't wrap because of windowing). |
| 108 ChannelBuffer<float> input_block_; |
| 109 |
| 110 // Space for the output block (can't wrap because of overlap/add). |
| 111 ChannelBuffer<float> output_block_; |
| 112 |
| 113 std::unique_ptr<float[]> window_; |
| 114 |
| 115 // The amount of frames between the start of contiguous blocks. For example, |
| 116 // |shift_amount_| = |block_size_| / 2 for a Hann window. |
| 117 size_t shift_amount_; |
| 118 |
| 119 BlockerCallback* callback_; |
| 120 }; |
| 121 |
| 122 } // namespace webrtc |
| 123 |
| 124 #endif // WEBRTC_INTERNAL_BEAMFORMER_BLOCKER_H_ |
OLD | NEW |