| OLD | NEW |
| 1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. | 1 // Copyright (c) 2012 The Chromium Authors. All rights reserved. |
| 2 // Use of this source code is governed by a BSD-style license that can be | 2 // Use of this source code is governed by a BSD-style license that can be |
| 3 // found in the LICENSE file. | 3 // found in the LICENSE file. |
| 4 | 4 |
| 5 #include "media/filters/audio_renderer_algorithm.h" | 5 #include "media/filters/audio_renderer_algorithm.h" |
| 6 | 6 |
| 7 #include <algorithm> | 7 #include <algorithm> |
| 8 #include <cmath> | 8 #include <cmath> |
| 9 | 9 |
| 10 #include "base/logging.h" | 10 #include "base/logging.h" |
| (...skipping 24 matching lines...) Expand all Loading... |
| 35 // | 35 // |
| 36 // 4) Update: | 36 // 4) Update: |
| 37 // |optimal_block_| = |transition_window_| * |target_block_| + | 37 // |optimal_block_| = |transition_window_| * |target_block_| + |
| 38 // (1 - |transition_window_|) * |optimal_block_|. | 38 // (1 - |transition_window_|) * |optimal_block_|. |
| 39 // | 39 // |
| 40 // 5) Overlap-and-add |optimal_block_| to the |wsola_output_|. | 40 // 5) Overlap-and-add |optimal_block_| to the |wsola_output_|. |
| 41 // | 41 // |
| 42 // 6) Update: | 42 // 6) Update: |
| 43 // |target_block_| = |optimal_index| + |ola_window_size_| / 2. | 43 // |target_block_| = |optimal_index| + |ola_window_size_| / 2. |
| 44 // |output_index_| = |output_index_| + |ola_window_size_| / 2, | 44 // |output_index_| = |output_index_| + |ola_window_size_| / 2, |
| 45 // |search_block_center_offset_| = |output_index_| * |playback_rate_|, and | 45 // |search_block_center_offset_| = |output_index_| * |playback_rate|, and |
| 46 // |search_block_index_| = |search_block_center_offset_| - | 46 // |search_block_index_| = |search_block_center_offset_| - |
| 47 // |search_block_center_offset_|. | 47 // |search_block_center_offset_|. |
| 48 | 48 |
| 49 // Max/min supported playback rates for fast/slow audio. Audio outside of these | 49 // Max/min supported playback rates for fast/slow audio. Audio outside of these |
| 50 // ranges are muted. | 50 // ranges are muted. |
| 51 // Audio at these speeds would sound better under a frequency domain algorithm. | 51 // Audio at these speeds would sound better under a frequency domain algorithm. |
| 52 static const float kMinPlaybackRate = 0.5f; | 52 static const float kMinPlaybackRate = 0.5f; |
| 53 static const float kMaxPlaybackRate = 4.0f; | 53 static const float kMaxPlaybackRate = 4.0f; |
| 54 | 54 |
| 55 // Overlap-and-add window size in milliseconds. | 55 // Overlap-and-add window size in milliseconds. |
| 56 static const int kOlaWindowSizeMs = 20; | 56 static const int kOlaWindowSizeMs = 20; |
| 57 | 57 |
| 58 // Size of search interval in milliseconds. The search interval is | 58 // Size of search interval in milliseconds. The search interval is |
| 59 // [-delta delta] around |output_index_| * |playback_rate_|. So the search | 59 // [-delta delta] around |output_index_| * |playback_rate|. So the search |
| 60 // interval is 2 * delta. | 60 // interval is 2 * delta. |
| 61 static const int kWsolaSearchIntervalMs = 30; | 61 static const int kWsolaSearchIntervalMs = 30; |
| 62 | 62 |
| 63 // The maximum size in seconds for the |audio_buffer_|. Arbitrarily determined. | 63 // The maximum size in seconds for the |audio_buffer_|. Arbitrarily determined. |
| 64 static const int kMaxCapacityInSeconds = 3; | 64 static const int kMaxCapacityInSeconds = 3; |
| 65 | 65 |
| 66 // The starting size in frames for |audio_buffer_|. Previous usage maintained a | 66 // The starting size in frames for |audio_buffer_|. Previous usage maintained a |
| 67 // queue of 16 AudioBuffers, each of 512 frames. This worked well, so we | 67 // queue of 16 AudioBuffers, each of 512 frames. This worked well, so we |
| 68 // maintain this number of frames. | 68 // maintain this number of frames. |
| 69 static const int kStartingBufferSizeInFrames = 16 * 512; | 69 static const int kStartingBufferSizeInFrames = 16 * 512; |
| 70 | 70 |
| 71 COMPILE_ASSERT(kStartingBufferSizeInFrames < | 71 COMPILE_ASSERT(kStartingBufferSizeInFrames < |
| 72 (kMaxCapacityInSeconds * limits::kMinSampleRate), | 72 (kMaxCapacityInSeconds * limits::kMinSampleRate), |
| 73 max_capacity_smaller_than_starting_buffer_size); | 73 max_capacity_smaller_than_starting_buffer_size); |
| 74 | 74 |
| 75 AudioRendererAlgorithm::AudioRendererAlgorithm() | 75 AudioRendererAlgorithm::AudioRendererAlgorithm() |
| 76 : channels_(0), | 76 : channels_(0), |
| 77 samples_per_second_(0), | 77 samples_per_second_(0), |
| 78 playback_rate_(0), | |
| 79 muted_(false), | |
| 80 muted_partial_frame_(0), | 78 muted_partial_frame_(0), |
| 81 capacity_(kStartingBufferSizeInFrames), | 79 capacity_(kStartingBufferSizeInFrames), |
| 82 output_time_(0.0), | 80 output_time_(0.0), |
| 83 search_block_center_offset_(0), | 81 search_block_center_offset_(0), |
| 84 search_block_index_(0), | 82 search_block_index_(0), |
| 85 num_candidate_blocks_(0), | 83 num_candidate_blocks_(0), |
| 86 target_block_index_(0), | 84 target_block_index_(0), |
| 87 ola_window_size_(0), | 85 ola_window_size_(0), |
| 88 ola_hop_size_(0), | 86 ola_hop_size_(0), |
| 89 num_complete_frames_(0) { | 87 num_complete_frames_(0) { |
| 90 } | 88 } |
| 91 | 89 |
| 92 AudioRendererAlgorithm::~AudioRendererAlgorithm() {} | 90 AudioRendererAlgorithm::~AudioRendererAlgorithm() {} |
| 93 | 91 |
| 94 void AudioRendererAlgorithm::Initialize(float initial_playback_rate, | 92 void AudioRendererAlgorithm::Initialize(const AudioParameters& params) { |
| 95 const AudioParameters& params) { | |
| 96 CHECK(params.IsValid()); | 93 CHECK(params.IsValid()); |
| 97 | 94 |
| 98 channels_ = params.channels(); | 95 channels_ = params.channels(); |
| 99 samples_per_second_ = params.sample_rate(); | 96 samples_per_second_ = params.sample_rate(); |
| 100 SetPlaybackRate(initial_playback_rate); | |
| 101 num_candidate_blocks_ = (kWsolaSearchIntervalMs * samples_per_second_) / 1000; | 97 num_candidate_blocks_ = (kWsolaSearchIntervalMs * samples_per_second_) / 1000; |
| 102 ola_window_size_ = kOlaWindowSizeMs * samples_per_second_ / 1000; | 98 ola_window_size_ = kOlaWindowSizeMs * samples_per_second_ / 1000; |
| 103 | 99 |
| 104 // Make sure window size in an even number. | 100 // Make sure window size in an even number. |
| 105 ola_window_size_ += ola_window_size_ & 1; | 101 ola_window_size_ += ola_window_size_ & 1; |
| 106 ola_hop_size_ = ola_window_size_ / 2; | 102 ola_hop_size_ = ola_window_size_ / 2; |
| 107 | 103 |
| 108 // |num_candidate_blocks_| / 2 is the offset of the center of the search | 104 // |num_candidate_blocks_| / 2 is the offset of the center of the search |
| 109 // block to the center of the first (left most) candidate block. The offset | 105 // block to the center of the first (left most) candidate block. The offset |
| 110 // of the center of a candidate block to its left most point is | 106 // of the center of a candidate block to its left most point is |
| (...skipping 27 matching lines...) Expand all Loading... |
| 138 wsola_output_ = AudioBus::Create(channels_, ola_window_size_ + ola_hop_size_); | 134 wsola_output_ = AudioBus::Create(channels_, ola_window_size_ + ola_hop_size_); |
| 139 wsola_output_->Zero(); // Initialize for overlap-and-add of the first block. | 135 wsola_output_->Zero(); // Initialize for overlap-and-add of the first block. |
| 140 | 136 |
| 141 // Auxiliary containers. | 137 // Auxiliary containers. |
| 142 optimal_block_ = AudioBus::Create(channels_, ola_window_size_); | 138 optimal_block_ = AudioBus::Create(channels_, ola_window_size_); |
| 143 search_block_ = AudioBus::Create( | 139 search_block_ = AudioBus::Create( |
| 144 channels_, num_candidate_blocks_ + (ola_window_size_ - 1)); | 140 channels_, num_candidate_blocks_ + (ola_window_size_ - 1)); |
| 145 target_block_ = AudioBus::Create(channels_, ola_window_size_); | 141 target_block_ = AudioBus::Create(channels_, ola_window_size_); |
| 146 } | 142 } |
| 147 | 143 |
| 148 int AudioRendererAlgorithm::FillBuffer(AudioBus* dest, int requested_frames) { | 144 int AudioRendererAlgorithm::FillBuffer(AudioBus* dest, |
| 149 if (playback_rate_ == 0) | 145 int requested_frames, |
| 146 float playback_rate) { |
| 147 if (playback_rate == 0) |
| 150 return 0; | 148 return 0; |
| 151 | 149 |
| 152 DCHECK_EQ(channels_, dest->channels()); | 150 DCHECK_EQ(channels_, dest->channels()); |
| 153 | 151 |
| 154 // Optimize the |muted_| case to issue a single clear instead of performing | 152 // Optimize the muted case to issue a single clear instead of performing |
| 155 // the full crossfade and clearing each crossfaded frame. | 153 // the full crossfade and clearing each crossfaded frame. |
| 156 if (muted_) { | 154 if (playback_rate < kMinPlaybackRate || playback_rate > kMaxPlaybackRate) { |
| 157 int frames_to_render = | 155 int frames_to_render = |
| 158 std::min(static_cast<int>(audio_buffer_.frames() / playback_rate_), | 156 std::min(static_cast<int>(audio_buffer_.frames() / playback_rate), |
| 159 requested_frames); | 157 requested_frames); |
| 160 | 158 |
| 161 // Compute accurate number of frames to actually skip in the source data. | 159 // Compute accurate number of frames to actually skip in the source data. |
| 162 // Includes the leftover partial frame from last request. However, we can | 160 // Includes the leftover partial frame from last request. However, we can |
| 163 // only skip over complete frames, so a partial frame may remain for next | 161 // only skip over complete frames, so a partial frame may remain for next |
| 164 // time. | 162 // time. |
| 165 muted_partial_frame_ += frames_to_render * playback_rate_; | 163 muted_partial_frame_ += frames_to_render * playback_rate; |
| 166 int seek_frames = static_cast<int>(muted_partial_frame_); | 164 int seek_frames = static_cast<int>(muted_partial_frame_); |
| 167 dest->ZeroFrames(frames_to_render); | 165 dest->ZeroFrames(frames_to_render); |
| 168 audio_buffer_.SeekFrames(seek_frames); | 166 audio_buffer_.SeekFrames(seek_frames); |
| 169 | 167 |
| 170 // Determine the partial frame that remains to be skipped for next call. If | 168 // Determine the partial frame that remains to be skipped for next call. If |
| 171 // the user switches back to playing, it may be off time by this partial | 169 // the user switches back to playing, it may be off time by this partial |
| 172 // frame, which would be undetectable. If they subsequently switch to | 170 // frame, which would be undetectable. If they subsequently switch to |
| 173 // another playback rate that mutes, the code will attempt to line up the | 171 // another playback rate that mutes, the code will attempt to line up the |
| 174 // frames again. | 172 // frames again. |
| 175 muted_partial_frame_ -= seek_frames; | 173 muted_partial_frame_ -= seek_frames; |
| 176 return frames_to_render; | 174 return frames_to_render; |
| 177 } | 175 } |
| 178 | 176 |
| 179 int slower_step = ceil(ola_window_size_ * playback_rate_); | 177 int slower_step = ceil(ola_window_size_ * playback_rate); |
| 180 int faster_step = ceil(ola_window_size_ / playback_rate_); | 178 int faster_step = ceil(ola_window_size_ / playback_rate); |
| 181 | 179 |
| 182 // Optimize the most common |playback_rate_| ~= 1 case to use a single copy | 180 // Optimize the most common |playback_rate| ~= 1 case to use a single copy |
| 183 // instead of copying frame by frame. | 181 // instead of copying frame by frame. |
| 184 if (ola_window_size_ <= faster_step && slower_step >= ola_window_size_) { | 182 if (ola_window_size_ <= faster_step && slower_step >= ola_window_size_) { |
| 185 const int frames_to_copy = | 183 const int frames_to_copy = |
| 186 std::min(audio_buffer_.frames(), requested_frames); | 184 std::min(audio_buffer_.frames(), requested_frames); |
| 187 const int frames_read = audio_buffer_.ReadFrames(frames_to_copy, 0, dest); | 185 const int frames_read = audio_buffer_.ReadFrames(frames_to_copy, 0, dest); |
| 188 DCHECK_EQ(frames_read, frames_to_copy); | 186 DCHECK_EQ(frames_read, frames_to_copy); |
| 189 return frames_read; | 187 return frames_read; |
| 190 } | 188 } |
| 191 | 189 |
| 192 int rendered_frames = 0; | 190 int rendered_frames = 0; |
| 193 do { | 191 do { |
| 194 rendered_frames += WriteCompletedFramesTo( | 192 rendered_frames += WriteCompletedFramesTo( |
| 195 requested_frames - rendered_frames, rendered_frames, dest); | 193 requested_frames - rendered_frames, rendered_frames, dest); |
| 196 } while (rendered_frames < requested_frames && RunOneWsolaIteration()); | 194 } while (rendered_frames < requested_frames && |
| 195 RunOneWsolaIteration(playback_rate)); |
| 197 return rendered_frames; | 196 return rendered_frames; |
| 198 } | 197 } |
| 199 | 198 |
| 200 void AudioRendererAlgorithm::SetPlaybackRate(float new_rate) { | |
| 201 DCHECK_GE(new_rate, 0); | |
| 202 playback_rate_ = new_rate; | |
| 203 muted_ = | |
| 204 playback_rate_ < kMinPlaybackRate || playback_rate_ > kMaxPlaybackRate; | |
| 205 } | |
| 206 | |
| 207 void AudioRendererAlgorithm::FlushBuffers() { | 199 void AudioRendererAlgorithm::FlushBuffers() { |
| 208 // Clear the queue of decoded packets (releasing the buffers). | 200 // Clear the queue of decoded packets (releasing the buffers). |
| 209 audio_buffer_.Clear(); | 201 audio_buffer_.Clear(); |
| 210 output_time_ = 0.0; | 202 output_time_ = 0.0; |
| 211 search_block_index_ = 0; | 203 search_block_index_ = 0; |
| 212 target_block_index_ = 0; | 204 target_block_index_ = 0; |
| 213 wsola_output_->Zero(); | 205 wsola_output_->Zero(); |
| 214 num_complete_frames_ = 0; | 206 num_complete_frames_ = 0; |
| 215 | 207 |
| 216 // Reset |capacity_| so growth triggered by underflows doesn't penalize | 208 // Reset |capacity_| so growth triggered by underflows doesn't penalize |
| (...skipping 22 matching lines...) Expand all Loading... |
| 239 capacity_ = std::min(2 * capacity_, max_capacity); | 231 capacity_ = std::min(2 * capacity_, max_capacity); |
| 240 } | 232 } |
| 241 | 233 |
| 242 bool AudioRendererAlgorithm::CanPerformWsola() const { | 234 bool AudioRendererAlgorithm::CanPerformWsola() const { |
| 243 const int search_block_size = num_candidate_blocks_ + (ola_window_size_ - 1); | 235 const int search_block_size = num_candidate_blocks_ + (ola_window_size_ - 1); |
| 244 const int frames = audio_buffer_.frames(); | 236 const int frames = audio_buffer_.frames(); |
| 245 return target_block_index_ + ola_window_size_ <= frames && | 237 return target_block_index_ + ola_window_size_ <= frames && |
| 246 search_block_index_ + search_block_size <= frames; | 238 search_block_index_ + search_block_size <= frames; |
| 247 } | 239 } |
| 248 | 240 |
| 249 bool AudioRendererAlgorithm::RunOneWsolaIteration() { | 241 bool AudioRendererAlgorithm::RunOneWsolaIteration(float playback_rate) { |
| 250 if (!CanPerformWsola()) | 242 if (!CanPerformWsola()) |
| 251 return false; | 243 return false; |
| 252 | 244 |
| 253 GetOptimalBlock(); | 245 GetOptimalBlock(); |
| 254 | 246 |
| 255 // Overlap-and-add. | 247 // Overlap-and-add. |
| 256 for (int k = 0; k < channels_; ++k) { | 248 for (int k = 0; k < channels_; ++k) { |
| 257 const float* const ch_opt_frame = optimal_block_->channel(k); | 249 const float* const ch_opt_frame = optimal_block_->channel(k); |
| 258 float* ch_output = wsola_output_->channel(k) + num_complete_frames_; | 250 float* ch_output = wsola_output_->channel(k) + num_complete_frames_; |
| 259 for (int n = 0; n < ola_hop_size_; ++n) { | 251 for (int n = 0; n < ola_hop_size_; ++n) { |
| 260 ch_output[n] = ch_output[n] * ola_window_[ola_hop_size_ + n] + | 252 ch_output[n] = ch_output[n] * ola_window_[ola_hop_size_ + n] + |
| 261 ch_opt_frame[n] * ola_window_[n]; | 253 ch_opt_frame[n] * ola_window_[n]; |
| 262 } | 254 } |
| 263 | 255 |
| 264 // Copy the second half to the output. | 256 // Copy the second half to the output. |
| 265 memcpy(&ch_output[ola_hop_size_], &ch_opt_frame[ola_hop_size_], | 257 memcpy(&ch_output[ola_hop_size_], &ch_opt_frame[ola_hop_size_], |
| 266 sizeof(*ch_opt_frame) * ola_hop_size_); | 258 sizeof(*ch_opt_frame) * ola_hop_size_); |
| 267 } | 259 } |
| 268 | 260 |
| 269 num_complete_frames_ += ola_hop_size_; | 261 num_complete_frames_ += ola_hop_size_; |
| 270 UpdateOutputTime(ola_hop_size_); | 262 UpdateOutputTime(playback_rate, ola_hop_size_); |
| 271 RemoveOldInputFrames(); | 263 RemoveOldInputFrames(playback_rate); |
| 272 return true; | 264 return true; |
| 273 } | 265 } |
| 274 | 266 |
| 275 void AudioRendererAlgorithm::UpdateOutputTime(double time_change) { | 267 void AudioRendererAlgorithm::UpdateOutputTime(float playback_rate, |
| 268 double time_change) { |
| 276 output_time_ += time_change; | 269 output_time_ += time_change; |
| 277 // Center of the search region, in frames. | 270 // Center of the search region, in frames. |
| 278 const int search_block_center_index = static_cast<int>( | 271 const int search_block_center_index = static_cast<int>( |
| 279 output_time_ * playback_rate_ + 0.5); | 272 output_time_ * playback_rate + 0.5); |
| 280 search_block_index_ = search_block_center_index - search_block_center_offset_; | 273 search_block_index_ = search_block_center_index - search_block_center_offset_; |
| 281 } | 274 } |
| 282 | 275 |
| 283 void AudioRendererAlgorithm::RemoveOldInputFrames() { | 276 void AudioRendererAlgorithm::RemoveOldInputFrames(float playback_rate) { |
| 284 const int earliest_used_index = std::min(target_block_index_, | 277 const int earliest_used_index = std::min(target_block_index_, |
| 285 search_block_index_); | 278 search_block_index_); |
| 286 if (earliest_used_index <= 0) | 279 if (earliest_used_index <= 0) |
| 287 return; // Nothing to remove. | 280 return; // Nothing to remove. |
| 288 | 281 |
| 289 // Remove frames from input and adjust indices accordingly. | 282 // Remove frames from input and adjust indices accordingly. |
| 290 audio_buffer_.SeekFrames(earliest_used_index); | 283 audio_buffer_.SeekFrames(earliest_used_index); |
| 291 target_block_index_ -= earliest_used_index; | 284 target_block_index_ -= earliest_used_index; |
| 292 | 285 |
| 293 // Adjust output index. | 286 // Adjust output index. |
| 294 double output_time_change = static_cast<double>(earliest_used_index) / | 287 double output_time_change = static_cast<double>(earliest_used_index) / |
| 295 playback_rate_; | 288 playback_rate; |
| 296 CHECK_GE(output_time_, output_time_change); | 289 CHECK_GE(output_time_, output_time_change); |
| 297 UpdateOutputTime(-output_time_change); | 290 UpdateOutputTime(playback_rate, -output_time_change); |
| 298 } | 291 } |
| 299 | 292 |
| 300 int AudioRendererAlgorithm::WriteCompletedFramesTo( | 293 int AudioRendererAlgorithm::WriteCompletedFramesTo( |
| 301 int requested_frames, int dest_offset, AudioBus* dest) { | 294 int requested_frames, int dest_offset, AudioBus* dest) { |
| 302 int rendered_frames = std::min(num_complete_frames_, requested_frames); | 295 int rendered_frames = std::min(num_complete_frames_, requested_frames); |
| 303 | 296 |
| 304 if (rendered_frames == 0) | 297 if (rendered_frames == 0) |
| 305 return 0; // There is nothing to read from |wsola_output_|, return. | 298 return 0; // There is nothing to read from |wsola_output_|, return. |
| 306 | 299 |
| 307 wsola_output_->CopyPartialFramesTo(0, rendered_frames, dest_offset, dest); | 300 wsola_output_->CopyPartialFramesTo(0, rendered_frames, dest_offset, dest); |
| (...skipping 79 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
| 387 read_offset_frames = 0; | 380 read_offset_frames = 0; |
| 388 num_frames_to_read -= num_zero_frames_appended; | 381 num_frames_to_read -= num_zero_frames_appended; |
| 389 write_offset = num_zero_frames_appended; | 382 write_offset = num_zero_frames_appended; |
| 390 dest->ZeroFrames(num_zero_frames_appended); | 383 dest->ZeroFrames(num_zero_frames_appended); |
| 391 } | 384 } |
| 392 audio_buffer_.PeekFrames(num_frames_to_read, read_offset_frames, | 385 audio_buffer_.PeekFrames(num_frames_to_read, read_offset_frames, |
| 393 write_offset, dest); | 386 write_offset, dest); |
| 394 } | 387 } |
| 395 | 388 |
| 396 } // namespace media | 389 } // namespace media |
| OLD | NEW |