OLD | NEW |
1 /* | 1 /* |
2 * Copyright (c) 2012 The WebRTC project authors. All Rights Reserved. | 2 * Copyright (c) 2012 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 #include "webrtc/modules/audio_coding/neteq/merge.h" | 11 #include "webrtc/modules/audio_coding/neteq/merge.h" |
12 | 12 |
13 #include <assert.h> | 13 #include <assert.h> |
14 #include <string.h> // memmove, memcpy, memset, size_t | 14 #include <string.h> // memmove, memcpy, memset, size_t |
15 | 15 |
16 #include <algorithm> // min, max | 16 #include <algorithm> // min, max |
17 #include <memory> | 17 #include <memory> |
18 | 18 |
19 #include "webrtc/common_audio/signal_processing/include/signal_processing_librar
y.h" | 19 #include "webrtc/common_audio/signal_processing/include/signal_processing_librar
y.h" |
20 #include "webrtc/modules/audio_coding/neteq/audio_multi_vector.h" | 20 #include "webrtc/modules/audio_coding/neteq/audio_multi_vector.h" |
21 #include "webrtc/modules/audio_coding/neteq/cross_correlation.h" | |
22 #include "webrtc/modules/audio_coding/neteq/dsp_helper.h" | 21 #include "webrtc/modules/audio_coding/neteq/dsp_helper.h" |
23 #include "webrtc/modules/audio_coding/neteq/expand.h" | 22 #include "webrtc/modules/audio_coding/neteq/expand.h" |
24 #include "webrtc/modules/audio_coding/neteq/sync_buffer.h" | 23 #include "webrtc/modules/audio_coding/neteq/sync_buffer.h" |
25 | 24 |
26 namespace webrtc { | 25 namespace webrtc { |
27 | 26 |
28 Merge::Merge(int fs_hz, | 27 Merge::Merge(int fs_hz, |
29 size_t num_channels, | 28 size_t num_channels, |
30 Expand* expand, | 29 Expand* expand, |
31 SyncBuffer* sync_buffer) | 30 SyncBuffer* sync_buffer) |
(...skipping 25 matching lines...) Expand all Loading... |
57 input_vector.PushBackInterleaved(input, input_length); | 56 input_vector.PushBackInterleaved(input, input_length); |
58 size_t input_length_per_channel = input_vector.Size(); | 57 size_t input_length_per_channel = input_vector.Size(); |
59 assert(input_length_per_channel == input_length / num_channels_); | 58 assert(input_length_per_channel == input_length / num_channels_); |
60 | 59 |
61 size_t best_correlation_index = 0; | 60 size_t best_correlation_index = 0; |
62 size_t output_length = 0; | 61 size_t output_length = 0; |
63 | 62 |
64 for (size_t channel = 0; channel < num_channels_; ++channel) { | 63 for (size_t channel = 0; channel < num_channels_; ++channel) { |
65 int16_t* input_channel = &input_vector[channel][0]; | 64 int16_t* input_channel = &input_vector[channel][0]; |
66 int16_t* expanded_channel = &expanded_[channel][0]; | 65 int16_t* expanded_channel = &expanded_[channel][0]; |
| 66 int16_t expanded_max, input_max; |
67 int16_t new_mute_factor = SignalScaling( | 67 int16_t new_mute_factor = SignalScaling( |
68 input_channel, input_length_per_channel, expanded_channel); | 68 input_channel, input_length_per_channel, expanded_channel, |
| 69 &expanded_max, &input_max); |
69 | 70 |
70 // Adjust muting factor (product of "main" muting factor and expand muting | 71 // Adjust muting factor (product of "main" muting factor and expand muting |
71 // factor). | 72 // factor). |
72 int16_t* external_mute_factor = &external_mute_factor_array[channel]; | 73 int16_t* external_mute_factor = &external_mute_factor_array[channel]; |
73 *external_mute_factor = | 74 *external_mute_factor = |
74 (*external_mute_factor * expand_->MuteFactor(channel)) >> 14; | 75 (*external_mute_factor * expand_->MuteFactor(channel)) >> 14; |
75 | 76 |
76 // Update |external_mute_factor| if it is lower than |new_mute_factor|. | 77 // Update |external_mute_factor| if it is lower than |new_mute_factor|. |
77 if (new_mute_factor > *external_mute_factor) { | 78 if (new_mute_factor > *external_mute_factor) { |
78 *external_mute_factor = std::min(new_mute_factor, | 79 *external_mute_factor = std::min(new_mute_factor, |
79 static_cast<int16_t>(16384)); | 80 static_cast<int16_t>(16384)); |
80 } | 81 } |
81 | 82 |
82 if (channel == 0) { | 83 if (channel == 0) { |
83 // Downsample, correlate, and find strongest correlation period for the | 84 // Downsample, correlate, and find strongest correlation period for the |
84 // master (i.e., first) channel only. | 85 // master (i.e., first) channel only. |
85 // Downsample to 4kHz sample rate. | 86 // Downsample to 4kHz sample rate. |
86 Downsample(input_channel, input_length_per_channel, expanded_channel, | 87 Downsample(input_channel, input_length_per_channel, expanded_channel, |
87 expanded_length); | 88 expanded_length); |
88 | 89 |
89 // Calculate the lag of the strongest correlation period. | 90 // Calculate the lag of the strongest correlation period. |
90 best_correlation_index = CorrelateAndPeakSearch( | 91 best_correlation_index = CorrelateAndPeakSearch( |
91 old_length, input_length_per_channel, expand_period); | 92 expanded_max, input_max, old_length, |
| 93 input_length_per_channel, expand_period); |
92 } | 94 } |
93 | 95 |
94 static const int kTempDataSize = 3600; | 96 static const int kTempDataSize = 3600; |
95 int16_t temp_data[kTempDataSize]; // TODO(hlundin) Remove this. | 97 int16_t temp_data[kTempDataSize]; // TODO(hlundin) Remove this. |
96 int16_t* decoded_output = temp_data + best_correlation_index; | 98 int16_t* decoded_output = temp_data + best_correlation_index; |
97 | 99 |
98 // Mute the new decoded data if needed (and unmute it linearly). | 100 // Mute the new decoded data if needed (and unmute it linearly). |
99 // This is the overlapping part of expanded_signal. | 101 // This is the overlapping part of expanded_signal. |
100 size_t interpolation_length = std::min( | 102 size_t interpolation_length = std::min( |
101 kMaxCorrelationLength * fs_mult_, | 103 kMaxCorrelationLength * fs_mult_, |
(...skipping 93 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
195 expanded_.PushBack(expanded_temp); | 197 expanded_.PushBack(expanded_temp); |
196 } | 198 } |
197 // Trim the length to exactly |required_length|. | 199 // Trim the length to exactly |required_length|. |
198 expanded_.PopBack(expanded_.Size() - required_length); | 200 expanded_.PopBack(expanded_.Size() - required_length); |
199 } | 201 } |
200 assert(expanded_.Size() >= required_length); | 202 assert(expanded_.Size() >= required_length); |
201 return required_length; | 203 return required_length; |
202 } | 204 } |
203 | 205 |
204 int16_t Merge::SignalScaling(const int16_t* input, size_t input_length, | 206 int16_t Merge::SignalScaling(const int16_t* input, size_t input_length, |
205 const int16_t* expanded_signal) const { | 207 const int16_t* expanded_signal, |
| 208 int16_t* expanded_max, int16_t* input_max) const { |
206 // Adjust muting factor if new vector is more or less of the BGN energy. | 209 // Adjust muting factor if new vector is more or less of the BGN energy. |
207 const size_t mod_input_length = | 210 const size_t mod_input_length = |
208 std::min(static_cast<size_t>(64 * fs_mult_), input_length); | 211 std::min(static_cast<size_t>(64 * fs_mult_), input_length); |
209 const int16_t expanded_max = | 212 *expanded_max = WebRtcSpl_MaxAbsValueW16(expanded_signal, mod_input_length); |
210 WebRtcSpl_MaxAbsValueW16(expanded_signal, mod_input_length); | 213 *input_max = WebRtcSpl_MaxAbsValueW16(input, mod_input_length); |
211 const int16_t input_max = WebRtcSpl_MaxAbsValueW16(input, mod_input_length); | |
212 | 214 |
213 // Calculate energy of expanded signal. | 215 // Calculate energy of expanded signal. |
214 // |log_fs_mult| is log2(fs_mult_), but is not exact for 48000 Hz. | 216 // |log_fs_mult| is log2(fs_mult_), but is not exact for 48000 Hz. |
215 int log_fs_mult = 30 - WebRtcSpl_NormW32(fs_mult_); | 217 int log_fs_mult = 30 - WebRtcSpl_NormW32(fs_mult_); |
216 int expanded_shift = 6 + log_fs_mult | 218 int expanded_shift = 6 + log_fs_mult |
217 - WebRtcSpl_NormW32(expanded_max * expanded_max); | 219 - WebRtcSpl_NormW32(*expanded_max * *expanded_max); |
218 expanded_shift = std::max(expanded_shift, 0); | 220 expanded_shift = std::max(expanded_shift, 0); |
219 int32_t energy_expanded = WebRtcSpl_DotProductWithScale(expanded_signal, | 221 int32_t energy_expanded = WebRtcSpl_DotProductWithScale(expanded_signal, |
220 expanded_signal, | 222 expanded_signal, |
221 mod_input_length, | 223 mod_input_length, |
222 expanded_shift); | 224 expanded_shift); |
223 | 225 |
224 // Calculate energy of input signal. | 226 // Calculate energy of input signal. |
225 int input_shift = 6 + log_fs_mult - WebRtcSpl_NormW32(input_max * input_max); | 227 int input_shift = 6 + log_fs_mult - |
| 228 WebRtcSpl_NormW32(*input_max * *input_max); |
226 input_shift = std::max(input_shift, 0); | 229 input_shift = std::max(input_shift, 0); |
227 int32_t energy_input = WebRtcSpl_DotProductWithScale(input, input, | 230 int32_t energy_input = WebRtcSpl_DotProductWithScale(input, input, |
228 mod_input_length, | 231 mod_input_length, |
229 input_shift); | 232 input_shift); |
230 | 233 |
231 // Align to the same Q-domain. | 234 // Align to the same Q-domain. |
232 if (input_shift > expanded_shift) { | 235 if (input_shift > expanded_shift) { |
233 energy_expanded = energy_expanded >> (input_shift - expanded_shift); | 236 energy_expanded = energy_expanded >> (input_shift - expanded_shift); |
234 } else { | 237 } else { |
235 energy_input = energy_input >> (expanded_shift - input_shift); | 238 energy_input = energy_input >> (expanded_shift - input_shift); |
(...skipping 61 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
297 sizeof(int16_t) * (kInputDownsampLength - downsamp_temp_len)); | 300 sizeof(int16_t) * (kInputDownsampLength - downsamp_temp_len)); |
298 } else { | 301 } else { |
299 WebRtcSpl_DownsampleFast(&input[signal_offset], | 302 WebRtcSpl_DownsampleFast(&input[signal_offset], |
300 input_length - signal_offset, input_downsampled_, | 303 input_length - signal_offset, input_downsampled_, |
301 kInputDownsampLength, filter_coefficients, | 304 kInputDownsampLength, filter_coefficients, |
302 num_coefficients, decimation_factor, | 305 num_coefficients, decimation_factor, |
303 kCompensateDelay); | 306 kCompensateDelay); |
304 } | 307 } |
305 } | 308 } |
306 | 309 |
307 size_t Merge::CorrelateAndPeakSearch(size_t start_position, size_t input_length, | 310 size_t Merge::CorrelateAndPeakSearch(int16_t expanded_max, int16_t input_max, |
| 311 size_t start_position, size_t input_length, |
308 size_t expand_period) const { | 312 size_t expand_period) const { |
309 // Calculate correlation without any normalization. | 313 // Calculate correlation without any normalization. |
310 const size_t max_corr_length = kMaxCorrelationLength; | 314 const size_t max_corr_length = kMaxCorrelationLength; |
311 size_t stop_position_downsamp = | 315 size_t stop_position_downsamp = |
312 std::min(max_corr_length, expand_->max_lag() / (fs_mult_ * 2) + 1); | 316 std::min(max_corr_length, expand_->max_lag() / (fs_mult_ * 2) + 1); |
| 317 int correlation_shift = 0; |
| 318 if (expanded_max * input_max > 26843546) { |
| 319 correlation_shift = 3; |
| 320 } |
313 | 321 |
314 int32_t correlation[kMaxCorrelationLength]; | 322 int32_t correlation[kMaxCorrelationLength]; |
315 CrossCorrelationWithAutoShift(input_downsampled_, expanded_downsampled_, | 323 WebRtcSpl_CrossCorrelation(correlation, input_downsampled_, |
316 kInputDownsampLength, stop_position_downsamp, 1, | 324 expanded_downsampled_, kInputDownsampLength, |
317 correlation); | 325 stop_position_downsamp, correlation_shift, 1); |
318 | 326 |
319 // Normalize correlation to 14 bits and copy to a 16-bit array. | 327 // Normalize correlation to 14 bits and copy to a 16-bit array. |
320 const size_t pad_length = expand_->overlap_length() - 1; | 328 const size_t pad_length = expand_->overlap_length() - 1; |
321 const size_t correlation_buffer_size = 2 * pad_length + kMaxCorrelationLength; | 329 const size_t correlation_buffer_size = 2 * pad_length + kMaxCorrelationLength; |
322 std::unique_ptr<int16_t[]> correlation16( | 330 std::unique_ptr<int16_t[]> correlation16( |
323 new int16_t[correlation_buffer_size]); | 331 new int16_t[correlation_buffer_size]); |
324 memset(correlation16.get(), 0, correlation_buffer_size * sizeof(int16_t)); | 332 memset(correlation16.get(), 0, correlation_buffer_size * sizeof(int16_t)); |
325 int16_t* correlation_ptr = &correlation16[pad_length]; | 333 int16_t* correlation_ptr = &correlation16[pad_length]; |
326 int32_t max_correlation = WebRtcSpl_MaxAbsValueW32(correlation, | 334 int32_t max_correlation = WebRtcSpl_MaxAbsValueW32(correlation, |
327 stop_position_downsamp); | 335 stop_position_downsamp); |
(...skipping 38 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
366 } | 374 } |
367 return best_correlation_index; | 375 return best_correlation_index; |
368 } | 376 } |
369 | 377 |
370 size_t Merge::RequiredFutureSamples() { | 378 size_t Merge::RequiredFutureSamples() { |
371 return fs_hz_ / 100 * num_channels_; // 10 ms. | 379 return fs_hz_ / 100 * num_channels_; // 10 ms. |
372 } | 380 } |
373 | 381 |
374 | 382 |
375 } // namespace webrtc | 383 } // namespace webrtc |
OLD | NEW |