OLD | NEW |
1 /* | 1 /* |
2 * Copyright (c) 2014 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 // | 11 // |
12 // Implements core class for intelligibility enhancer. | 12 // Implements core class for intelligibility enhancer. |
13 // | 13 // |
14 // Details of the model and algorithm can be found in the original paper: | 14 // Details of the model and algorithm can be found in the original paper: |
15 // http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6882788 | 15 // http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=6882788 |
16 // | 16 // |
17 | 17 |
18 #include "webrtc/modules/audio_processing/intelligibility/intelligibility_enhanc
er.h" | 18 #include "webrtc/modules/audio_processing/intelligibility/intelligibility_enhanc
er.h" |
19 | 19 |
20 #include <math.h> | 20 #include <math.h> |
21 #include <stdlib.h> | 21 #include <stdlib.h> |
22 #include <algorithm> | 22 #include <algorithm> |
| 23 #include <limits> |
23 #include <numeric> | 24 #include <numeric> |
24 | 25 |
25 #include "webrtc/base/checks.h" | 26 #include "webrtc/base/checks.h" |
26 #include "webrtc/common_audio/include/audio_util.h" | 27 #include "webrtc/common_audio/include/audio_util.h" |
27 #include "webrtc/common_audio/window_generator.h" | 28 #include "webrtc/common_audio/window_generator.h" |
28 | 29 |
29 namespace webrtc { | 30 namespace webrtc { |
30 | 31 |
31 namespace { | 32 namespace { |
32 | 33 |
(...skipping 184 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
217 UpdateErbGains(); | 218 UpdateErbGains(); |
218 } // Else experiencing variance underflow, so do nothing. | 219 } // Else experiencing variance underflow, so do nothing. |
219 } | 220 } |
220 | 221 |
221 void IntelligibilityEnhancer::SolveForLambda(float power_target, | 222 void IntelligibilityEnhancer::SolveForLambda(float power_target, |
222 float power_bot, | 223 float power_bot, |
223 float power_top) { | 224 float power_top) { |
224 const float kConvergeThresh = 0.001f; // TODO(ekmeyerson): Find best values | 225 const float kConvergeThresh = 0.001f; // TODO(ekmeyerson): Find best values |
225 const int kMaxIters = 100; // for these, based on experiments. | 226 const int kMaxIters = 100; // for these, based on experiments. |
226 | 227 |
227 const float reciprocal_power_target = 1.f / power_target; | 228 const float reciprocal_power_target = |
| 229 1.f / (power_target + std::numeric_limits<float>::epsilon()); |
228 float lambda_bot = kLambdaBot; | 230 float lambda_bot = kLambdaBot; |
229 float lambda_top = kLambdaTop; | 231 float lambda_top = kLambdaTop; |
230 float power_ratio = 2.0f; // Ratio of achieved power to target power. | 232 float power_ratio = 2.0f; // Ratio of achieved power to target power. |
231 int iters = 0; | 233 int iters = 0; |
232 while (std::fabs(power_ratio - 1.0f) > kConvergeThresh && | 234 while (std::fabs(power_ratio - 1.0f) > kConvergeThresh && |
233 iters <= kMaxIters) { | 235 iters <= kMaxIters) { |
234 const float lambda = lambda_bot + (lambda_top - lambda_bot) / 2.0f; | 236 const float lambda = lambda_bot + (lambda_top - lambda_bot) / 2.0f; |
235 SolveForGainsGivenLambda(lambda, start_freq_, gains_eq_.get()); | 237 SolveForGainsGivenLambda(lambda, start_freq_, gains_eq_.get()); |
236 const float power = | 238 const float power = |
237 DotProduct(gains_eq_.get(), filtered_clear_var_.get(), bank_size_); | 239 DotProduct(gains_eq_.get(), filtered_clear_var_.get(), bank_size_); |
(...skipping 61 matching lines...) Expand 10 before | Expand all | Expand 10 after Loading... |
299 center_freqs_[min(bank_size_, i + rf) - 1] * num_freqs / | 301 center_freqs_[min(bank_size_, i + rf) - 1] * num_freqs / |
300 (0.5f * sample_rate_hz_))); | 302 (0.5f * sample_rate_hz_))); |
301 rr = static_cast<size_t>(round( | 303 rr = static_cast<size_t>(round( |
302 center_freqs_[min(bank_size_, i + 1) - 1] * num_freqs / | 304 center_freqs_[min(bank_size_, i + 1) - 1] * num_freqs / |
303 (0.5f * sample_rate_hz_))); | 305 (0.5f * sample_rate_hz_))); |
304 rrr = min(num_freqs, max(rrr, kOne)) - 1; | 306 rrr = min(num_freqs, max(rrr, kOne)) - 1; |
305 rr = min(num_freqs, max(rr, kOne)) - 1; | 307 rr = min(num_freqs, max(rr, kOne)) - 1; |
306 | 308 |
307 float step, element; | 309 float step, element; |
308 | 310 |
309 step = 1.0f / (ll - lll); | 311 step = ll == lll ? 0.f : 1.f / (ll - lll); |
310 element = 0.0f; | 312 element = 0.0f; |
311 for (size_t j = lll; j <= ll; ++j) { | 313 for (size_t j = lll; j <= ll; ++j) { |
312 filter_bank[i - 1][j] = element; | 314 filter_bank[i - 1][j] = element; |
313 element += step; | 315 element += step; |
314 } | 316 } |
315 step = 1.0f / (rrr - rr); | 317 step = rr == rrr ? 0.f : 1.f / (rrr - rr); |
316 element = 1.0f; | 318 element = 1.0f; |
317 for (size_t j = rr; j <= rrr; ++j) { | 319 for (size_t j = rr; j <= rrr; ++j) { |
318 filter_bank[i - 1][j] = element; | 320 filter_bank[i - 1][j] = element; |
319 element -= step; | 321 element -= step; |
320 } | 322 } |
321 for (size_t j = ll; j <= rr; ++j) { | 323 for (size_t j = ll; j <= rr; ++j) { |
322 filter_bank[i - 1][j] = 1.0f; | 324 filter_bank[i - 1][j] = 1.0f; |
323 } | 325 } |
324 } | 326 } |
325 | 327 |
(...skipping 23 matching lines...) Expand all Loading... |
349 | 351 |
350 // Analytic solution for optimal gains. See paper for derivation. | 352 // Analytic solution for optimal gains. See paper for derivation. |
351 for (size_t n = start_freq - 1; n < bank_size_; ++n) { | 353 for (size_t n = start_freq - 1; n < bank_size_; ++n) { |
352 float alpha0, beta0, gamma0; | 354 float alpha0, beta0, gamma0; |
353 gamma0 = 0.5f * rho_[n] * var_x0[n] * var_n0[n] + | 355 gamma0 = 0.5f * rho_[n] * var_x0[n] * var_n0[n] + |
354 lambda * var_x0[n] * var_n0[n] * var_n0[n]; | 356 lambda * var_x0[n] * var_n0[n] * var_n0[n]; |
355 beta0 = lambda * var_x0[n] * (2 - rho_[n]) * var_x0[n] * var_n0[n]; | 357 beta0 = lambda * var_x0[n] * (2 - rho_[n]) * var_x0[n] * var_n0[n]; |
356 if (quadratic) { | 358 if (quadratic) { |
357 alpha0 = lambda * var_x0[n] * (1 - rho_[n]) * var_x0[n] * var_x0[n]; | 359 alpha0 = lambda * var_x0[n] * (1 - rho_[n]) * var_x0[n] * var_x0[n]; |
358 sols[n] = | 360 sols[n] = |
359 (-beta0 - sqrtf(beta0 * beta0 - 4 * alpha0 * gamma0)) / (2 * alpha0); | 361 (-beta0 - sqrtf(beta0 * beta0 - 4 * alpha0 * gamma0)) / |
| 362 (2 * alpha0 + std::numeric_limits<float>::epsilon()); |
360 } else { | 363 } else { |
361 sols[n] = -gamma0 / beta0; | 364 sols[n] = -gamma0 / beta0; |
362 } | 365 } |
363 sols[n] = fmax(0, sols[n]); | 366 sols[n] = fmax(0, sols[n]); |
364 } | 367 } |
365 } | 368 } |
366 | 369 |
367 bool IntelligibilityEnhancer::active() const { | 370 bool IntelligibilityEnhancer::active() const { |
368 return active_; | 371 return active_; |
369 } | 372 } |
370 | 373 |
371 } // namespace webrtc | 374 } // namespace webrtc |
OLD | NEW |