| OLD | NEW |
|---|
| (Empty) |
| 1 /* | |
| 2 * Copyright (c) 2012 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 #include <math.h> | |
| 12 #include <stdio.h> | |
| 13 #include <string.h> | |
| 14 #ifdef WEBRTC_ANDROID | |
| 15 #include <sys/stat.h> | |
| 16 #endif | |
| 17 | |
| 18 #include <algorithm> | |
| 19 #include <memory> | |
| 20 | |
| 21 #include "webrtc/base/format_macros.h" | |
| 22 #include "webrtc/base/ignore_wundef.h" | |
| 23 #include "webrtc/base/timeutils.h" | |
| 24 #include "webrtc/modules/audio_processing/include/audio_processing.h" | |
| 25 #include "webrtc/modules/audio_processing/include/config.h" | |
| 26 #include "webrtc/modules/audio_processing/test/protobuf_utils.h" | |
| 27 #include "webrtc/modules/audio_processing/test/test_utils.h" | |
| 28 #include "webrtc/modules/include/module_common_types.h" | |
| 29 #include "webrtc/system_wrappers/include/cpu_features_wrapper.h" | |
| 30 #include "webrtc/test/gtest.h" | |
| 31 #include "webrtc/test/testsupport/fileutils.h" | |
| 32 #include "webrtc/test/testsupport/perf_test.h" | |
| 33 | |
| 34 RTC_PUSH_IGNORING_WUNDEF() | |
| 35 #ifdef WEBRTC_ANDROID_PLATFORM_BUILD | |
| 36 #include "external/webrtc/webrtc/modules/audio_processing/debug.pb.h" | |
| 37 #else | |
| 38 #include "webrtc/modules/audio_processing/debug.pb.h" | |
| 39 #endif | |
| 40 RTC_POP_IGNORING_WUNDEF() | |
| 41 | |
| 42 namespace webrtc { | |
| 43 | |
| 44 using webrtc::audioproc::Event; | |
| 45 using webrtc::audioproc::Init; | |
| 46 using webrtc::audioproc::ReverseStream; | |
| 47 using webrtc::audioproc::Stream; | |
| 48 | |
| 49 namespace { | |
| 50 | |
| 51 void PrintStat(const AudioProcessing::Statistic& stat) { | |
| 52 printf("%d, %d, %d\n", stat.average, | |
| 53 stat.maximum, | |
| 54 stat.minimum); | |
| 55 } | |
| 56 | |
| 57 void usage() { | |
| 58 printf( | |
| 59 "Usage: process_test [options] [-pb PROTOBUF_FILE]\n" | |
| 60 " [-ir REVERSE_FILE] [-i PRIMARY_FILE] [-o OUT_FILE]\n"); | |
| 61 printf( | |
| 62 "process_test is a test application for AudioProcessing.\n\n" | |
| 63 "When a protobuf debug file is available, specify it with -pb. Alternately,\n" | |
| 64 "when -ir or -i is used, the specified files will be processed directly in\n" | |
| 65 "a simulation mode. Otherwise the full set of legacy test files is expected\n" | |
| 66 "to be present in the working directory. OUT_FILE should be specified\n" | |
| 67 "without extension to support both raw and wav output.\n\n"); | |
| 68 printf("Options\n"); | |
| 69 printf("General configuration (only used for the simulation mode):\n"); | |
| 70 printf(" -fs SAMPLE_RATE_HZ\n"); | |
| 71 printf(" -ch CHANNELS_IN CHANNELS_OUT\n"); | |
| 72 printf(" -rch REVERSE_CHANNELS\n"); | |
| 73 printf("\n"); | |
| 74 printf("Component configuration:\n"); | |
| 75 printf( | |
| 76 "All components are disabled by default. Each block below begins with a\n" | |
| 77 "flag to enable the component with default settings. The subsequent flags\n" | |
| 78 "in the block are used to provide configuration settings.\n"); | |
| 79 printf("\n -aec Echo cancellation\n"); | |
| 80 printf(" --drift_compensation\n"); | |
| 81 printf(" --no_drift_compensation\n"); | |
| 82 printf(" --no_echo_metrics\n"); | |
| 83 printf(" --no_delay_logging\n"); | |
| 84 printf(" --aec_suppression_level LEVEL [0 - 2]\n"); | |
| 85 printf(" --extended_filter\n"); | |
| 86 printf(" --no_reported_delay\n"); | |
| 87 printf(" --aec3\n"); | |
| 88 printf(" --refined_adaptive_filter\n"); | |
| 89 printf("\n -aecm Echo control mobile\n"); | |
| 90 printf(" --aecm_echo_path_in_file FILE\n"); | |
| 91 printf(" --aecm_echo_path_out_file FILE\n"); | |
| 92 printf(" --no_comfort_noise\n"); | |
| 93 printf(" --routing_mode MODE [0 - 4]\n"); | |
| 94 printf("\n -agc Gain control\n"); | |
| 95 printf(" --analog\n"); | |
| 96 printf(" --adaptive_digital\n"); | |
| 97 printf(" --fixed_digital\n"); | |
| 98 printf(" --target_level LEVEL\n"); | |
| 99 printf(" --compression_gain GAIN\n"); | |
| 100 printf(" --limiter\n"); | |
| 101 printf(" --no_limiter\n"); | |
| 102 printf("\n -hpf High pass filter\n"); | |
| 103 printf("\n -ns Noise suppression\n"); | |
| 104 printf(" --ns_low\n"); | |
| 105 printf(" --ns_moderate\n"); | |
| 106 printf(" --ns_high\n"); | |
| 107 printf(" --ns_very_high\n"); | |
| 108 printf(" --ns_prob_file FILE\n"); | |
| 109 printf("\n -vad Voice activity detection\n"); | |
| 110 printf(" --vad_out_file FILE\n"); | |
| 111 printf("\n -expns Experimental noise suppression\n"); | |
| 112 printf("\n Level metrics (enabled by default)\n"); | |
| 113 printf(" --no_level_metrics\n"); | |
| 114 printf(" --level_control\n"); | |
| 115 printf("\n"); | |
| 116 printf("Modifiers:\n"); | |
| 117 printf(" --noasm Disable SSE optimization.\n"); | |
| 118 printf(" --add_delay DELAY Add DELAY ms to input value.\n"); | |
| 119 printf(" --delay DELAY Override input delay with DELAY ms.\n"); | |
| 120 printf(" --perf Measure performance.\n"); | |
| 121 printf(" --quiet Suppress text output.\n"); | |
| 122 printf(" --no_progress Suppress progress.\n"); | |
| 123 printf(" --raw_output Raw output instead of WAV file.\n"); | |
| 124 printf(" --debug_file FILE Dump a debug recording.\n"); | |
| 125 } | |
| 126 | |
| 127 static float MicLevel2Gain(int level) { | |
| 128 return pow(10.0f, ((level - 127.0f) / 128.0f * 40.0f) / 20.0f); | |
| 129 } | |
| 130 | |
| 131 static void SimulateMic(int mic_level, AudioFrame* frame) { | |
| 132 mic_level = std::min(std::max(mic_level, 0), 255); | |
| 133 float mic_gain = MicLevel2Gain(mic_level); | |
| 134 int num_samples = frame->samples_per_channel_ * frame->num_channels_; | |
| 135 float v; | |
| 136 for (int n = 0; n < num_samples; n++) { | |
| 137 v = floor(frame->data_[n] * mic_gain + 0.5); | |
| 138 v = std::max(std::min(32767.0f, v), -32768.0f); | |
| 139 frame->data_[n] = static_cast<int16_t>(v); | |
| 140 } | |
| 141 } | |
| 142 | |
| 143 // void function for gtest. | |
| 144 void void_main(int argc, char* argv[]) { | |
| 145 if (argc > 1 && strcmp(argv[1], "--help") == 0) { | |
| 146 usage(); | |
| 147 return; | |
| 148 } | |
| 149 | |
| 150 if (argc < 2) { | |
| 151 printf("Did you mean to run without arguments?\n"); | |
| 152 printf("Try `process_test --help' for more information.\n\n"); | |
| 153 } | |
| 154 | |
| 155 std::unique_ptr<AudioProcessing> apm(AudioProcessing::Create()); | |
| 156 ASSERT_TRUE(apm.get() != NULL); | |
| 157 | |
| 158 const char* pb_filename = NULL; | |
| 159 const char* far_filename = NULL; | |
| 160 const char* near_filename = NULL; | |
| 161 std::string out_filename; | |
| 162 const char* vad_out_filename = NULL; | |
| 163 const char* ns_prob_filename = NULL; | |
| 164 const char* aecm_echo_path_in_filename = NULL; | |
| 165 const char* aecm_echo_path_out_filename = NULL; | |
| 166 | |
| 167 int32_t sample_rate_hz = 16000; | |
| 168 | |
| 169 size_t num_capture_input_channels = 1; | |
| 170 size_t num_capture_output_channels = 1; | |
| 171 size_t num_render_channels = 1; | |
| 172 | |
| 173 int samples_per_channel = sample_rate_hz / 100; | |
| 174 | |
| 175 bool simulating = false; | |
| 176 bool perf_testing = false; | |
| 177 bool verbose = true; | |
| 178 bool progress = true; | |
| 179 bool raw_output = false; | |
| 180 int extra_delay_ms = 0; | |
| 181 int override_delay_ms = 0; | |
| 182 Config config; | |
| 183 AudioProcessing::Config apm_config; | |
| 184 | |
| 185 ASSERT_EQ(apm->kNoError, apm->level_estimator()->Enable(true)); | |
| 186 for (int i = 1; i < argc; i++) { | |
| 187 if (strcmp(argv[i], "-pb") == 0) { | |
| 188 i++; | |
| 189 ASSERT_LT(i, argc) << "Specify protobuf filename after -pb"; | |
| 190 pb_filename = argv[i]; | |
| 191 | |
| 192 } else if (strcmp(argv[i], "-ir") == 0) { | |
| 193 i++; | |
| 194 ASSERT_LT(i, argc) << "Specify filename after -ir"; | |
| 195 far_filename = argv[i]; | |
| 196 simulating = true; | |
| 197 | |
| 198 } else if (strcmp(argv[i], "-i") == 0) { | |
| 199 i++; | |
| 200 ASSERT_LT(i, argc) << "Specify filename after -i"; | |
| 201 near_filename = argv[i]; | |
| 202 simulating = true; | |
| 203 | |
| 204 } else if (strcmp(argv[i], "-o") == 0) { | |
| 205 i++; | |
| 206 ASSERT_LT(i, argc) << "Specify filename without extension after -o"; | |
| 207 out_filename = argv[i]; | |
| 208 | |
| 209 } else if (strcmp(argv[i], "-fs") == 0) { | |
| 210 i++; | |
| 211 ASSERT_LT(i, argc) << "Specify sample rate after -fs"; | |
| 212 ASSERT_EQ(1, sscanf(argv[i], "%d", &sample_rate_hz)); | |
| 213 samples_per_channel = sample_rate_hz / 100; | |
| 214 | |
| 215 } else if (strcmp(argv[i], "-ch") == 0) { | |
| 216 i++; | |
| 217 ASSERT_LT(i + 1, argc) << "Specify number of channels after -ch"; | |
| 218 ASSERT_EQ(1, sscanf(argv[i], "%" PRIuS, &num_capture_input_channels)); | |
| 219 i++; | |
| 220 ASSERT_EQ(1, sscanf(argv[i], "%" PRIuS, &num_capture_output_channels)); | |
| 221 | |
| 222 } else if (strcmp(argv[i], "-rch") == 0) { | |
| 223 i++; | |
| 224 ASSERT_LT(i, argc) << "Specify number of channels after -rch"; | |
| 225 ASSERT_EQ(1, sscanf(argv[i], "%" PRIuS, &num_render_channels)); | |
| 226 | |
| 227 } else if (strcmp(argv[i], "-aec") == 0) { | |
| 228 ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true)); | |
| 229 ASSERT_EQ(apm->kNoError, | |
| 230 apm->echo_cancellation()->enable_metrics(true)); | |
| 231 ASSERT_EQ(apm->kNoError, | |
| 232 apm->echo_cancellation()->enable_delay_logging(true)); | |
| 233 | |
| 234 } else if (strcmp(argv[i], "--drift_compensation") == 0) { | |
| 235 ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true)); | |
| 236 // TODO(ajm): this is enabled in the VQE test app by default. Investigate | |
| 237 // why it can give better performance despite passing zeros. | |
| 238 ASSERT_EQ(apm->kNoError, | |
| 239 apm->echo_cancellation()->enable_drift_compensation(true)); | |
| 240 } else if (strcmp(argv[i], "--no_drift_compensation") == 0) { | |
| 241 ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true)); | |
| 242 ASSERT_EQ(apm->kNoError, | |
| 243 apm->echo_cancellation()->enable_drift_compensation(false)); | |
| 244 | |
| 245 } else if (strcmp(argv[i], "--no_echo_metrics") == 0) { | |
| 246 ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true)); | |
| 247 ASSERT_EQ(apm->kNoError, | |
| 248 apm->echo_cancellation()->enable_metrics(false)); | |
| 249 | |
| 250 } else if (strcmp(argv[i], "--no_delay_logging") == 0) { | |
| 251 ASSERT_EQ(apm->kNoError, apm->echo_cancellation()->Enable(true)); | |
| 252 ASSERT_EQ(apm->kNoError, | |
| 253 apm->echo_cancellation()->enable_delay_logging(false)); | |
| 254 | |
| 255 } else if (strcmp(argv[i], "--no_level_metrics") == 0) { | |
| 256 ASSERT_EQ(apm->kNoError, apm->level_estimator()->Enable(false)); | |
| 257 | |
| 258 } else if (strcmp(argv[i], "--aec_suppression_level") == 0) { | |
| 259 i++; | |
| 260 ASSERT_LT(i, argc) << "Specify level after --aec_suppression_level"; | |
| 261 int suppression_level; | |
| 262 ASSERT_EQ(1, sscanf(argv[i], "%d", &suppression_level)); | |
| 263 ASSERT_EQ(apm->kNoError, | |
| 264 apm->echo_cancellation()->set_suppression_level( | |
| 265 static_cast<webrtc::EchoCancellation::SuppressionLevel>( | |
| 266 suppression_level))); | |
| 267 | |
| 268 } else if (strcmp(argv[i], "--level_control") == 0) { | |
| 269 apm_config.level_controller.enabled = true; | |
| 270 } else if (strcmp(argv[i], "--extended_filter") == 0) { | |
| 271 config.Set<ExtendedFilter>(new ExtendedFilter(true)); | |
| 272 | |
| 273 } else if (strcmp(argv[i], "--no_reported_delay") == 0) { | |
| 274 config.Set<DelayAgnostic>(new DelayAgnostic(true)); | |
| 275 | |
| 276 } else if (strcmp(argv[i], "--delay_agnostic") == 0) { | |
| 277 config.Set<DelayAgnostic>(new DelayAgnostic(true)); | |
| 278 | |
| 279 } else if (strcmp(argv[i], "--aec3") == 0) { | |
| 280 config.Set<EchoCanceller3>(new EchoCanceller3(true)); | |
| 281 | |
| 282 } else if (strcmp(argv[i], "--refined_adaptive_filter") == 0) { | |
| 283 config.Set<RefinedAdaptiveFilter>(new RefinedAdaptiveFilter(true)); | |
| 284 | |
| 285 } else if (strcmp(argv[i], "-aecm") == 0) { | |
| 286 ASSERT_EQ(apm->kNoError, apm->echo_control_mobile()->Enable(true)); | |
| 287 | |
| 288 } else if (strcmp(argv[i], "--aecm_echo_path_in_file") == 0) { | |
| 289 i++; | |
| 290 ASSERT_LT(i, argc) << "Specify filename after --aecm_echo_path_in_file"; | |
| 291 aecm_echo_path_in_filename = argv[i]; | |
| 292 | |
| 293 } else if (strcmp(argv[i], "--aecm_echo_path_out_file") == 0) { | |
| 294 i++; | |
| 295 ASSERT_LT(i, argc) << "Specify filename after --aecm_echo_path_out_file"; | |
| 296 aecm_echo_path_out_filename = argv[i]; | |
| 297 | |
| 298 } else if (strcmp(argv[i], "--no_comfort_noise") == 0) { | |
| 299 ASSERT_EQ(apm->kNoError, | |
| 300 apm->echo_control_mobile()->enable_comfort_noise(false)); | |
| 301 | |
| 302 } else if (strcmp(argv[i], "--routing_mode") == 0) { | |
| 303 i++; | |
| 304 ASSERT_LT(i, argc) << "Specify mode after --routing_mode"; | |
| 305 int routing_mode; | |
| 306 ASSERT_EQ(1, sscanf(argv[i], "%d", &routing_mode)); | |
| 307 ASSERT_EQ(apm->kNoError, | |
| 308 apm->echo_control_mobile()->set_routing_mode( | |
| 309 static_cast<webrtc::EchoControlMobile::RoutingMode>( | |
| 310 routing_mode))); | |
| 311 | |
| 312 } else if (strcmp(argv[i], "-agc") == 0) { | |
| 313 ASSERT_EQ(apm->kNoError, apm->gain_control()->Enable(true)); | |
| 314 | |
| 315 } else if (strcmp(argv[i], "--analog") == 0) { | |
| 316 ASSERT_EQ(apm->kNoError, apm->gain_control()->Enable(true)); | |
| 317 ASSERT_EQ(apm->kNoError, | |
| 318 apm->gain_control()->set_mode(GainControl::kAdaptiveAnalog)); | |
| 319 | |
| 320 } else if (strcmp(argv[i], "--adaptive_digital") == 0) { | |
| 321 ASSERT_EQ(apm->kNoError, apm->gain_control()->Enable(true)); | |
| 322 ASSERT_EQ(apm->kNoError, | |
| 323 apm->gain_control()->set_mode(GainControl::kAdaptiveDigital)); | |
| 324 | |
| 325 } else if (strcmp(argv[i], "--fixed_digital") == 0) { | |
| 326 ASSERT_EQ(apm->kNoError, apm->gain_control()->Enable(true)); | |
| 327 ASSERT_EQ(apm->kNoError, | |
| 328 apm->gain_control()->set_mode(GainControl::kFixedDigital)); | |
| 329 | |
| 330 } else if (strcmp(argv[i], "--target_level") == 0) { | |
| 331 i++; | |
| 332 int level; | |
| 333 ASSERT_EQ(1, sscanf(argv[i], "%d", &level)); | |
| 334 | |
| 335 ASSERT_EQ(apm->kNoError, apm->gain_control()->Enable(true)); | |
| 336 ASSERT_EQ(apm->kNoError, | |
| 337 apm->gain_control()->set_target_level_dbfs(level)); | |
| 338 | |
| 339 } else if (strcmp(argv[i], "--compression_gain") == 0) { | |
| 340 i++; | |
| 341 int gain; | |
| 342 ASSERT_EQ(1, sscanf(argv[i], "%d", &gain)); | |
| 343 | |
| 344 ASSERT_EQ(apm->kNoError, apm->gain_control()->Enable(true)); | |
| 345 ASSERT_EQ(apm->kNoError, | |
| 346 apm->gain_control()->set_compression_gain_db(gain)); | |
| 347 | |
| 348 } else if (strcmp(argv[i], "--limiter") == 0) { | |
| 349 ASSERT_EQ(apm->kNoError, apm->gain_control()->Enable(true)); | |
| 350 ASSERT_EQ(apm->kNoError, | |
| 351 apm->gain_control()->enable_limiter(true)); | |
| 352 | |
| 353 } else if (strcmp(argv[i], "--no_limiter") == 0) { | |
| 354 ASSERT_EQ(apm->kNoError, apm->gain_control()->Enable(true)); | |
| 355 ASSERT_EQ(apm->kNoError, | |
| 356 apm->gain_control()->enable_limiter(false)); | |
| 357 | |
| 358 } else if (strcmp(argv[i], "-hpf") == 0) { | |
| 359 ASSERT_EQ(apm->kNoError, apm->high_pass_filter()->Enable(true)); | |
| 360 | |
| 361 } else if (strcmp(argv[i], "-ns") == 0) { | |
| 362 ASSERT_EQ(apm->kNoError, apm->noise_suppression()->Enable(true)); | |
| 363 | |
| 364 } else if (strcmp(argv[i], "--ns_low") == 0) { | |
| 365 ASSERT_EQ(apm->kNoError, apm->noise_suppression()->Enable(true)); | |
| 366 ASSERT_EQ(apm->kNoError, | |
| 367 apm->noise_suppression()->set_level(NoiseSuppression::kLow)); | |
| 368 | |
| 369 } else if (strcmp(argv[i], "--ns_moderate") == 0) { | |
| 370 ASSERT_EQ(apm->kNoError, apm->noise_suppression()->Enable(true)); | |
| 371 ASSERT_EQ(apm->kNoError, | |
| 372 apm->noise_suppression()->set_level(NoiseSuppression::kModerate)); | |
| 373 | |
| 374 } else if (strcmp(argv[i], "--ns_high") == 0) { | |
| 375 ASSERT_EQ(apm->kNoError, apm->noise_suppression()->Enable(true)); | |
| 376 ASSERT_EQ(apm->kNoError, | |
| 377 apm->noise_suppression()->set_level(NoiseSuppression::kHigh)); | |
| 378 | |
| 379 } else if (strcmp(argv[i], "--ns_very_high") == 0) { | |
| 380 ASSERT_EQ(apm->kNoError, apm->noise_suppression()->Enable(true)); | |
| 381 ASSERT_EQ(apm->kNoError, | |
| 382 apm->noise_suppression()->set_level(NoiseSuppression::kVeryHigh)); | |
| 383 | |
| 384 } else if (strcmp(argv[i], "--ns_prob_file") == 0) { | |
| 385 i++; | |
| 386 ASSERT_LT(i, argc) << "Specify filename after --ns_prob_file"; | |
| 387 ns_prob_filename = argv[i]; | |
| 388 | |
| 389 } else if (strcmp(argv[i], "-vad") == 0) { | |
| 390 ASSERT_EQ(apm->kNoError, apm->voice_detection()->Enable(true)); | |
| 391 | |
| 392 } else if (strcmp(argv[i], "--vad_very_low") == 0) { | |
| 393 ASSERT_EQ(apm->kNoError, apm->voice_detection()->Enable(true)); | |
| 394 ASSERT_EQ(apm->kNoError, | |
| 395 apm->voice_detection()->set_likelihood( | |
| 396 VoiceDetection::kVeryLowLikelihood)); | |
| 397 | |
| 398 } else if (strcmp(argv[i], "--vad_low") == 0) { | |
| 399 ASSERT_EQ(apm->kNoError, apm->voice_detection()->Enable(true)); | |
| 400 ASSERT_EQ(apm->kNoError, | |
| 401 apm->voice_detection()->set_likelihood( | |
| 402 VoiceDetection::kLowLikelihood)); | |
| 403 | |
| 404 } else if (strcmp(argv[i], "--vad_moderate") == 0) { | |
| 405 ASSERT_EQ(apm->kNoError, apm->voice_detection()->Enable(true)); | |
| 406 ASSERT_EQ(apm->kNoError, | |
| 407 apm->voice_detection()->set_likelihood( | |
| 408 VoiceDetection::kModerateLikelihood)); | |
| 409 | |
| 410 } else if (strcmp(argv[i], "--vad_high") == 0) { | |
| 411 ASSERT_EQ(apm->kNoError, apm->voice_detection()->Enable(true)); | |
| 412 ASSERT_EQ(apm->kNoError, | |
| 413 apm->voice_detection()->set_likelihood( | |
| 414 VoiceDetection::kHighLikelihood)); | |
| 415 | |
| 416 } else if (strcmp(argv[i], "--vad_out_file") == 0) { | |
| 417 i++; | |
| 418 ASSERT_LT(i, argc) << "Specify filename after --vad_out_file"; | |
| 419 vad_out_filename = argv[i]; | |
| 420 | |
| 421 } else if (strcmp(argv[i], "-expns") == 0) { | |
| 422 config.Set<ExperimentalNs>(new ExperimentalNs(true)); | |
| 423 | |
| 424 } else if (strcmp(argv[i], "--noasm") == 0) { | |
| 425 WebRtc_GetCPUInfo = WebRtc_GetCPUInfoNoASM; | |
| 426 // We need to reinitialize here if components have already been enabled. | |
| 427 ASSERT_EQ(apm->kNoError, apm->Initialize()); | |
| 428 | |
| 429 } else if (strcmp(argv[i], "--add_delay") == 0) { | |
| 430 i++; | |
| 431 ASSERT_EQ(1, sscanf(argv[i], "%d", &extra_delay_ms)); | |
| 432 | |
| 433 } else if (strcmp(argv[i], "--delay") == 0) { | |
| 434 i++; | |
| 435 ASSERT_EQ(1, sscanf(argv[i], "%d", &override_delay_ms)); | |
| 436 | |
| 437 } else if (strcmp(argv[i], "--perf") == 0) { | |
| 438 perf_testing = true; | |
| 439 | |
| 440 } else if (strcmp(argv[i], "--quiet") == 0) { | |
| 441 verbose = false; | |
| 442 progress = false; | |
| 443 | |
| 444 } else if (strcmp(argv[i], "--no_progress") == 0) { | |
| 445 progress = false; | |
| 446 | |
| 447 } else if (strcmp(argv[i], "--raw_output") == 0) { | |
| 448 raw_output = true; | |
| 449 | |
| 450 } else if (strcmp(argv[i], "--debug_file") == 0) { | |
| 451 i++; | |
| 452 ASSERT_LT(i, argc) << "Specify filename after --debug_file"; | |
| 453 ASSERT_EQ(apm->kNoError, apm->StartDebugRecording(argv[i], -1)); | |
| 454 } else { | |
| 455 FAIL() << "Unrecognized argument " << argv[i]; | |
| 456 } | |
| 457 } | |
| 458 apm->ApplyConfig(apm_config); | |
| 459 apm->SetExtraOptions(config); | |
| 460 | |
| 461 // If we're reading a protobuf file, ensure a simulation hasn't also | |
| 462 // been requested (which makes no sense...) | |
| 463 ASSERT_FALSE(pb_filename && simulating); | |
| 464 | |
| 465 if (verbose) { | |
| 466 printf("Sample rate: %d Hz\n", sample_rate_hz); | |
| 467 printf("Primary channels: %" PRIuS " (in), %" PRIuS " (out)\n", | |
| 468 num_capture_input_channels, | |
| 469 num_capture_output_channels); | |
| 470 printf("Reverse channels: %" PRIuS "\n", num_render_channels); | |
| 471 } | |
| 472 | |
| 473 const std::string out_path = webrtc::test::OutputPath(); | |
| 474 const char far_file_default[] = "apm_far.pcm"; | |
| 475 const char near_file_default[] = "apm_near.pcm"; | |
| 476 const char event_filename[] = "apm_event.dat"; | |
| 477 const char delay_filename[] = "apm_delay.dat"; | |
| 478 const char drift_filename[] = "apm_drift.dat"; | |
| 479 const std::string vad_file_default = out_path + "vad_out.dat"; | |
| 480 const std::string ns_prob_file_default = out_path + "ns_prob.dat"; | |
| 481 | |
| 482 if (!simulating) { | |
| 483 far_filename = far_file_default; | |
| 484 near_filename = near_file_default; | |
| 485 } | |
| 486 | |
| 487 if (out_filename.size() == 0) { | |
| 488 out_filename = out_path + "out"; | |
| 489 } | |
| 490 | |
| 491 if (!vad_out_filename) { | |
| 492 vad_out_filename = vad_file_default.c_str(); | |
| 493 } | |
| 494 | |
| 495 if (!ns_prob_filename) { | |
| 496 ns_prob_filename = ns_prob_file_default.c_str(); | |
| 497 } | |
| 498 | |
| 499 FILE* pb_file = NULL; | |
| 500 FILE* far_file = NULL; | |
| 501 FILE* near_file = NULL; | |
| 502 FILE* event_file = NULL; | |
| 503 FILE* delay_file = NULL; | |
| 504 FILE* drift_file = NULL; | |
| 505 FILE* vad_out_file = NULL; | |
| 506 FILE* ns_prob_file = NULL; | |
| 507 FILE* aecm_echo_path_in_file = NULL; | |
| 508 FILE* aecm_echo_path_out_file = NULL; | |
| 509 | |
| 510 std::unique_ptr<WavWriter> output_wav_file; | |
| 511 std::unique_ptr<RawFile> output_raw_file; | |
| 512 | |
| 513 if (pb_filename) { | |
| 514 pb_file = OpenFile(pb_filename, "rb"); | |
| 515 } else { | |
| 516 if (far_filename) { | |
| 517 far_file = OpenFile(far_filename, "rb"); | |
| 518 } | |
| 519 | |
| 520 near_file = OpenFile(near_filename, "rb"); | |
| 521 if (!simulating) { | |
| 522 event_file = OpenFile(event_filename, "rb"); | |
| 523 delay_file = OpenFile(delay_filename, "rb"); | |
| 524 drift_file = OpenFile(drift_filename, "rb"); | |
| 525 } | |
| 526 } | |
| 527 | |
| 528 int near_size_bytes = 0; | |
| 529 if (pb_file) { | |
| 530 struct stat st; | |
| 531 stat(pb_filename, &st); | |
| 532 // Crude estimate, but should be good enough. | |
| 533 near_size_bytes = st.st_size / 3; | |
| 534 } else { | |
| 535 struct stat st; | |
| 536 stat(near_filename, &st); | |
| 537 near_size_bytes = st.st_size; | |
| 538 } | |
| 539 | |
| 540 if (apm->voice_detection()->is_enabled()) { | |
| 541 vad_out_file = OpenFile(vad_out_filename, "wb"); | |
| 542 } | |
| 543 | |
| 544 if (apm->noise_suppression()->is_enabled()) { | |
| 545 ns_prob_file = OpenFile(ns_prob_filename, "wb"); | |
| 546 } | |
| 547 | |
| 548 if (aecm_echo_path_in_filename != NULL) { | |
| 549 aecm_echo_path_in_file = OpenFile(aecm_echo_path_in_filename, "rb"); | |
| 550 | |
| 551 const size_t path_size = | |
| 552 apm->echo_control_mobile()->echo_path_size_bytes(); | |
| 553 std::unique_ptr<char[]> echo_path(new char[path_size]); | |
| 554 ASSERT_EQ(path_size, fread(echo_path.get(), | |
| 555 sizeof(char), | |
| 556 path_size, | |
| 557 aecm_echo_path_in_file)); | |
| 558 EXPECT_EQ(apm->kNoError, | |
| 559 apm->echo_control_mobile()->SetEchoPath(echo_path.get(), | |
| 560 path_size)); | |
| 561 fclose(aecm_echo_path_in_file); | |
| 562 aecm_echo_path_in_file = NULL; | |
| 563 } | |
| 564 | |
| 565 if (aecm_echo_path_out_filename != NULL) { | |
| 566 aecm_echo_path_out_file = OpenFile(aecm_echo_path_out_filename, "wb"); | |
| 567 } | |
| 568 | |
| 569 size_t read_count = 0; | |
| 570 int reverse_count = 0; | |
| 571 int primary_count = 0; | |
| 572 int near_read_bytes = 0; | |
| 573 int64_t acc_nanos = 0; | |
| 574 | |
| 575 AudioFrame far_frame; | |
| 576 AudioFrame near_frame; | |
| 577 | |
| 578 int delay_ms = 0; | |
| 579 int drift_samples = 0; | |
| 580 int capture_level = 127; | |
| 581 int8_t stream_has_voice = 0; | |
| 582 float ns_speech_prob = 0.0f; | |
| 583 | |
| 584 int64_t t0 = rtc::TimeNanos(); | |
| 585 int64_t t1 = t0; | |
| 586 int64_t max_time_us = 0; | |
| 587 int64_t max_time_reverse_us = 0; | |
| 588 int64_t min_time_us = 1e6; | |
| 589 int64_t min_time_reverse_us = 1e6; | |
| 590 | |
| 591 // TODO(ajm): Ideally we would refactor this block into separate functions, | |
| 592 // but for now we want to share the variables. | |
| 593 if (pb_file) { | |
| 594 Event event_msg; | |
| 595 std::unique_ptr<ChannelBuffer<float> > reverse_cb; | |
| 596 std::unique_ptr<ChannelBuffer<float> > primary_cb; | |
| 597 int output_sample_rate = 32000; | |
| 598 AudioProcessing::ChannelLayout output_layout = AudioProcessing::kMono; | |
| 599 while (ReadMessageFromFile(pb_file, &event_msg)) { | |
| 600 std::ostringstream trace_stream; | |
| 601 trace_stream << "Processed frames: " << reverse_count << " (reverse), " | |
| 602 << primary_count << " (primary)"; | |
| 603 SCOPED_TRACE(trace_stream.str()); | |
| 604 | |
| 605 if (event_msg.type() == Event::INIT) { | |
| 606 ASSERT_TRUE(event_msg.has_init()); | |
| 607 const Init msg = event_msg.init(); | |
| 608 | |
| 609 ASSERT_TRUE(msg.has_sample_rate()); | |
| 610 ASSERT_TRUE(msg.has_num_input_channels()); | |
| 611 ASSERT_TRUE(msg.has_num_output_channels()); | |
| 612 ASSERT_TRUE(msg.has_num_reverse_channels()); | |
| 613 int reverse_sample_rate = msg.sample_rate(); | |
| 614 if (msg.has_reverse_sample_rate()) { | |
| 615 reverse_sample_rate = msg.reverse_sample_rate(); | |
| 616 } | |
| 617 output_sample_rate = msg.sample_rate(); | |
| 618 if (msg.has_output_sample_rate()) { | |
| 619 output_sample_rate = msg.output_sample_rate(); | |
| 620 } | |
| 621 output_layout = | |
| 622 LayoutFromChannels(static_cast<size_t>(msg.num_output_channels())); | |
| 623 ASSERT_EQ(kNoErr, | |
| 624 apm->Initialize( | |
| 625 msg.sample_rate(), | |
| 626 output_sample_rate, | |
| 627 reverse_sample_rate, | |
| 628 LayoutFromChannels( | |
| 629 static_cast<size_t>(msg.num_input_channels())), | |
| 630 output_layout, | |
| 631 LayoutFromChannels( | |
| 632 static_cast<size_t>(msg.num_reverse_channels())))); | |
| 633 | |
| 634 samples_per_channel = msg.sample_rate() / 100; | |
| 635 far_frame.sample_rate_hz_ = reverse_sample_rate; | |
| 636 far_frame.samples_per_channel_ = reverse_sample_rate / 100; | |
| 637 far_frame.num_channels_ = msg.num_reverse_channels(); | |
| 638 near_frame.sample_rate_hz_ = msg.sample_rate(); | |
| 639 near_frame.samples_per_channel_ = samples_per_channel; | |
| 640 near_frame.num_channels_ = msg.num_input_channels(); | |
| 641 reverse_cb.reset(new ChannelBuffer<float>( | |
| 642 far_frame.samples_per_channel_, | |
| 643 msg.num_reverse_channels())); | |
| 644 primary_cb.reset(new ChannelBuffer<float>(samples_per_channel, | |
| 645 msg.num_input_channels())); | |
| 646 | |
| 647 if (verbose) { | |
| 648 printf("Init at frame: %d (primary), %d (reverse)\n", | |
| 649 primary_count, reverse_count); | |
| 650 printf(" Primary rates: %d Hz (in), %d Hz (out)\n", | |
| 651 msg.sample_rate(), output_sample_rate); | |
| 652 printf(" Primary channels: %d (in), %d (out)\n", | |
| 653 msg.num_input_channels(), | |
| 654 msg.num_output_channels()); | |
| 655 printf(" Reverse rate: %d\n", reverse_sample_rate); | |
| 656 printf(" Reverse channels: %d\n", msg.num_reverse_channels()); | |
| 657 } | |
| 658 | |
| 659 if (!raw_output) { | |
| 660 // The WAV file needs to be reset every time, because it can't change | |
| 661 // its sample rate or number of channels. | |
| 662 output_wav_file.reset(new WavWriter( | |
| 663 out_filename + ".wav", output_sample_rate, | |
| 664 static_cast<size_t>(msg.num_output_channels()))); | |
| 665 } | |
| 666 | |
| 667 } else if (event_msg.type() == Event::REVERSE_STREAM) { | |
| 668 ASSERT_TRUE(event_msg.has_reverse_stream()); | |
| 669 ReverseStream msg = event_msg.reverse_stream(); | |
| 670 reverse_count++; | |
| 671 | |
| 672 ASSERT_TRUE(msg.has_data() ^ (msg.channel_size() > 0)); | |
| 673 if (msg.has_data()) { | |
| 674 ASSERT_EQ(sizeof(int16_t) * far_frame.samples_per_channel_ * | |
| 675 far_frame.num_channels_, msg.data().size()); | |
| 676 memcpy(far_frame.data_, msg.data().data(), msg.data().size()); | |
| 677 } else { | |
| 678 for (int i = 0; i < msg.channel_size(); ++i) { | |
| 679 memcpy(reverse_cb->channels()[i], | |
| 680 msg.channel(i).data(), | |
| 681 reverse_cb->num_frames() * | |
| 682 sizeof(reverse_cb->channels()[i][0])); | |
| 683 } | |
| 684 } | |
| 685 | |
| 686 if (perf_testing) { | |
| 687 t0 = rtc::TimeNanos(); | |
| 688 } | |
| 689 | |
| 690 if (msg.has_data()) { | |
| 691 ASSERT_EQ(apm->kNoError, | |
| 692 apm->ProcessReverseStream(&far_frame)); | |
| 693 } else { | |
| 694 ASSERT_EQ(apm->kNoError, | |
| 695 apm->AnalyzeReverseStream( | |
| 696 reverse_cb->channels(), | |
| 697 far_frame.samples_per_channel_, | |
| 698 far_frame.sample_rate_hz_, | |
| 699 LayoutFromChannels(far_frame.num_channels_))); | |
| 700 } | |
| 701 | |
| 702 if (perf_testing) { | |
| 703 t1 = rtc::TimeNanos(); | |
| 704 int64_t diff_nanos = t1 - t0; | |
| 705 acc_nanos += diff_nanos; | |
| 706 int64_t diff_us = diff_nanos / rtc::kNumNanosecsPerMicrosec; | |
| 707 if (diff_us > max_time_reverse_us) { | |
| 708 max_time_reverse_us = diff_us; | |
| 709 } | |
| 710 if (diff_us < min_time_reverse_us) { | |
| 711 min_time_reverse_us = diff_us; | |
| 712 } | |
| 713 } | |
| 714 | |
| 715 } else if (event_msg.type() == Event::STREAM) { | |
| 716 ASSERT_TRUE(event_msg.has_stream()); | |
| 717 const Stream msg = event_msg.stream(); | |
| 718 primary_count++; | |
| 719 | |
| 720 ASSERT_TRUE(msg.has_input_data() ^ (msg.input_channel_size() > 0)); | |
| 721 if (msg.has_input_data()) { | |
| 722 ASSERT_EQ(sizeof(int16_t) * samples_per_channel * | |
| 723 near_frame.num_channels_, msg.input_data().size()); | |
| 724 memcpy(near_frame.data_, | |
| 725 msg.input_data().data(), | |
| 726 msg.input_data().size()); | |
| 727 near_read_bytes += msg.input_data().size(); | |
| 728 } else { | |
| 729 for (int i = 0; i < msg.input_channel_size(); ++i) { | |
| 730 memcpy(primary_cb->channels()[i], | |
| 731 msg.input_channel(i).data(), | |
| 732 primary_cb->num_frames() * | |
| 733 sizeof(primary_cb->channels()[i][0])); | |
| 734 near_read_bytes += msg.input_channel(i).size(); | |
| 735 } | |
| 736 } | |
| 737 | |
| 738 if (progress && primary_count % 100 == 0) { | |
| 739 near_read_bytes = std::min(near_read_bytes, near_size_bytes); | |
| 740 printf("%.0f%% complete\r", | |
| 741 (near_read_bytes * 100.0) / near_size_bytes); | |
| 742 fflush(stdout); | |
| 743 } | |
| 744 | |
| 745 if (perf_testing) { | |
| 746 t0 = rtc::TimeNanos(); | |
| 747 } | |
| 748 | |
| 749 ASSERT_EQ(apm->kNoError, | |
| 750 apm->gain_control()->set_stream_analog_level(msg.level())); | |
| 751 delay_ms = msg.delay() + extra_delay_ms; | |
| 752 if (override_delay_ms) { | |
| 753 delay_ms = override_delay_ms; | |
| 754 } | |
| 755 ASSERT_EQ(apm->kNoError, | |
| 756 apm->set_stream_delay_ms(delay_ms)); | |
| 757 apm->echo_cancellation()->set_stream_drift_samples(msg.drift()); | |
| 758 | |
| 759 if (msg.has_keypress()) { | |
| 760 apm->set_stream_key_pressed(msg.keypress()); | |
| 761 } else { | |
| 762 apm->set_stream_key_pressed(true); | |
| 763 } | |
| 764 | |
| 765 int err = apm->kNoError; | |
| 766 if (msg.has_input_data()) { | |
| 767 err = apm->ProcessStream(&near_frame); | |
| 768 ASSERT_TRUE(near_frame.num_channels_ == apm->num_output_channels()); | |
| 769 } else { | |
| 770 err = apm->ProcessStream( | |
| 771 primary_cb->channels(), | |
| 772 near_frame.samples_per_channel_, | |
| 773 near_frame.sample_rate_hz_, | |
| 774 LayoutFromChannels(near_frame.num_channels_), | |
| 775 output_sample_rate, | |
| 776 output_layout, | |
| 777 primary_cb->channels()); | |
| 778 } | |
| 779 | |
| 780 if (err == apm->kBadStreamParameterWarning) { | |
| 781 printf("Bad parameter warning. %s\n", trace_stream.str().c_str()); | |
| 782 } | |
| 783 ASSERT_TRUE(err == apm->kNoError || | |
| 784 err == apm->kBadStreamParameterWarning); | |
| 785 | |
| 786 stream_has_voice = | |
| 787 static_cast<int8_t>(apm->voice_detection()->stream_has_voice()); | |
| 788 if (vad_out_file != NULL) { | |
| 789 ASSERT_EQ(1u, fwrite(&stream_has_voice, | |
| 790 sizeof(stream_has_voice), | |
| 791 1, | |
| 792 vad_out_file)); | |
| 793 } | |
| 794 | |
| 795 if (ns_prob_file != NULL) { | |
| 796 ns_speech_prob = apm->noise_suppression()->speech_probability(); | |
| 797 ASSERT_EQ(1u, fwrite(&ns_speech_prob, | |
| 798 sizeof(ns_speech_prob), | |
| 799 1, | |
| 800 ns_prob_file)); | |
| 801 } | |
| 802 | |
| 803 if (perf_testing) { | |
| 804 t1 = rtc::TimeNanos(); | |
| 805 int64_t diff_nanos = t1 - t0; | |
| 806 acc_nanos += diff_nanos; | |
| 807 int64_t diff_us = diff_nanos / rtc::kNumNanosecsPerMicrosec; | |
| 808 if (diff_us > max_time_us) { | |
| 809 max_time_us = diff_us; | |
| 810 } | |
| 811 if (diff_us < min_time_us) { | |
| 812 min_time_us = diff_us; | |
| 813 } | |
| 814 } | |
| 815 | |
| 816 const size_t samples_per_channel = output_sample_rate / 100; | |
| 817 if (msg.has_input_data()) { | |
| 818 if (raw_output && !output_raw_file) { | |
| 819 output_raw_file.reset(new RawFile(out_filename + ".pcm")); | |
| 820 } | |
| 821 WriteIntData(near_frame.data_, | |
| 822 apm->num_output_channels() * samples_per_channel, | |
| 823 output_wav_file.get(), | |
| 824 output_raw_file.get()); | |
| 825 } else { | |
| 826 if (raw_output && !output_raw_file) { | |
| 827 output_raw_file.reset(new RawFile(out_filename + ".float")); | |
| 828 } | |
| 829 WriteFloatData(primary_cb->channels(), | |
| 830 samples_per_channel, | |
| 831 apm->num_output_channels(), | |
| 832 output_wav_file.get(), | |
| 833 output_raw_file.get()); | |
| 834 } | |
| 835 } | |
| 836 } | |
| 837 | |
| 838 ASSERT_TRUE(feof(pb_file)); | |
| 839 | |
| 840 } else { | |
| 841 enum Events { | |
| 842 kInitializeEvent, | |
| 843 kRenderEvent, | |
| 844 kCaptureEvent, | |
| 845 kResetEventDeprecated | |
| 846 }; | |
| 847 int16_t event = 0; | |
| 848 while (simulating || feof(event_file) == 0) { | |
| 849 std::ostringstream trace_stream; | |
| 850 trace_stream << "Processed frames: " << reverse_count << " (reverse), " | |
| 851 << primary_count << " (primary)"; | |
| 852 SCOPED_TRACE(trace_stream.str()); | |
| 853 | |
| 854 if (simulating) { | |
| 855 if (far_file == NULL) { | |
| 856 event = kCaptureEvent; | |
| 857 } else { | |
| 858 event = (event == kCaptureEvent) ? kRenderEvent : kCaptureEvent; | |
| 859 } | |
| 860 } else { | |
| 861 read_count = fread(&event, sizeof(event), 1, event_file); | |
| 862 if (read_count != 1) { | |
| 863 break; | |
| 864 } | |
| 865 } | |
| 866 | |
| 867 far_frame.sample_rate_hz_ = sample_rate_hz; | |
| 868 far_frame.samples_per_channel_ = samples_per_channel; | |
| 869 far_frame.num_channels_ = num_render_channels; | |
| 870 near_frame.sample_rate_hz_ = sample_rate_hz; | |
| 871 near_frame.samples_per_channel_ = samples_per_channel; | |
| 872 | |
| 873 if (event == kInitializeEvent || event == kResetEventDeprecated) { | |
| 874 ASSERT_EQ(1u, | |
| 875 fread(&sample_rate_hz, sizeof(sample_rate_hz), 1, event_file)); | |
| 876 samples_per_channel = sample_rate_hz / 100; | |
| 877 | |
| 878 int32_t unused_device_sample_rate_hz; | |
| 879 ASSERT_EQ(1u, | |
| 880 fread(&unused_device_sample_rate_hz, | |
| 881 sizeof(unused_device_sample_rate_hz), | |
| 882 1, | |
| 883 event_file)); | |
| 884 | |
| 885 ASSERT_EQ(kNoErr, apm->Initialize( | |
| 886 sample_rate_hz, | |
| 887 sample_rate_hz, | |
| 888 sample_rate_hz, | |
| 889 LayoutFromChannels(num_capture_input_channels), | |
| 890 LayoutFromChannels(num_capture_output_channels), | |
| 891 LayoutFromChannels(num_render_channels))); | |
| 892 | |
| 893 far_frame.sample_rate_hz_ = sample_rate_hz; | |
| 894 far_frame.samples_per_channel_ = samples_per_channel; | |
| 895 far_frame.num_channels_ = num_render_channels; | |
| 896 near_frame.sample_rate_hz_ = sample_rate_hz; | |
| 897 near_frame.samples_per_channel_ = samples_per_channel; | |
| 898 | |
| 899 if (!raw_output) { | |
| 900 // The WAV file needs to be reset every time, because it can't change | |
| 901 // it's sample rate or number of channels. | |
| 902 output_wav_file.reset(new WavWriter(out_filename + ".wav", | |
| 903 sample_rate_hz, | |
| 904 num_capture_output_channels)); | |
| 905 } | |
| 906 | |
| 907 if (verbose) { | |
| 908 printf("Init at frame: %d (primary), %d (reverse)\n", | |
| 909 primary_count, reverse_count); | |
| 910 printf(" Sample rate: %d Hz\n", sample_rate_hz); | |
| 911 } | |
| 912 | |
| 913 } else if (event == kRenderEvent) { | |
| 914 reverse_count++; | |
| 915 | |
| 916 size_t size = samples_per_channel * num_render_channels; | |
| 917 read_count = fread(far_frame.data_, | |
| 918 sizeof(int16_t), | |
| 919 size, | |
| 920 far_file); | |
| 921 | |
| 922 if (simulating) { | |
| 923 if (read_count != size) { | |
| 924 // Read an equal amount from the near file to avoid errors due to | |
| 925 // not reaching end-of-file. | |
| 926 EXPECT_EQ(0, fseek(near_file, read_count * sizeof(int16_t), | |
| 927 SEEK_CUR)); | |
| 928 break; // This is expected. | |
| 929 } | |
| 930 } else { | |
| 931 ASSERT_EQ(size, read_count); | |
| 932 } | |
| 933 | |
| 934 if (perf_testing) { | |
| 935 t0 = rtc::TimeNanos(); | |
| 936 } | |
| 937 | |
| 938 ASSERT_EQ(apm->kNoError, | |
| 939 apm->ProcessReverseStream(&far_frame)); | |
| 940 | |
| 941 if (perf_testing) { | |
| 942 t1 = rtc::TimeNanos(); | |
| 943 int64_t diff_nanos = t1 - t0; | |
| 944 acc_nanos += diff_nanos; | |
| 945 int64_t diff_us = diff_nanos / rtc::kNumNanosecsPerMicrosec; | |
| 946 if (diff_us > max_time_reverse_us) { | |
| 947 max_time_reverse_us = diff_us; | |
| 948 } | |
| 949 if (diff_us < min_time_reverse_us) { | |
| 950 min_time_reverse_us = diff_us; | |
| 951 } | |
| 952 } | |
| 953 | |
| 954 } else if (event == kCaptureEvent) { | |
| 955 primary_count++; | |
| 956 near_frame.num_channels_ = num_capture_input_channels; | |
| 957 | |
| 958 size_t size = samples_per_channel * num_capture_input_channels; | |
| 959 read_count = fread(near_frame.data_, | |
| 960 sizeof(int16_t), | |
| 961 size, | |
| 962 near_file); | |
| 963 | |
| 964 near_read_bytes += read_count * sizeof(int16_t); | |
| 965 if (progress && primary_count % 100 == 0) { | |
| 966 printf("%.0f%% complete\r", | |
| 967 (near_read_bytes * 100.0) / near_size_bytes); | |
| 968 fflush(stdout); | |
| 969 } | |
| 970 if (simulating) { | |
| 971 if (read_count != size) { | |
| 972 break; // This is expected. | |
| 973 } | |
| 974 | |
| 975 delay_ms = 0; | |
| 976 drift_samples = 0; | |
| 977 } else { | |
| 978 ASSERT_EQ(size, read_count); | |
| 979 | |
| 980 // TODO(ajm): sizeof(delay_ms) for current files? | |
| 981 ASSERT_EQ(1u, | |
| 982 fread(&delay_ms, 2, 1, delay_file)); | |
| 983 ASSERT_EQ(1u, | |
| 984 fread(&drift_samples, sizeof(drift_samples), 1, drift_file)); | |
| 985 } | |
| 986 | |
| 987 if (apm->gain_control()->is_enabled() && | |
| 988 apm->gain_control()->mode() == GainControl::kAdaptiveAnalog) { | |
| 989 SimulateMic(capture_level, &near_frame); | |
| 990 } | |
| 991 | |
| 992 if (perf_testing) { | |
| 993 t0 = rtc::TimeNanos(); | |
| 994 } | |
| 995 | |
| 996 const int capture_level_in = capture_level; | |
| 997 ASSERT_EQ(apm->kNoError, | |
| 998 apm->gain_control()->set_stream_analog_level(capture_level)); | |
| 999 delay_ms += extra_delay_ms; | |
| 1000 if (override_delay_ms) { | |
| 1001 delay_ms = override_delay_ms; | |
| 1002 } | |
| 1003 ASSERT_EQ(apm->kNoError, | |
| 1004 apm->set_stream_delay_ms(delay_ms)); | |
| 1005 apm->echo_cancellation()->set_stream_drift_samples(drift_samples); | |
| 1006 | |
| 1007 apm->set_stream_key_pressed(true); | |
| 1008 | |
| 1009 int err = apm->ProcessStream(&near_frame); | |
| 1010 if (err == apm->kBadStreamParameterWarning) { | |
| 1011 printf("Bad parameter warning. %s\n", trace_stream.str().c_str()); | |
| 1012 } | |
| 1013 ASSERT_TRUE(err == apm->kNoError || | |
| 1014 err == apm->kBadStreamParameterWarning); | |
| 1015 ASSERT_TRUE(near_frame.num_channels_ == apm->num_output_channels()); | |
| 1016 | |
| 1017 capture_level = apm->gain_control()->stream_analog_level(); | |
| 1018 | |
| 1019 stream_has_voice = | |
| 1020 static_cast<int8_t>(apm->voice_detection()->stream_has_voice()); | |
| 1021 if (vad_out_file != NULL) { | |
| 1022 ASSERT_EQ(1u, fwrite(&stream_has_voice, | |
| 1023 sizeof(stream_has_voice), | |
| 1024 1, | |
| 1025 vad_out_file)); | |
| 1026 } | |
| 1027 | |
| 1028 if (ns_prob_file != NULL) { | |
| 1029 ns_speech_prob = apm->noise_suppression()->speech_probability(); | |
| 1030 ASSERT_EQ(1u, fwrite(&ns_speech_prob, | |
| 1031 sizeof(ns_speech_prob), | |
| 1032 1, | |
| 1033 ns_prob_file)); | |
| 1034 } | |
| 1035 | |
| 1036 if (apm->gain_control()->mode() != GainControl::kAdaptiveAnalog) { | |
| 1037 ASSERT_EQ(capture_level_in, capture_level); | |
| 1038 } | |
| 1039 | |
| 1040 if (perf_testing) { | |
| 1041 t1 = rtc::TimeNanos(); | |
| 1042 int64_t diff_nanos = t1 - t0; | |
| 1043 acc_nanos += diff_nanos; | |
| 1044 int64_t diff_us = diff_nanos / rtc::kNumNanosecsPerMicrosec; | |
| 1045 if (diff_us > max_time_us) { | |
| 1046 max_time_us = diff_us; | |
| 1047 } | |
| 1048 if (diff_us < min_time_us) { | |
| 1049 min_time_us = diff_us; | |
| 1050 } | |
| 1051 } | |
| 1052 | |
| 1053 if (raw_output && !output_raw_file) { | |
| 1054 output_raw_file.reset(new RawFile(out_filename + ".pcm")); | |
| 1055 } | |
| 1056 if (!raw_output && !output_wav_file) { | |
| 1057 output_wav_file.reset(new WavWriter(out_filename + ".wav", | |
| 1058 sample_rate_hz, | |
| 1059 num_capture_output_channels)); | |
| 1060 } | |
| 1061 WriteIntData(near_frame.data_, | |
| 1062 size, | |
| 1063 output_wav_file.get(), | |
| 1064 output_raw_file.get()); | |
| 1065 } else { | |
| 1066 FAIL() << "Event " << event << " is unrecognized"; | |
| 1067 } | |
| 1068 } | |
| 1069 } | |
| 1070 if (progress) { | |
| 1071 printf("100%% complete\r"); | |
| 1072 } | |
| 1073 | |
| 1074 if (aecm_echo_path_out_file != NULL) { | |
| 1075 const size_t path_size = | |
| 1076 apm->echo_control_mobile()->echo_path_size_bytes(); | |
| 1077 std::unique_ptr<char[]> echo_path(new char[path_size]); | |
| 1078 apm->echo_control_mobile()->GetEchoPath(echo_path.get(), path_size); | |
| 1079 ASSERT_EQ(path_size, fwrite(echo_path.get(), | |
| 1080 sizeof(char), | |
| 1081 path_size, | |
| 1082 aecm_echo_path_out_file)); | |
| 1083 fclose(aecm_echo_path_out_file); | |
| 1084 aecm_echo_path_out_file = NULL; | |
| 1085 } | |
| 1086 | |
| 1087 if (verbose) { | |
| 1088 printf("\nProcessed frames: %d (primary), %d (reverse)\n", | |
| 1089 primary_count, reverse_count); | |
| 1090 | |
| 1091 if (apm->level_estimator()->is_enabled()) { | |
| 1092 printf("\n--Level metrics--\n"); | |
| 1093 printf("RMS: %d dBFS\n", -apm->level_estimator()->RMS()); | |
| 1094 } | |
| 1095 if (apm->echo_cancellation()->are_metrics_enabled()) { | |
| 1096 EchoCancellation::Metrics metrics; | |
| 1097 apm->echo_cancellation()->GetMetrics(&metrics); | |
| 1098 printf("\n--Echo metrics--\n"); | |
| 1099 printf("(avg, max, min)\n"); | |
| 1100 printf("ERL: "); | |
| 1101 PrintStat(metrics.echo_return_loss); | |
| 1102 printf("ERLE: "); | |
| 1103 PrintStat(metrics.echo_return_loss_enhancement); | |
| 1104 printf("ANLP: "); | |
| 1105 PrintStat(metrics.a_nlp); | |
| 1106 } | |
| 1107 if (apm->echo_cancellation()->is_delay_logging_enabled()) { | |
| 1108 int median = 0; | |
| 1109 int std = 0; | |
| 1110 float fraction_poor_delays = 0; | |
| 1111 apm->echo_cancellation()->GetDelayMetrics(&median, &std, | |
| 1112 &fraction_poor_delays); | |
| 1113 printf("\n--Delay metrics--\n"); | |
| 1114 printf("Median: %3d\n", median); | |
| 1115 printf("Standard deviation: %3d\n", std); | |
| 1116 printf("Poor delay values: %3.1f%%\n", fraction_poor_delays * 100); | |
| 1117 } | |
| 1118 } | |
| 1119 | |
| 1120 if (!pb_file) { | |
| 1121 int8_t temp_int8; | |
| 1122 if (far_file) { | |
| 1123 read_count = fread(&temp_int8, sizeof(temp_int8), 1, far_file); | |
| 1124 EXPECT_NE(0, feof(far_file)) << "Far-end file not fully processed"; | |
| 1125 } | |
| 1126 | |
| 1127 read_count = fread(&temp_int8, sizeof(temp_int8), 1, near_file); | |
| 1128 EXPECT_NE(0, feof(near_file)) << "Near-end file not fully processed"; | |
| 1129 | |
| 1130 if (!simulating) { | |
| 1131 read_count = fread(&temp_int8, sizeof(temp_int8), 1, event_file); | |
| 1132 EXPECT_NE(0, feof(event_file)) << "Event file not fully processed"; | |
| 1133 read_count = fread(&temp_int8, sizeof(temp_int8), 1, delay_file); | |
| 1134 EXPECT_NE(0, feof(delay_file)) << "Delay file not fully processed"; | |
| 1135 read_count = fread(&temp_int8, sizeof(temp_int8), 1, drift_file); | |
| 1136 EXPECT_NE(0, feof(drift_file)) << "Drift file not fully processed"; | |
| 1137 } | |
| 1138 } | |
| 1139 | |
| 1140 if (perf_testing) { | |
| 1141 if (primary_count > 0) { | |
| 1142 int64_t exec_time = acc_nanos / rtc::kNumNanosecsPerMillisec; | |
| 1143 printf("\nTotal time: %.3f s, file time: %.2f s\n", | |
| 1144 exec_time * 0.001, primary_count * 0.01); | |
| 1145 printf("Time per frame: %.3f ms (average), %.3f ms (max)," | |
| 1146 " %.3f ms (min)\n", | |
| 1147 (exec_time * 1.0) / primary_count, | |
| 1148 (max_time_us + max_time_reverse_us) / 1000.0, | |
| 1149 (min_time_us + min_time_reverse_us) / 1000.0); | |
| 1150 // Record the results with Perf test tools. | |
| 1151 webrtc::test::PrintResult("audioproc", "", "time_per_10ms_frame", | |
| 1152 (exec_time * 1000) / primary_count, "us", false); | |
| 1153 } else { | |
| 1154 printf("Warning: no capture frames\n"); | |
| 1155 } | |
| 1156 } | |
| 1157 } | |
| 1158 | |
| 1159 } // namespace | |
| 1160 } // namespace webrtc | |
| 1161 | |
| 1162 int main(int argc, char* argv[]) { | |
| 1163 webrtc::void_main(argc, argv); | |
| 1164 | |
| 1165 // Optional, but removes memory leak noise from Valgrind. | |
| 1166 google::protobuf::ShutdownProtobufLibrary(); | |
| 1167 return 0; | |
| 1168 } | |
| OLD | NEW |
|---|
Chromium Code Reviews
Issue 2425583002:
Removed the deprecated audioproc executable (Closed)
