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 <algorithm>
14 #include <limits>
15 #include <queue>
16
17 #include "webrtc/base/arraysize.h"
18 #include "webrtc/base/scoped_ptr.h"
19 #include "webrtc/common_audio/include/audio_util.h"
20 #include "webrtc/common_audio/resampler/include/push_resampler.h"
21 #include "webrtc/common_audio/resampler/push_sinc_resampler.h"
22 #include "webrtc/common_audio/signal_processing/include/signal_processing_library.h"
23 #include "webrtc/modules/audio_processing/common.h"
24 #include "webrtc/modules/audio_processing/include/audio_processing.h"
25 #include "webrtc/modules/audio_processing/test/protobuf_utils.h"
26 #include "webrtc/modules/audio_processing/test/test_utils.h"
27 #include "webrtc/modules/include/module_common_types.h"
28 #include "webrtc/system_wrappers/include/event_wrapper.h"
29 #include "webrtc/system_wrappers/include/trace.h"
30 #include "webrtc/test/testsupport/fileutils.h"
31 #ifdef WEBRTC_ANDROID_PLATFORM_BUILD
32 #include "gtest/gtest.h"
33 #include "external/webrtc/webrtc/modules/audio_processing/test/unittest.pb.h"
34 #else
35 #include "testing/gtest/include/gtest/gtest.h"
36 #include "webrtc/audio_processing/unittest.pb.h"
37 #endif
38
39 #if !defined(WEBRTC_ANDROID) && !defined(WEBRTC_IOS)
40 #include "webrtc/modules/audio_processing/beamformer/mock_nonlinear_beamformer.h"
41 #endif
42
43 namespace webrtc {
44 namespace {
45
46 // TODO(ekmeyerson): Switch to using StreamConfig and ProcessingConfig where
47 // applicable.
48
49 // TODO(bjornv): This is not feasible until the functionality has been
50 // re-implemented; see comment at the bottom of this file. For now, the user has
51 // to hard code the |write_ref_data| value.
52 // When false, this will compare the output data with the results stored to
53 // file. This is the typical case. When the file should be updated, it can
54 // be set to true with the command-line switch --write_ref_data.
55 bool write_ref_data = false;
56 const google::protobuf::int32 kChannels[] = {1, 2};
57 const int kSampleRates[] = {8000, 16000, 32000, 48000};
58
59 #if defined(WEBRTC_AUDIOPROC_FIXED_PROFILE)
60 // AECM doesn't support super-wb.
61 const int kProcessSampleRates[] = {8000, 16000};
62 #elif defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
63 const int kProcessSampleRates[] = {8000, 16000, 32000, 48000};
64 #endif
65
66 enum StreamDirection { kForward = 0, kReverse };
67
ConvertToFloat(const int16_t * int_data,ChannelBuffer<float> * cb)68 void ConvertToFloat(const int16_t* int_data, ChannelBuffer<float>* cb) {
69 ChannelBuffer<int16_t> cb_int(cb->num_frames(),
70 cb->num_channels());
71 Deinterleave(int_data,
72 cb->num_frames(),
73 cb->num_channels(),
74 cb_int.channels());
75 for (size_t i = 0; i < cb->num_channels(); ++i) {
76 S16ToFloat(cb_int.channels()[i],
77 cb->num_frames(),
78 cb->channels()[i]);
79 }
80 }
81
ConvertToFloat(const AudioFrame & frame,ChannelBuffer<float> * cb)82 void ConvertToFloat(const AudioFrame& frame, ChannelBuffer<float>* cb) {
83 ConvertToFloat(frame.data_, cb);
84 }
85
86 // Number of channels including the keyboard channel.
TotalChannelsFromLayout(AudioProcessing::ChannelLayout layout)87 size_t TotalChannelsFromLayout(AudioProcessing::ChannelLayout layout) {
88 switch (layout) {
89 case AudioProcessing::kMono:
90 return 1;
91 case AudioProcessing::kMonoAndKeyboard:
92 case AudioProcessing::kStereo:
93 return 2;
94 case AudioProcessing::kStereoAndKeyboard:
95 return 3;
96 }
97 assert(false);
98 return 0;
99 }
100
TruncateToMultipleOf10(int value)101 int TruncateToMultipleOf10(int value) {
102 return (value / 10) * 10;
103 }
104
MixStereoToMono(const float * stereo,float * mono,size_t samples_per_channel)105 void MixStereoToMono(const float* stereo, float* mono,
106 size_t samples_per_channel) {
107 for (size_t i = 0; i < samples_per_channel; ++i)
108 mono[i] = (stereo[i * 2] + stereo[i * 2 + 1]) / 2;
109 }
110
MixStereoToMono(const int16_t * stereo,int16_t * mono,size_t samples_per_channel)111 void MixStereoToMono(const int16_t* stereo, int16_t* mono,
112 size_t samples_per_channel) {
113 for (size_t i = 0; i < samples_per_channel; ++i)
114 mono[i] = (stereo[i * 2] + stereo[i * 2 + 1]) >> 1;
115 }
116
CopyLeftToRightChannel(int16_t * stereo,size_t samples_per_channel)117 void CopyLeftToRightChannel(int16_t* stereo, size_t samples_per_channel) {
118 for (size_t i = 0; i < samples_per_channel; i++) {
119 stereo[i * 2 + 1] = stereo[i * 2];
120 }
121 }
122
VerifyChannelsAreEqual(int16_t * stereo,size_t samples_per_channel)123 void VerifyChannelsAreEqual(int16_t* stereo, size_t samples_per_channel) {
124 for (size_t i = 0; i < samples_per_channel; i++) {
125 EXPECT_EQ(stereo[i * 2 + 1], stereo[i * 2]);
126 }
127 }
128
SetFrameTo(AudioFrame * frame,int16_t value)129 void SetFrameTo(AudioFrame* frame, int16_t value) {
130 for (size_t i = 0; i < frame->samples_per_channel_ * frame->num_channels_;
131 ++i) {
132 frame->data_[i] = value;
133 }
134 }
135
SetFrameTo(AudioFrame * frame,int16_t left,int16_t right)136 void SetFrameTo(AudioFrame* frame, int16_t left, int16_t right) {
137 ASSERT_EQ(2u, frame->num_channels_);
138 for (size_t i = 0; i < frame->samples_per_channel_ * 2; i += 2) {
139 frame->data_[i] = left;
140 frame->data_[i + 1] = right;
141 }
142 }
143
ScaleFrame(AudioFrame * frame,float scale)144 void ScaleFrame(AudioFrame* frame, float scale) {
145 for (size_t i = 0; i < frame->samples_per_channel_ * frame->num_channels_;
146 ++i) {
147 frame->data_[i] = FloatS16ToS16(frame->data_[i] * scale);
148 }
149 }
150
FrameDataAreEqual(const AudioFrame & frame1,const AudioFrame & frame2)151 bool FrameDataAreEqual(const AudioFrame& frame1, const AudioFrame& frame2) {
152 if (frame1.samples_per_channel_ != frame2.samples_per_channel_) {
153 return false;
154 }
155 if (frame1.num_channels_ != frame2.num_channels_) {
156 return false;
157 }
158 if (memcmp(frame1.data_, frame2.data_,
159 frame1.samples_per_channel_ * frame1.num_channels_ *
160 sizeof(int16_t))) {
161 return false;
162 }
163 return true;
164 }
165
EnableAllAPComponents(AudioProcessing * ap)166 void EnableAllAPComponents(AudioProcessing* ap) {
167 #if defined(WEBRTC_AUDIOPROC_FIXED_PROFILE)
168 EXPECT_NOERR(ap->echo_control_mobile()->Enable(true));
169
170 EXPECT_NOERR(ap->gain_control()->set_mode(GainControl::kAdaptiveDigital));
171 EXPECT_NOERR(ap->gain_control()->Enable(true));
172 #elif defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
173 EXPECT_NOERR(ap->echo_cancellation()->enable_drift_compensation(true));
174 EXPECT_NOERR(ap->echo_cancellation()->enable_metrics(true));
175 EXPECT_NOERR(ap->echo_cancellation()->enable_delay_logging(true));
176 EXPECT_NOERR(ap->echo_cancellation()->Enable(true));
177
178 EXPECT_NOERR(ap->gain_control()->set_mode(GainControl::kAdaptiveAnalog));
179 EXPECT_NOERR(ap->gain_control()->set_analog_level_limits(0, 255));
180 EXPECT_NOERR(ap->gain_control()->Enable(true));
181 #endif
182
183 EXPECT_NOERR(ap->high_pass_filter()->Enable(true));
184 EXPECT_NOERR(ap->level_estimator()->Enable(true));
185 EXPECT_NOERR(ap->noise_suppression()->Enable(true));
186
187 EXPECT_NOERR(ap->voice_detection()->Enable(true));
188 }
189
190 // These functions are only used by ApmTest.Process.
191 template <class T>
AbsValue(T a)192 T AbsValue(T a) {
193 return a > 0 ? a: -a;
194 }
195
MaxAudioFrame(const AudioFrame & frame)196 int16_t MaxAudioFrame(const AudioFrame& frame) {
197 const size_t length = frame.samples_per_channel_ * frame.num_channels_;
198 int16_t max_data = AbsValue(frame.data_[0]);
199 for (size_t i = 1; i < length; i++) {
200 max_data = std::max(max_data, AbsValue(frame.data_[i]));
201 }
202
203 return max_data;
204 }
205
206 #if defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
TestStats(const AudioProcessing::Statistic & test,const audioproc::Test::Statistic & reference)207 void TestStats(const AudioProcessing::Statistic& test,
208 const audioproc::Test::Statistic& reference) {
209 EXPECT_EQ(reference.instant(), test.instant);
210 EXPECT_EQ(reference.average(), test.average);
211 EXPECT_EQ(reference.maximum(), test.maximum);
212 EXPECT_EQ(reference.minimum(), test.minimum);
213 }
214
WriteStatsMessage(const AudioProcessing::Statistic & output,audioproc::Test::Statistic * msg)215 void WriteStatsMessage(const AudioProcessing::Statistic& output,
216 audioproc::Test::Statistic* msg) {
217 msg->set_instant(output.instant);
218 msg->set_average(output.average);
219 msg->set_maximum(output.maximum);
220 msg->set_minimum(output.minimum);
221 }
222 #endif
223
OpenFileAndWriteMessage(const std::string filename,const::google::protobuf::MessageLite & msg)224 void OpenFileAndWriteMessage(const std::string filename,
225 const ::google::protobuf::MessageLite& msg) {
226 #if defined(WEBRTC_LINUX) && !defined(WEBRTC_ANDROID)
227 FILE* file = fopen(filename.c_str(), "wb");
228 ASSERT_TRUE(file != NULL);
229
230 int32_t size = msg.ByteSize();
231 ASSERT_GT(size, 0);
232 rtc::scoped_ptr<uint8_t[]> array(new uint8_t[size]);
233 ASSERT_TRUE(msg.SerializeToArray(array.get(), size));
234
235 ASSERT_EQ(1u, fwrite(&size, sizeof(size), 1, file));
236 ASSERT_EQ(static_cast<size_t>(size),
237 fwrite(array.get(), sizeof(array[0]), size, file));
238 fclose(file);
239 #else
240 std::cout << "Warning: Writing new reference is only allowed on Linux!"
241 << std::endl;
242 #endif
243 }
244
ResourceFilePath(std::string name,int sample_rate_hz)245 std::string ResourceFilePath(std::string name, int sample_rate_hz) {
246 std::ostringstream ss;
247 // Resource files are all stereo.
248 ss << name << sample_rate_hz / 1000 << "_stereo";
249 return test::ResourcePath(ss.str(), "pcm");
250 }
251
252 // Temporary filenames unique to this process. Used to be able to run these
253 // tests in parallel as each process needs to be running in isolation they can't
254 // have competing filenames.
255 std::map<std::string, std::string> temp_filenames;
256
OutputFilePath(std::string name,int input_rate,int output_rate,int reverse_input_rate,int reverse_output_rate,size_t num_input_channels,size_t num_output_channels,size_t num_reverse_input_channels,size_t num_reverse_output_channels,StreamDirection file_direction)257 std::string OutputFilePath(std::string name,
258 int input_rate,
259 int output_rate,
260 int reverse_input_rate,
261 int reverse_output_rate,
262 size_t num_input_channels,
263 size_t num_output_channels,
264 size_t num_reverse_input_channels,
265 size_t num_reverse_output_channels,
266 StreamDirection file_direction) {
267 std::ostringstream ss;
268 ss << name << "_i" << num_input_channels << "_" << input_rate / 1000 << "_ir"
269 << num_reverse_input_channels << "_" << reverse_input_rate / 1000 << "_";
270 if (num_output_channels == 1) {
271 ss << "mono";
272 } else if (num_output_channels == 2) {
273 ss << "stereo";
274 } else {
275 assert(false);
276 }
277 ss << output_rate / 1000;
278 if (num_reverse_output_channels == 1) {
279 ss << "_rmono";
280 } else if (num_reverse_output_channels == 2) {
281 ss << "_rstereo";
282 } else {
283 assert(false);
284 }
285 ss << reverse_output_rate / 1000;
286 ss << "_d" << file_direction << "_pcm";
287
288 std::string filename = ss.str();
289 if (temp_filenames[filename].empty())
290 temp_filenames[filename] = test::TempFilename(test::OutputPath(), filename);
291 return temp_filenames[filename];
292 }
293
ClearTempFiles()294 void ClearTempFiles() {
295 for (auto& kv : temp_filenames)
296 remove(kv.second.c_str());
297 }
298
OpenFileAndReadMessage(const std::string filename,::google::protobuf::MessageLite * msg)299 void OpenFileAndReadMessage(const std::string filename,
300 ::google::protobuf::MessageLite* msg) {
301 FILE* file = fopen(filename.c_str(), "rb");
302 ASSERT_TRUE(file != NULL);
303 ReadMessageFromFile(file, msg);
304 fclose(file);
305 }
306
307 // Reads a 10 ms chunk of int16 interleaved audio from the given (assumed
308 // stereo) file, converts to deinterleaved float (optionally downmixing) and
309 // returns the result in |cb|. Returns false if the file ended (or on error) and
310 // true otherwise.
311 //
312 // |int_data| and |float_data| are just temporary space that must be
313 // sufficiently large to hold the 10 ms chunk.
ReadChunk(FILE * file,int16_t * int_data,float * float_data,ChannelBuffer<float> * cb)314 bool ReadChunk(FILE* file, int16_t* int_data, float* float_data,
315 ChannelBuffer<float>* cb) {
316 // The files always contain stereo audio.
317 size_t frame_size = cb->num_frames() * 2;
318 size_t read_count = fread(int_data, sizeof(int16_t), frame_size, file);
319 if (read_count != frame_size) {
320 // Check that the file really ended.
321 assert(feof(file));
322 return false; // This is expected.
323 }
324
325 S16ToFloat(int_data, frame_size, float_data);
326 if (cb->num_channels() == 1) {
327 MixStereoToMono(float_data, cb->channels()[0], cb->num_frames());
328 } else {
329 Deinterleave(float_data, cb->num_frames(), 2,
330 cb->channels());
331 }
332
333 return true;
334 }
335
336 class ApmTest : public ::testing::Test {
337 protected:
338 ApmTest();
339 virtual void SetUp();
340 virtual void TearDown();
341
SetUpTestCase()342 static void SetUpTestCase() {
343 Trace::CreateTrace();
344 }
345
TearDownTestCase()346 static void TearDownTestCase() {
347 Trace::ReturnTrace();
348 ClearTempFiles();
349 }
350
351 // Used to select between int and float interface tests.
352 enum Format {
353 kIntFormat,
354 kFloatFormat
355 };
356
357 void Init(int sample_rate_hz,
358 int output_sample_rate_hz,
359 int reverse_sample_rate_hz,
360 size_t num_input_channels,
361 size_t num_output_channels,
362 size_t num_reverse_channels,
363 bool open_output_file);
364 void Init(AudioProcessing* ap);
365 void EnableAllComponents();
366 bool ReadFrame(FILE* file, AudioFrame* frame);
367 bool ReadFrame(FILE* file, AudioFrame* frame, ChannelBuffer<float>* cb);
368 void ReadFrameWithRewind(FILE* file, AudioFrame* frame);
369 void ReadFrameWithRewind(FILE* file, AudioFrame* frame,
370 ChannelBuffer<float>* cb);
371 void ProcessWithDefaultStreamParameters(AudioFrame* frame);
372 void ProcessDelayVerificationTest(int delay_ms, int system_delay_ms,
373 int delay_min, int delay_max);
374 void TestChangingChannelsInt16Interface(
375 size_t num_channels,
376 AudioProcessing::Error expected_return);
377 void TestChangingForwardChannels(size_t num_in_channels,
378 size_t num_out_channels,
379 AudioProcessing::Error expected_return);
380 void TestChangingReverseChannels(size_t num_rev_channels,
381 AudioProcessing::Error expected_return);
382 void RunQuantizedVolumeDoesNotGetStuckTest(int sample_rate);
383 void RunManualVolumeChangeIsPossibleTest(int sample_rate);
384 void StreamParametersTest(Format format);
385 int ProcessStreamChooser(Format format);
386 int AnalyzeReverseStreamChooser(Format format);
387 void ProcessDebugDump(const std::string& in_filename,
388 const std::string& out_filename,
389 Format format);
390 void VerifyDebugDumpTest(Format format);
391
392 const std::string output_path_;
393 const std::string ref_path_;
394 const std::string ref_filename_;
395 rtc::scoped_ptr<AudioProcessing> apm_;
396 AudioFrame* frame_;
397 AudioFrame* revframe_;
398 rtc::scoped_ptr<ChannelBuffer<float> > float_cb_;
399 rtc::scoped_ptr<ChannelBuffer<float> > revfloat_cb_;
400 int output_sample_rate_hz_;
401 size_t num_output_channels_;
402 FILE* far_file_;
403 FILE* near_file_;
404 FILE* out_file_;
405 };
406
ApmTest()407 ApmTest::ApmTest()
408 : output_path_(test::OutputPath()),
409 ref_path_(test::ProjectRootPath() + "data/audio_processing/"),
410 #if defined(WEBRTC_AUDIOPROC_FIXED_PROFILE)
411 ref_filename_(ref_path_ + "output_data_fixed.pb"),
412 #elif defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
413 #if defined(WEBRTC_MAC)
414 // A different file for Mac is needed because on this platform the AEC
415 // constant |kFixedDelayMs| value is 20 and not 50 as it is on the rest.
416 ref_filename_(ref_path_ + "output_data_mac.pb"),
417 #else
418 ref_filename_(ref_path_ + "output_data_float.pb"),
419 #endif
420 #endif
421 frame_(NULL),
422 revframe_(NULL),
423 output_sample_rate_hz_(0),
424 num_output_channels_(0),
425 far_file_(NULL),
426 near_file_(NULL),
427 out_file_(NULL) {
428 Config config;
429 config.Set<ExperimentalAgc>(new ExperimentalAgc(false));
430 apm_.reset(AudioProcessing::Create(config));
431 }
432
SetUp()433 void ApmTest::SetUp() {
434 ASSERT_TRUE(apm_.get() != NULL);
435
436 frame_ = new AudioFrame();
437 revframe_ = new AudioFrame();
438
439 #if defined(WEBRTC_AUDIOPROC_FIXED_PROFILE)
440 Init(16000, 16000, 16000, 2, 2, 2, false);
441 #else
442 Init(32000, 32000, 32000, 2, 2, 2, false);
443 #endif
444 }
445
TearDown()446 void ApmTest::TearDown() {
447 if (frame_) {
448 delete frame_;
449 }
450 frame_ = NULL;
451
452 if (revframe_) {
453 delete revframe_;
454 }
455 revframe_ = NULL;
456
457 if (far_file_) {
458 ASSERT_EQ(0, fclose(far_file_));
459 }
460 far_file_ = NULL;
461
462 if (near_file_) {
463 ASSERT_EQ(0, fclose(near_file_));
464 }
465 near_file_ = NULL;
466
467 if (out_file_) {
468 ASSERT_EQ(0, fclose(out_file_));
469 }
470 out_file_ = NULL;
471 }
472
Init(AudioProcessing * ap)473 void ApmTest::Init(AudioProcessing* ap) {
474 ASSERT_EQ(kNoErr,
475 ap->Initialize(
476 {{{frame_->sample_rate_hz_, frame_->num_channels_},
477 {output_sample_rate_hz_, num_output_channels_},
478 {revframe_->sample_rate_hz_, revframe_->num_channels_},
479 {revframe_->sample_rate_hz_, revframe_->num_channels_}}}));
480 }
481
Init(int sample_rate_hz,int output_sample_rate_hz,int reverse_sample_rate_hz,size_t num_input_channels,size_t num_output_channels,size_t num_reverse_channels,bool open_output_file)482 void ApmTest::Init(int sample_rate_hz,
483 int output_sample_rate_hz,
484 int reverse_sample_rate_hz,
485 size_t num_input_channels,
486 size_t num_output_channels,
487 size_t num_reverse_channels,
488 bool open_output_file) {
489 SetContainerFormat(sample_rate_hz, num_input_channels, frame_, &float_cb_);
490 output_sample_rate_hz_ = output_sample_rate_hz;
491 num_output_channels_ = num_output_channels;
492
493 SetContainerFormat(reverse_sample_rate_hz, num_reverse_channels, revframe_,
494 &revfloat_cb_);
495 Init(apm_.get());
496
497 if (far_file_) {
498 ASSERT_EQ(0, fclose(far_file_));
499 }
500 std::string filename = ResourceFilePath("far", sample_rate_hz);
501 far_file_ = fopen(filename.c_str(), "rb");
502 ASSERT_TRUE(far_file_ != NULL) << "Could not open file " <<
503 filename << "\n";
504
505 if (near_file_) {
506 ASSERT_EQ(0, fclose(near_file_));
507 }
508 filename = ResourceFilePath("near", sample_rate_hz);
509 near_file_ = fopen(filename.c_str(), "rb");
510 ASSERT_TRUE(near_file_ != NULL) << "Could not open file " <<
511 filename << "\n";
512
513 if (open_output_file) {
514 if (out_file_) {
515 ASSERT_EQ(0, fclose(out_file_));
516 }
517 filename = OutputFilePath(
518 "out", sample_rate_hz, output_sample_rate_hz, reverse_sample_rate_hz,
519 reverse_sample_rate_hz, num_input_channels, num_output_channels,
520 num_reverse_channels, num_reverse_channels, kForward);
521 out_file_ = fopen(filename.c_str(), "wb");
522 ASSERT_TRUE(out_file_ != NULL) << "Could not open file " <<
523 filename << "\n";
524 }
525 }
526
EnableAllComponents()527 void ApmTest::EnableAllComponents() {
528 EnableAllAPComponents(apm_.get());
529 }
530
ReadFrame(FILE * file,AudioFrame * frame,ChannelBuffer<float> * cb)531 bool ApmTest::ReadFrame(FILE* file, AudioFrame* frame,
532 ChannelBuffer<float>* cb) {
533 // The files always contain stereo audio.
534 size_t frame_size = frame->samples_per_channel_ * 2;
535 size_t read_count = fread(frame->data_,
536 sizeof(int16_t),
537 frame_size,
538 file);
539 if (read_count != frame_size) {
540 // Check that the file really ended.
541 EXPECT_NE(0, feof(file));
542 return false; // This is expected.
543 }
544
545 if (frame->num_channels_ == 1) {
546 MixStereoToMono(frame->data_, frame->data_,
547 frame->samples_per_channel_);
548 }
549
550 if (cb) {
551 ConvertToFloat(*frame, cb);
552 }
553 return true;
554 }
555
ReadFrame(FILE * file,AudioFrame * frame)556 bool ApmTest::ReadFrame(FILE* file, AudioFrame* frame) {
557 return ReadFrame(file, frame, NULL);
558 }
559
560 // If the end of the file has been reached, rewind it and attempt to read the
561 // frame again.
ReadFrameWithRewind(FILE * file,AudioFrame * frame,ChannelBuffer<float> * cb)562 void ApmTest::ReadFrameWithRewind(FILE* file, AudioFrame* frame,
563 ChannelBuffer<float>* cb) {
564 if (!ReadFrame(near_file_, frame_, cb)) {
565 rewind(near_file_);
566 ASSERT_TRUE(ReadFrame(near_file_, frame_, cb));
567 }
568 }
569
ReadFrameWithRewind(FILE * file,AudioFrame * frame)570 void ApmTest::ReadFrameWithRewind(FILE* file, AudioFrame* frame) {
571 ReadFrameWithRewind(file, frame, NULL);
572 }
573
ProcessWithDefaultStreamParameters(AudioFrame * frame)574 void ApmTest::ProcessWithDefaultStreamParameters(AudioFrame* frame) {
575 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(0));
576 apm_->echo_cancellation()->set_stream_drift_samples(0);
577 EXPECT_EQ(apm_->kNoError,
578 apm_->gain_control()->set_stream_analog_level(127));
579 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame));
580 }
581
ProcessStreamChooser(Format format)582 int ApmTest::ProcessStreamChooser(Format format) {
583 if (format == kIntFormat) {
584 return apm_->ProcessStream(frame_);
585 }
586 return apm_->ProcessStream(float_cb_->channels(),
587 frame_->samples_per_channel_,
588 frame_->sample_rate_hz_,
589 LayoutFromChannels(frame_->num_channels_),
590 output_sample_rate_hz_,
591 LayoutFromChannels(num_output_channels_),
592 float_cb_->channels());
593 }
594
AnalyzeReverseStreamChooser(Format format)595 int ApmTest::AnalyzeReverseStreamChooser(Format format) {
596 if (format == kIntFormat) {
597 return apm_->AnalyzeReverseStream(revframe_);
598 }
599 return apm_->AnalyzeReverseStream(
600 revfloat_cb_->channels(),
601 revframe_->samples_per_channel_,
602 revframe_->sample_rate_hz_,
603 LayoutFromChannels(revframe_->num_channels_));
604 }
605
ProcessDelayVerificationTest(int delay_ms,int system_delay_ms,int delay_min,int delay_max)606 void ApmTest::ProcessDelayVerificationTest(int delay_ms, int system_delay_ms,
607 int delay_min, int delay_max) {
608 // The |revframe_| and |frame_| should include the proper frame information,
609 // hence can be used for extracting information.
610 AudioFrame tmp_frame;
611 std::queue<AudioFrame*> frame_queue;
612 bool causal = true;
613
614 tmp_frame.CopyFrom(*revframe_);
615 SetFrameTo(&tmp_frame, 0);
616
617 EXPECT_EQ(apm_->kNoError, apm_->Initialize());
618 // Initialize the |frame_queue| with empty frames.
619 int frame_delay = delay_ms / 10;
620 while (frame_delay < 0) {
621 AudioFrame* frame = new AudioFrame();
622 frame->CopyFrom(tmp_frame);
623 frame_queue.push(frame);
624 frame_delay++;
625 causal = false;
626 }
627 while (frame_delay > 0) {
628 AudioFrame* frame = new AudioFrame();
629 frame->CopyFrom(tmp_frame);
630 frame_queue.push(frame);
631 frame_delay--;
632 }
633 // Run for 4.5 seconds, skipping statistics from the first 2.5 seconds. We
634 // need enough frames with audio to have reliable estimates, but as few as
635 // possible to keep processing time down. 4.5 seconds seemed to be a good
636 // compromise for this recording.
637 for (int frame_count = 0; frame_count < 450; ++frame_count) {
638 AudioFrame* frame = new AudioFrame();
639 frame->CopyFrom(tmp_frame);
640 // Use the near end recording, since that has more speech in it.
641 ASSERT_TRUE(ReadFrame(near_file_, frame));
642 frame_queue.push(frame);
643 AudioFrame* reverse_frame = frame;
644 AudioFrame* process_frame = frame_queue.front();
645 if (!causal) {
646 reverse_frame = frame_queue.front();
647 // When we call ProcessStream() the frame is modified, so we can't use the
648 // pointer directly when things are non-causal. Use an intermediate frame
649 // and copy the data.
650 process_frame = &tmp_frame;
651 process_frame->CopyFrom(*frame);
652 }
653 EXPECT_EQ(apm_->kNoError, apm_->AnalyzeReverseStream(reverse_frame));
654 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(system_delay_ms));
655 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(process_frame));
656 frame = frame_queue.front();
657 frame_queue.pop();
658 delete frame;
659
660 if (frame_count == 250) {
661 int median;
662 int std;
663 float poor_fraction;
664 // Discard the first delay metrics to avoid convergence effects.
665 EXPECT_EQ(apm_->kNoError,
666 apm_->echo_cancellation()->GetDelayMetrics(&median, &std,
667 &poor_fraction));
668 }
669 }
670
671 rewind(near_file_);
672 while (!frame_queue.empty()) {
673 AudioFrame* frame = frame_queue.front();
674 frame_queue.pop();
675 delete frame;
676 }
677 // Calculate expected delay estimate and acceptable regions. Further,
678 // limit them w.r.t. AEC delay estimation support.
679 const size_t samples_per_ms =
680 std::min(static_cast<size_t>(16), frame_->samples_per_channel_ / 10);
681 int expected_median = std::min(std::max(delay_ms - system_delay_ms,
682 delay_min), delay_max);
683 int expected_median_high = std::min(
684 std::max(expected_median + static_cast<int>(96 / samples_per_ms),
685 delay_min),
686 delay_max);
687 int expected_median_low = std::min(
688 std::max(expected_median - static_cast<int>(96 / samples_per_ms),
689 delay_min),
690 delay_max);
691 // Verify delay metrics.
692 int median;
693 int std;
694 float poor_fraction;
695 EXPECT_EQ(apm_->kNoError,
696 apm_->echo_cancellation()->GetDelayMetrics(&median, &std,
697 &poor_fraction));
698 EXPECT_GE(expected_median_high, median);
699 EXPECT_LE(expected_median_low, median);
700 }
701
StreamParametersTest(Format format)702 void ApmTest::StreamParametersTest(Format format) {
703 // No errors when the components are disabled.
704 EXPECT_EQ(apm_->kNoError, ProcessStreamChooser(format));
705
706 // -- Missing AGC level --
707 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->Enable(true));
708 EXPECT_EQ(apm_->kStreamParameterNotSetError,
709 ProcessStreamChooser(format));
710
711 // Resets after successful ProcessStream().
712 EXPECT_EQ(apm_->kNoError,
713 apm_->gain_control()->set_stream_analog_level(127));
714 EXPECT_EQ(apm_->kNoError, ProcessStreamChooser(format));
715 EXPECT_EQ(apm_->kStreamParameterNotSetError,
716 ProcessStreamChooser(format));
717
718 // Other stream parameters set correctly.
719 EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(true));
720 EXPECT_EQ(apm_->kNoError,
721 apm_->echo_cancellation()->enable_drift_compensation(true));
722 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(100));
723 apm_->echo_cancellation()->set_stream_drift_samples(0);
724 EXPECT_EQ(apm_->kStreamParameterNotSetError,
725 ProcessStreamChooser(format));
726 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->Enable(false));
727 EXPECT_EQ(apm_->kNoError,
728 apm_->echo_cancellation()->enable_drift_compensation(false));
729
730 // -- Missing delay --
731 EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(true));
732 EXPECT_EQ(apm_->kNoError, ProcessStreamChooser(format));
733 EXPECT_EQ(apm_->kStreamParameterNotSetError,
734 ProcessStreamChooser(format));
735
736 // Resets after successful ProcessStream().
737 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(100));
738 EXPECT_EQ(apm_->kNoError, ProcessStreamChooser(format));
739 EXPECT_EQ(apm_->kStreamParameterNotSetError,
740 ProcessStreamChooser(format));
741
742 // Other stream parameters set correctly.
743 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->Enable(true));
744 EXPECT_EQ(apm_->kNoError,
745 apm_->echo_cancellation()->enable_drift_compensation(true));
746 apm_->echo_cancellation()->set_stream_drift_samples(0);
747 EXPECT_EQ(apm_->kNoError,
748 apm_->gain_control()->set_stream_analog_level(127));
749 EXPECT_EQ(apm_->kStreamParameterNotSetError,
750 ProcessStreamChooser(format));
751 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->Enable(false));
752
753 // -- Missing drift --
754 EXPECT_EQ(apm_->kStreamParameterNotSetError,
755 ProcessStreamChooser(format));
756
757 // Resets after successful ProcessStream().
758 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(100));
759 apm_->echo_cancellation()->set_stream_drift_samples(0);
760 EXPECT_EQ(apm_->kNoError, ProcessStreamChooser(format));
761 EXPECT_EQ(apm_->kStreamParameterNotSetError,
762 ProcessStreamChooser(format));
763
764 // Other stream parameters set correctly.
765 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->Enable(true));
766 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(100));
767 EXPECT_EQ(apm_->kNoError,
768 apm_->gain_control()->set_stream_analog_level(127));
769 EXPECT_EQ(apm_->kStreamParameterNotSetError,
770 ProcessStreamChooser(format));
771
772 // -- No stream parameters --
773 EXPECT_EQ(apm_->kNoError,
774 AnalyzeReverseStreamChooser(format));
775 EXPECT_EQ(apm_->kStreamParameterNotSetError,
776 ProcessStreamChooser(format));
777
778 // -- All there --
779 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(100));
780 apm_->echo_cancellation()->set_stream_drift_samples(0);
781 EXPECT_EQ(apm_->kNoError,
782 apm_->gain_control()->set_stream_analog_level(127));
783 EXPECT_EQ(apm_->kNoError, ProcessStreamChooser(format));
784 }
785
TEST_F(ApmTest,StreamParametersInt)786 TEST_F(ApmTest, StreamParametersInt) {
787 StreamParametersTest(kIntFormat);
788 }
789
TEST_F(ApmTest,StreamParametersFloat)790 TEST_F(ApmTest, StreamParametersFloat) {
791 StreamParametersTest(kFloatFormat);
792 }
793
TEST_F(ApmTest,DefaultDelayOffsetIsZero)794 TEST_F(ApmTest, DefaultDelayOffsetIsZero) {
795 EXPECT_EQ(0, apm_->delay_offset_ms());
796 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(50));
797 EXPECT_EQ(50, apm_->stream_delay_ms());
798 }
799
TEST_F(ApmTest,DelayOffsetWithLimitsIsSetProperly)800 TEST_F(ApmTest, DelayOffsetWithLimitsIsSetProperly) {
801 // High limit of 500 ms.
802 apm_->set_delay_offset_ms(100);
803 EXPECT_EQ(100, apm_->delay_offset_ms());
804 EXPECT_EQ(apm_->kBadStreamParameterWarning, apm_->set_stream_delay_ms(450));
805 EXPECT_EQ(500, apm_->stream_delay_ms());
806 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(100));
807 EXPECT_EQ(200, apm_->stream_delay_ms());
808
809 // Low limit of 0 ms.
810 apm_->set_delay_offset_ms(-50);
811 EXPECT_EQ(-50, apm_->delay_offset_ms());
812 EXPECT_EQ(apm_->kBadStreamParameterWarning, apm_->set_stream_delay_ms(20));
813 EXPECT_EQ(0, apm_->stream_delay_ms());
814 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(100));
815 EXPECT_EQ(50, apm_->stream_delay_ms());
816 }
817
TestChangingChannelsInt16Interface(size_t num_channels,AudioProcessing::Error expected_return)818 void ApmTest::TestChangingChannelsInt16Interface(
819 size_t num_channels,
820 AudioProcessing::Error expected_return) {
821 frame_->num_channels_ = num_channels;
822 EXPECT_EQ(expected_return, apm_->ProcessStream(frame_));
823 EXPECT_EQ(expected_return, apm_->AnalyzeReverseStream(frame_));
824 }
825
TestChangingForwardChannels(size_t num_in_channels,size_t num_out_channels,AudioProcessing::Error expected_return)826 void ApmTest::TestChangingForwardChannels(
827 size_t num_in_channels,
828 size_t num_out_channels,
829 AudioProcessing::Error expected_return) {
830 const StreamConfig input_stream = {frame_->sample_rate_hz_, num_in_channels};
831 const StreamConfig output_stream = {output_sample_rate_hz_, num_out_channels};
832
833 EXPECT_EQ(expected_return,
834 apm_->ProcessStream(float_cb_->channels(), input_stream,
835 output_stream, float_cb_->channels()));
836 }
837
TestChangingReverseChannels(size_t num_rev_channels,AudioProcessing::Error expected_return)838 void ApmTest::TestChangingReverseChannels(
839 size_t num_rev_channels,
840 AudioProcessing::Error expected_return) {
841 const ProcessingConfig processing_config = {
842 {{frame_->sample_rate_hz_, apm_->num_input_channels()},
843 {output_sample_rate_hz_, apm_->num_output_channels()},
844 {frame_->sample_rate_hz_, num_rev_channels},
845 {frame_->sample_rate_hz_, num_rev_channels}}};
846
847 EXPECT_EQ(
848 expected_return,
849 apm_->ProcessReverseStream(
850 float_cb_->channels(), processing_config.reverse_input_stream(),
851 processing_config.reverse_output_stream(), float_cb_->channels()));
852 }
853
TEST_F(ApmTest,ChannelsInt16Interface)854 TEST_F(ApmTest, ChannelsInt16Interface) {
855 // Testing number of invalid and valid channels.
856 Init(16000, 16000, 16000, 4, 4, 4, false);
857
858 TestChangingChannelsInt16Interface(0, apm_->kBadNumberChannelsError);
859
860 for (size_t i = 1; i < 4; i++) {
861 TestChangingChannelsInt16Interface(i, kNoErr);
862 EXPECT_EQ(i, apm_->num_input_channels());
863 // We always force the number of reverse channels used for processing to 1.
864 EXPECT_EQ(1u, apm_->num_reverse_channels());
865 }
866 }
867
TEST_F(ApmTest,Channels)868 TEST_F(ApmTest, Channels) {
869 // Testing number of invalid and valid channels.
870 Init(16000, 16000, 16000, 4, 4, 4, false);
871
872 TestChangingForwardChannels(0, 1, apm_->kBadNumberChannelsError);
873 TestChangingReverseChannels(0, apm_->kBadNumberChannelsError);
874
875 for (size_t i = 1; i < 4; ++i) {
876 for (size_t j = 0; j < 1; ++j) {
877 // Output channels much be one or match input channels.
878 if (j == 1 || i == j) {
879 TestChangingForwardChannels(i, j, kNoErr);
880 TestChangingReverseChannels(i, kNoErr);
881
882 EXPECT_EQ(i, apm_->num_input_channels());
883 EXPECT_EQ(j, apm_->num_output_channels());
884 // The number of reverse channels used for processing to is always 1.
885 EXPECT_EQ(1u, apm_->num_reverse_channels());
886 } else {
887 TestChangingForwardChannels(i, j,
888 AudioProcessing::kBadNumberChannelsError);
889 }
890 }
891 }
892 }
893
TEST_F(ApmTest,SampleRatesInt)894 TEST_F(ApmTest, SampleRatesInt) {
895 // Testing invalid sample rates
896 SetContainerFormat(10000, 2, frame_, &float_cb_);
897 EXPECT_EQ(apm_->kBadSampleRateError, ProcessStreamChooser(kIntFormat));
898 // Testing valid sample rates
899 int fs[] = {8000, 16000, 32000, 48000};
900 for (size_t i = 0; i < arraysize(fs); i++) {
901 SetContainerFormat(fs[i], 2, frame_, &float_cb_);
902 EXPECT_NOERR(ProcessStreamChooser(kIntFormat));
903 }
904 }
905
TEST_F(ApmTest,EchoCancellation)906 TEST_F(ApmTest, EchoCancellation) {
907 EXPECT_EQ(apm_->kNoError,
908 apm_->echo_cancellation()->enable_drift_compensation(true));
909 EXPECT_TRUE(apm_->echo_cancellation()->is_drift_compensation_enabled());
910 EXPECT_EQ(apm_->kNoError,
911 apm_->echo_cancellation()->enable_drift_compensation(false));
912 EXPECT_FALSE(apm_->echo_cancellation()->is_drift_compensation_enabled());
913
914 EchoCancellation::SuppressionLevel level[] = {
915 EchoCancellation::kLowSuppression,
916 EchoCancellation::kModerateSuppression,
917 EchoCancellation::kHighSuppression,
918 };
919 for (size_t i = 0; i < arraysize(level); i++) {
920 EXPECT_EQ(apm_->kNoError,
921 apm_->echo_cancellation()->set_suppression_level(level[i]));
922 EXPECT_EQ(level[i],
923 apm_->echo_cancellation()->suppression_level());
924 }
925
926 EchoCancellation::Metrics metrics;
927 EXPECT_EQ(apm_->kNotEnabledError,
928 apm_->echo_cancellation()->GetMetrics(&metrics));
929
930 EXPECT_EQ(apm_->kNoError,
931 apm_->echo_cancellation()->enable_metrics(true));
932 EXPECT_TRUE(apm_->echo_cancellation()->are_metrics_enabled());
933 EXPECT_EQ(apm_->kNoError,
934 apm_->echo_cancellation()->enable_metrics(false));
935 EXPECT_FALSE(apm_->echo_cancellation()->are_metrics_enabled());
936
937 int median = 0;
938 int std = 0;
939 float poor_fraction = 0;
940 EXPECT_EQ(apm_->kNotEnabledError,
941 apm_->echo_cancellation()->GetDelayMetrics(&median, &std,
942 &poor_fraction));
943
944 EXPECT_EQ(apm_->kNoError,
945 apm_->echo_cancellation()->enable_delay_logging(true));
946 EXPECT_TRUE(apm_->echo_cancellation()->is_delay_logging_enabled());
947 EXPECT_EQ(apm_->kNoError,
948 apm_->echo_cancellation()->enable_delay_logging(false));
949 EXPECT_FALSE(apm_->echo_cancellation()->is_delay_logging_enabled());
950
951 EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(true));
952 EXPECT_TRUE(apm_->echo_cancellation()->is_enabled());
953 EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(false));
954 EXPECT_FALSE(apm_->echo_cancellation()->is_enabled());
955
956 EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(true));
957 EXPECT_TRUE(apm_->echo_cancellation()->is_enabled());
958 EXPECT_TRUE(apm_->echo_cancellation()->aec_core() != NULL);
959 EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(false));
960 EXPECT_FALSE(apm_->echo_cancellation()->is_enabled());
961 EXPECT_FALSE(apm_->echo_cancellation()->aec_core() != NULL);
962 }
963
TEST_F(ApmTest,DISABLED_EchoCancellationReportsCorrectDelays)964 TEST_F(ApmTest, DISABLED_EchoCancellationReportsCorrectDelays) {
965 // TODO(bjornv): Fix this test to work with DA-AEC.
966 // Enable AEC only.
967 EXPECT_EQ(apm_->kNoError,
968 apm_->echo_cancellation()->enable_drift_compensation(false));
969 EXPECT_EQ(apm_->kNoError,
970 apm_->echo_cancellation()->enable_metrics(false));
971 EXPECT_EQ(apm_->kNoError,
972 apm_->echo_cancellation()->enable_delay_logging(true));
973 EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(true));
974 Config config;
975 config.Set<DelayAgnostic>(new DelayAgnostic(false));
976 apm_->SetExtraOptions(config);
977
978 // Internally in the AEC the amount of lookahead the delay estimation can
979 // handle is 15 blocks and the maximum delay is set to 60 blocks.
980 const int kLookaheadBlocks = 15;
981 const int kMaxDelayBlocks = 60;
982 // The AEC has a startup time before it actually starts to process. This
983 // procedure can flush the internal far-end buffer, which of course affects
984 // the delay estimation. Therefore, we set a system_delay high enough to
985 // avoid that. The smallest system_delay you can report without flushing the
986 // buffer is 66 ms in 8 kHz.
987 //
988 // It is known that for 16 kHz (and 32 kHz) sampling frequency there is an
989 // additional stuffing of 8 ms on the fly, but it seems to have no impact on
990 // delay estimation. This should be noted though. In case of test failure,
991 // this could be the cause.
992 const int kSystemDelayMs = 66;
993 // Test a couple of corner cases and verify that the estimated delay is
994 // within a valid region (set to +-1.5 blocks). Note that these cases are
995 // sampling frequency dependent.
996 for (size_t i = 0; i < arraysize(kProcessSampleRates); i++) {
997 Init(kProcessSampleRates[i],
998 kProcessSampleRates[i],
999 kProcessSampleRates[i],
1000 2,
1001 2,
1002 2,
1003 false);
1004 // Sampling frequency dependent variables.
1005 const int num_ms_per_block =
1006 std::max(4, static_cast<int>(640 / frame_->samples_per_channel_));
1007 const int delay_min_ms = -kLookaheadBlocks * num_ms_per_block;
1008 const int delay_max_ms = (kMaxDelayBlocks - 1) * num_ms_per_block;
1009
1010 // 1) Verify correct delay estimate at lookahead boundary.
1011 int delay_ms = TruncateToMultipleOf10(kSystemDelayMs + delay_min_ms);
1012 ProcessDelayVerificationTest(delay_ms, kSystemDelayMs, delay_min_ms,
1013 delay_max_ms);
1014 // 2) A delay less than maximum lookahead should give an delay estimate at
1015 // the boundary (= -kLookaheadBlocks * num_ms_per_block).
1016 delay_ms -= 20;
1017 ProcessDelayVerificationTest(delay_ms, kSystemDelayMs, delay_min_ms,
1018 delay_max_ms);
1019 // 3) Three values around zero delay. Note that we need to compensate for
1020 // the fake system_delay.
1021 delay_ms = TruncateToMultipleOf10(kSystemDelayMs - 10);
1022 ProcessDelayVerificationTest(delay_ms, kSystemDelayMs, delay_min_ms,
1023 delay_max_ms);
1024 delay_ms = TruncateToMultipleOf10(kSystemDelayMs);
1025 ProcessDelayVerificationTest(delay_ms, kSystemDelayMs, delay_min_ms,
1026 delay_max_ms);
1027 delay_ms = TruncateToMultipleOf10(kSystemDelayMs + 10);
1028 ProcessDelayVerificationTest(delay_ms, kSystemDelayMs, delay_min_ms,
1029 delay_max_ms);
1030 // 4) Verify correct delay estimate at maximum delay boundary.
1031 delay_ms = TruncateToMultipleOf10(kSystemDelayMs + delay_max_ms);
1032 ProcessDelayVerificationTest(delay_ms, kSystemDelayMs, delay_min_ms,
1033 delay_max_ms);
1034 // 5) A delay above the maximum delay should give an estimate at the
1035 // boundary (= (kMaxDelayBlocks - 1) * num_ms_per_block).
1036 delay_ms += 20;
1037 ProcessDelayVerificationTest(delay_ms, kSystemDelayMs, delay_min_ms,
1038 delay_max_ms);
1039 }
1040 }
1041
TEST_F(ApmTest,EchoControlMobile)1042 TEST_F(ApmTest, EchoControlMobile) {
1043 // AECM won't use super-wideband.
1044 SetFrameSampleRate(frame_, 32000);
1045 EXPECT_NOERR(apm_->ProcessStream(frame_));
1046 EXPECT_EQ(apm_->kBadSampleRateError,
1047 apm_->echo_control_mobile()->Enable(true));
1048 SetFrameSampleRate(frame_, 16000);
1049 EXPECT_NOERR(apm_->ProcessStream(frame_));
1050 EXPECT_EQ(apm_->kNoError,
1051 apm_->echo_control_mobile()->Enable(true));
1052 SetFrameSampleRate(frame_, 32000);
1053 EXPECT_EQ(apm_->kUnsupportedComponentError, apm_->ProcessStream(frame_));
1054
1055 // Turn AECM on (and AEC off)
1056 Init(16000, 16000, 16000, 2, 2, 2, false);
1057 EXPECT_EQ(apm_->kNoError, apm_->echo_control_mobile()->Enable(true));
1058 EXPECT_TRUE(apm_->echo_control_mobile()->is_enabled());
1059
1060 // Toggle routing modes
1061 EchoControlMobile::RoutingMode mode[] = {
1062 EchoControlMobile::kQuietEarpieceOrHeadset,
1063 EchoControlMobile::kEarpiece,
1064 EchoControlMobile::kLoudEarpiece,
1065 EchoControlMobile::kSpeakerphone,
1066 EchoControlMobile::kLoudSpeakerphone,
1067 };
1068 for (size_t i = 0; i < arraysize(mode); i++) {
1069 EXPECT_EQ(apm_->kNoError,
1070 apm_->echo_control_mobile()->set_routing_mode(mode[i]));
1071 EXPECT_EQ(mode[i],
1072 apm_->echo_control_mobile()->routing_mode());
1073 }
1074 // Turn comfort noise off/on
1075 EXPECT_EQ(apm_->kNoError,
1076 apm_->echo_control_mobile()->enable_comfort_noise(false));
1077 EXPECT_FALSE(apm_->echo_control_mobile()->is_comfort_noise_enabled());
1078 EXPECT_EQ(apm_->kNoError,
1079 apm_->echo_control_mobile()->enable_comfort_noise(true));
1080 EXPECT_TRUE(apm_->echo_control_mobile()->is_comfort_noise_enabled());
1081 // Set and get echo path
1082 const size_t echo_path_size =
1083 apm_->echo_control_mobile()->echo_path_size_bytes();
1084 rtc::scoped_ptr<char[]> echo_path_in(new char[echo_path_size]);
1085 rtc::scoped_ptr<char[]> echo_path_out(new char[echo_path_size]);
1086 EXPECT_EQ(apm_->kNullPointerError,
1087 apm_->echo_control_mobile()->SetEchoPath(NULL, echo_path_size));
1088 EXPECT_EQ(apm_->kNullPointerError,
1089 apm_->echo_control_mobile()->GetEchoPath(NULL, echo_path_size));
1090 EXPECT_EQ(apm_->kBadParameterError,
1091 apm_->echo_control_mobile()->GetEchoPath(echo_path_out.get(), 1));
1092 EXPECT_EQ(apm_->kNoError,
1093 apm_->echo_control_mobile()->GetEchoPath(echo_path_out.get(),
1094 echo_path_size));
1095 for (size_t i = 0; i < echo_path_size; i++) {
1096 echo_path_in[i] = echo_path_out[i] + 1;
1097 }
1098 EXPECT_EQ(apm_->kBadParameterError,
1099 apm_->echo_control_mobile()->SetEchoPath(echo_path_in.get(), 1));
1100 EXPECT_EQ(apm_->kNoError,
1101 apm_->echo_control_mobile()->SetEchoPath(echo_path_in.get(),
1102 echo_path_size));
1103 EXPECT_EQ(apm_->kNoError,
1104 apm_->echo_control_mobile()->GetEchoPath(echo_path_out.get(),
1105 echo_path_size));
1106 for (size_t i = 0; i < echo_path_size; i++) {
1107 EXPECT_EQ(echo_path_in[i], echo_path_out[i]);
1108 }
1109
1110 // Process a few frames with NS in the default disabled state. This exercises
1111 // a different codepath than with it enabled.
1112 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(0));
1113 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1114 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(0));
1115 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1116
1117 // Turn AECM off
1118 EXPECT_EQ(apm_->kNoError, apm_->echo_control_mobile()->Enable(false));
1119 EXPECT_FALSE(apm_->echo_control_mobile()->is_enabled());
1120 }
1121
TEST_F(ApmTest,GainControl)1122 TEST_F(ApmTest, GainControl) {
1123 // Testing gain modes
1124 EXPECT_EQ(apm_->kNoError,
1125 apm_->gain_control()->set_mode(
1126 apm_->gain_control()->mode()));
1127
1128 GainControl::Mode mode[] = {
1129 GainControl::kAdaptiveAnalog,
1130 GainControl::kAdaptiveDigital,
1131 GainControl::kFixedDigital
1132 };
1133 for (size_t i = 0; i < arraysize(mode); i++) {
1134 EXPECT_EQ(apm_->kNoError,
1135 apm_->gain_control()->set_mode(mode[i]));
1136 EXPECT_EQ(mode[i], apm_->gain_control()->mode());
1137 }
1138 // Testing invalid target levels
1139 EXPECT_EQ(apm_->kBadParameterError,
1140 apm_->gain_control()->set_target_level_dbfs(-3));
1141 EXPECT_EQ(apm_->kBadParameterError,
1142 apm_->gain_control()->set_target_level_dbfs(-40));
1143 // Testing valid target levels
1144 EXPECT_EQ(apm_->kNoError,
1145 apm_->gain_control()->set_target_level_dbfs(
1146 apm_->gain_control()->target_level_dbfs()));
1147
1148 int level_dbfs[] = {0, 6, 31};
1149 for (size_t i = 0; i < arraysize(level_dbfs); i++) {
1150 EXPECT_EQ(apm_->kNoError,
1151 apm_->gain_control()->set_target_level_dbfs(level_dbfs[i]));
1152 EXPECT_EQ(level_dbfs[i], apm_->gain_control()->target_level_dbfs());
1153 }
1154
1155 // Testing invalid compression gains
1156 EXPECT_EQ(apm_->kBadParameterError,
1157 apm_->gain_control()->set_compression_gain_db(-1));
1158 EXPECT_EQ(apm_->kBadParameterError,
1159 apm_->gain_control()->set_compression_gain_db(100));
1160
1161 // Testing valid compression gains
1162 EXPECT_EQ(apm_->kNoError,
1163 apm_->gain_control()->set_compression_gain_db(
1164 apm_->gain_control()->compression_gain_db()));
1165
1166 int gain_db[] = {0, 10, 90};
1167 for (size_t i = 0; i < arraysize(gain_db); i++) {
1168 EXPECT_EQ(apm_->kNoError,
1169 apm_->gain_control()->set_compression_gain_db(gain_db[i]));
1170 EXPECT_EQ(gain_db[i], apm_->gain_control()->compression_gain_db());
1171 }
1172
1173 // Testing limiter off/on
1174 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->enable_limiter(false));
1175 EXPECT_FALSE(apm_->gain_control()->is_limiter_enabled());
1176 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->enable_limiter(true));
1177 EXPECT_TRUE(apm_->gain_control()->is_limiter_enabled());
1178
1179 // Testing invalid level limits
1180 EXPECT_EQ(apm_->kBadParameterError,
1181 apm_->gain_control()->set_analog_level_limits(-1, 512));
1182 EXPECT_EQ(apm_->kBadParameterError,
1183 apm_->gain_control()->set_analog_level_limits(100000, 512));
1184 EXPECT_EQ(apm_->kBadParameterError,
1185 apm_->gain_control()->set_analog_level_limits(512, -1));
1186 EXPECT_EQ(apm_->kBadParameterError,
1187 apm_->gain_control()->set_analog_level_limits(512, 100000));
1188 EXPECT_EQ(apm_->kBadParameterError,
1189 apm_->gain_control()->set_analog_level_limits(512, 255));
1190
1191 // Testing valid level limits
1192 EXPECT_EQ(apm_->kNoError,
1193 apm_->gain_control()->set_analog_level_limits(
1194 apm_->gain_control()->analog_level_minimum(),
1195 apm_->gain_control()->analog_level_maximum()));
1196
1197 int min_level[] = {0, 255, 1024};
1198 for (size_t i = 0; i < arraysize(min_level); i++) {
1199 EXPECT_EQ(apm_->kNoError,
1200 apm_->gain_control()->set_analog_level_limits(min_level[i], 1024));
1201 EXPECT_EQ(min_level[i], apm_->gain_control()->analog_level_minimum());
1202 }
1203
1204 int max_level[] = {0, 1024, 65535};
1205 for (size_t i = 0; i < arraysize(min_level); i++) {
1206 EXPECT_EQ(apm_->kNoError,
1207 apm_->gain_control()->set_analog_level_limits(0, max_level[i]));
1208 EXPECT_EQ(max_level[i], apm_->gain_control()->analog_level_maximum());
1209 }
1210
1211 // TODO(ajm): stream_is_saturated() and stream_analog_level()
1212
1213 // Turn AGC off
1214 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->Enable(false));
1215 EXPECT_FALSE(apm_->gain_control()->is_enabled());
1216 }
1217
RunQuantizedVolumeDoesNotGetStuckTest(int sample_rate)1218 void ApmTest::RunQuantizedVolumeDoesNotGetStuckTest(int sample_rate) {
1219 Init(sample_rate, sample_rate, sample_rate, 2, 2, 2, false);
1220 EXPECT_EQ(apm_->kNoError,
1221 apm_->gain_control()->set_mode(GainControl::kAdaptiveAnalog));
1222 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->Enable(true));
1223
1224 int out_analog_level = 0;
1225 for (int i = 0; i < 2000; ++i) {
1226 ReadFrameWithRewind(near_file_, frame_);
1227 // Ensure the audio is at a low level, so the AGC will try to increase it.
1228 ScaleFrame(frame_, 0.25);
1229
1230 // Always pass in the same volume.
1231 EXPECT_EQ(apm_->kNoError,
1232 apm_->gain_control()->set_stream_analog_level(100));
1233 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1234 out_analog_level = apm_->gain_control()->stream_analog_level();
1235 }
1236
1237 // Ensure the AGC is still able to reach the maximum.
1238 EXPECT_EQ(255, out_analog_level);
1239 }
1240
1241 // Verifies that despite volume slider quantization, the AGC can continue to
1242 // increase its volume.
TEST_F(ApmTest,QuantizedVolumeDoesNotGetStuck)1243 TEST_F(ApmTest, QuantizedVolumeDoesNotGetStuck) {
1244 for (size_t i = 0; i < arraysize(kSampleRates); ++i) {
1245 RunQuantizedVolumeDoesNotGetStuckTest(kSampleRates[i]);
1246 }
1247 }
1248
RunManualVolumeChangeIsPossibleTest(int sample_rate)1249 void ApmTest::RunManualVolumeChangeIsPossibleTest(int sample_rate) {
1250 Init(sample_rate, sample_rate, sample_rate, 2, 2, 2, false);
1251 EXPECT_EQ(apm_->kNoError,
1252 apm_->gain_control()->set_mode(GainControl::kAdaptiveAnalog));
1253 EXPECT_EQ(apm_->kNoError, apm_->gain_control()->Enable(true));
1254
1255 int out_analog_level = 100;
1256 for (int i = 0; i < 1000; ++i) {
1257 ReadFrameWithRewind(near_file_, frame_);
1258 // Ensure the audio is at a low level, so the AGC will try to increase it.
1259 ScaleFrame(frame_, 0.25);
1260
1261 EXPECT_EQ(apm_->kNoError,
1262 apm_->gain_control()->set_stream_analog_level(out_analog_level));
1263 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1264 out_analog_level = apm_->gain_control()->stream_analog_level();
1265 }
1266
1267 // Ensure the volume was raised.
1268 EXPECT_GT(out_analog_level, 100);
1269 int highest_level_reached = out_analog_level;
1270 // Simulate a user manual volume change.
1271 out_analog_level = 100;
1272
1273 for (int i = 0; i < 300; ++i) {
1274 ReadFrameWithRewind(near_file_, frame_);
1275 ScaleFrame(frame_, 0.25);
1276
1277 EXPECT_EQ(apm_->kNoError,
1278 apm_->gain_control()->set_stream_analog_level(out_analog_level));
1279 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1280 out_analog_level = apm_->gain_control()->stream_analog_level();
1281 // Check that AGC respected the manually adjusted volume.
1282 EXPECT_LT(out_analog_level, highest_level_reached);
1283 }
1284 // Check that the volume was still raised.
1285 EXPECT_GT(out_analog_level, 100);
1286 }
1287
TEST_F(ApmTest,ManualVolumeChangeIsPossible)1288 TEST_F(ApmTest, ManualVolumeChangeIsPossible) {
1289 for (size_t i = 0; i < arraysize(kSampleRates); ++i) {
1290 RunManualVolumeChangeIsPossibleTest(kSampleRates[i]);
1291 }
1292 }
1293
1294 #if !defined(WEBRTC_ANDROID) && !defined(WEBRTC_IOS)
TEST_F(ApmTest,AgcOnlyAdaptsWhenTargetSignalIsPresent)1295 TEST_F(ApmTest, AgcOnlyAdaptsWhenTargetSignalIsPresent) {
1296 const int kSampleRateHz = 16000;
1297 const size_t kSamplesPerChannel =
1298 static_cast<size_t>(AudioProcessing::kChunkSizeMs * kSampleRateHz / 1000);
1299 const size_t kNumInputChannels = 2;
1300 const size_t kNumOutputChannels = 1;
1301 const size_t kNumChunks = 700;
1302 const float kScaleFactor = 0.25f;
1303 Config config;
1304 std::vector<webrtc::Point> geometry;
1305 geometry.push_back(webrtc::Point(0.f, 0.f, 0.f));
1306 geometry.push_back(webrtc::Point(0.05f, 0.f, 0.f));
1307 config.Set<Beamforming>(new Beamforming(true, geometry));
1308 testing::NiceMock<MockNonlinearBeamformer>* beamformer =
1309 new testing::NiceMock<MockNonlinearBeamformer>(geometry);
1310 rtc::scoped_ptr<AudioProcessing> apm(
1311 AudioProcessing::Create(config, beamformer));
1312 EXPECT_EQ(kNoErr, apm->gain_control()->Enable(true));
1313 ChannelBuffer<float> src_buf(kSamplesPerChannel, kNumInputChannels);
1314 ChannelBuffer<float> dest_buf(kSamplesPerChannel, kNumOutputChannels);
1315 const size_t max_length = kSamplesPerChannel * std::max(kNumInputChannels,
1316 kNumOutputChannels);
1317 rtc::scoped_ptr<int16_t[]> int_data(new int16_t[max_length]);
1318 rtc::scoped_ptr<float[]> float_data(new float[max_length]);
1319 std::string filename = ResourceFilePath("far", kSampleRateHz);
1320 FILE* far_file = fopen(filename.c_str(), "rb");
1321 ASSERT_TRUE(far_file != NULL) << "Could not open file " << filename << "\n";
1322 const int kDefaultVolume = apm->gain_control()->stream_analog_level();
1323 const int kDefaultCompressionGain =
1324 apm->gain_control()->compression_gain_db();
1325 bool is_target = false;
1326 EXPECT_CALL(*beamformer, is_target_present())
1327 .WillRepeatedly(testing::ReturnPointee(&is_target));
1328 for (size_t i = 0; i < kNumChunks; ++i) {
1329 ASSERT_TRUE(ReadChunk(far_file,
1330 int_data.get(),
1331 float_data.get(),
1332 &src_buf));
1333 for (size_t j = 0; j < kNumInputChannels; ++j) {
1334 for (size_t k = 0; k < kSamplesPerChannel; ++k) {
1335 src_buf.channels()[j][k] *= kScaleFactor;
1336 }
1337 }
1338 EXPECT_EQ(kNoErr,
1339 apm->ProcessStream(src_buf.channels(),
1340 src_buf.num_frames(),
1341 kSampleRateHz,
1342 LayoutFromChannels(src_buf.num_channels()),
1343 kSampleRateHz,
1344 LayoutFromChannels(dest_buf.num_channels()),
1345 dest_buf.channels()));
1346 }
1347 EXPECT_EQ(kDefaultVolume,
1348 apm->gain_control()->stream_analog_level());
1349 EXPECT_EQ(kDefaultCompressionGain,
1350 apm->gain_control()->compression_gain_db());
1351 rewind(far_file);
1352 is_target = true;
1353 for (size_t i = 0; i < kNumChunks; ++i) {
1354 ASSERT_TRUE(ReadChunk(far_file,
1355 int_data.get(),
1356 float_data.get(),
1357 &src_buf));
1358 for (size_t j = 0; j < kNumInputChannels; ++j) {
1359 for (size_t k = 0; k < kSamplesPerChannel; ++k) {
1360 src_buf.channels()[j][k] *= kScaleFactor;
1361 }
1362 }
1363 EXPECT_EQ(kNoErr,
1364 apm->ProcessStream(src_buf.channels(),
1365 src_buf.num_frames(),
1366 kSampleRateHz,
1367 LayoutFromChannels(src_buf.num_channels()),
1368 kSampleRateHz,
1369 LayoutFromChannels(dest_buf.num_channels()),
1370 dest_buf.channels()));
1371 }
1372 EXPECT_LT(kDefaultVolume,
1373 apm->gain_control()->stream_analog_level());
1374 EXPECT_LT(kDefaultCompressionGain,
1375 apm->gain_control()->compression_gain_db());
1376 ASSERT_EQ(0, fclose(far_file));
1377 }
1378 #endif
1379
TEST_F(ApmTest,NoiseSuppression)1380 TEST_F(ApmTest, NoiseSuppression) {
1381 // Test valid suppression levels.
1382 NoiseSuppression::Level level[] = {
1383 NoiseSuppression::kLow,
1384 NoiseSuppression::kModerate,
1385 NoiseSuppression::kHigh,
1386 NoiseSuppression::kVeryHigh
1387 };
1388 for (size_t i = 0; i < arraysize(level); i++) {
1389 EXPECT_EQ(apm_->kNoError,
1390 apm_->noise_suppression()->set_level(level[i]));
1391 EXPECT_EQ(level[i], apm_->noise_suppression()->level());
1392 }
1393
1394 // Turn NS on/off
1395 EXPECT_EQ(apm_->kNoError, apm_->noise_suppression()->Enable(true));
1396 EXPECT_TRUE(apm_->noise_suppression()->is_enabled());
1397 EXPECT_EQ(apm_->kNoError, apm_->noise_suppression()->Enable(false));
1398 EXPECT_FALSE(apm_->noise_suppression()->is_enabled());
1399 }
1400
TEST_F(ApmTest,HighPassFilter)1401 TEST_F(ApmTest, HighPassFilter) {
1402 // Turn HP filter on/off
1403 EXPECT_EQ(apm_->kNoError, apm_->high_pass_filter()->Enable(true));
1404 EXPECT_TRUE(apm_->high_pass_filter()->is_enabled());
1405 EXPECT_EQ(apm_->kNoError, apm_->high_pass_filter()->Enable(false));
1406 EXPECT_FALSE(apm_->high_pass_filter()->is_enabled());
1407 }
1408
TEST_F(ApmTest,LevelEstimator)1409 TEST_F(ApmTest, LevelEstimator) {
1410 // Turn level estimator on/off
1411 EXPECT_EQ(apm_->kNoError, apm_->level_estimator()->Enable(false));
1412 EXPECT_FALSE(apm_->level_estimator()->is_enabled());
1413
1414 EXPECT_EQ(apm_->kNotEnabledError, apm_->level_estimator()->RMS());
1415
1416 EXPECT_EQ(apm_->kNoError, apm_->level_estimator()->Enable(true));
1417 EXPECT_TRUE(apm_->level_estimator()->is_enabled());
1418
1419 // Run this test in wideband; in super-wb, the splitting filter distorts the
1420 // audio enough to cause deviation from the expectation for small values.
1421 frame_->samples_per_channel_ = 160;
1422 frame_->num_channels_ = 2;
1423 frame_->sample_rate_hz_ = 16000;
1424
1425 // Min value if no frames have been processed.
1426 EXPECT_EQ(127, apm_->level_estimator()->RMS());
1427
1428 // Min value on zero frames.
1429 SetFrameTo(frame_, 0);
1430 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1431 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1432 EXPECT_EQ(127, apm_->level_estimator()->RMS());
1433
1434 // Try a few RMS values.
1435 // (These also test that the value resets after retrieving it.)
1436 SetFrameTo(frame_, 32767);
1437 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1438 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1439 EXPECT_EQ(0, apm_->level_estimator()->RMS());
1440
1441 SetFrameTo(frame_, 30000);
1442 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1443 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1444 EXPECT_EQ(1, apm_->level_estimator()->RMS());
1445
1446 SetFrameTo(frame_, 10000);
1447 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1448 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1449 EXPECT_EQ(10, apm_->level_estimator()->RMS());
1450
1451 SetFrameTo(frame_, 10);
1452 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1453 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1454 EXPECT_EQ(70, apm_->level_estimator()->RMS());
1455
1456 // Verify reset after enable/disable.
1457 SetFrameTo(frame_, 32767);
1458 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1459 EXPECT_EQ(apm_->kNoError, apm_->level_estimator()->Enable(false));
1460 EXPECT_EQ(apm_->kNoError, apm_->level_estimator()->Enable(true));
1461 SetFrameTo(frame_, 1);
1462 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1463 EXPECT_EQ(90, apm_->level_estimator()->RMS());
1464
1465 // Verify reset after initialize.
1466 SetFrameTo(frame_, 32767);
1467 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1468 EXPECT_EQ(apm_->kNoError, apm_->Initialize());
1469 SetFrameTo(frame_, 1);
1470 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1471 EXPECT_EQ(90, apm_->level_estimator()->RMS());
1472 }
1473
TEST_F(ApmTest,VoiceDetection)1474 TEST_F(ApmTest, VoiceDetection) {
1475 // Test external VAD
1476 EXPECT_EQ(apm_->kNoError,
1477 apm_->voice_detection()->set_stream_has_voice(true));
1478 EXPECT_TRUE(apm_->voice_detection()->stream_has_voice());
1479 EXPECT_EQ(apm_->kNoError,
1480 apm_->voice_detection()->set_stream_has_voice(false));
1481 EXPECT_FALSE(apm_->voice_detection()->stream_has_voice());
1482
1483 // Test valid likelihoods
1484 VoiceDetection::Likelihood likelihood[] = {
1485 VoiceDetection::kVeryLowLikelihood,
1486 VoiceDetection::kLowLikelihood,
1487 VoiceDetection::kModerateLikelihood,
1488 VoiceDetection::kHighLikelihood
1489 };
1490 for (size_t i = 0; i < arraysize(likelihood); i++) {
1491 EXPECT_EQ(apm_->kNoError,
1492 apm_->voice_detection()->set_likelihood(likelihood[i]));
1493 EXPECT_EQ(likelihood[i], apm_->voice_detection()->likelihood());
1494 }
1495
1496 /* TODO(bjornv): Enable once VAD supports other frame lengths than 10 ms
1497 // Test invalid frame sizes
1498 EXPECT_EQ(apm_->kBadParameterError,
1499 apm_->voice_detection()->set_frame_size_ms(12));
1500
1501 // Test valid frame sizes
1502 for (int i = 10; i <= 30; i += 10) {
1503 EXPECT_EQ(apm_->kNoError,
1504 apm_->voice_detection()->set_frame_size_ms(i));
1505 EXPECT_EQ(i, apm_->voice_detection()->frame_size_ms());
1506 }
1507 */
1508
1509 // Turn VAD on/off
1510 EXPECT_EQ(apm_->kNoError, apm_->voice_detection()->Enable(true));
1511 EXPECT_TRUE(apm_->voice_detection()->is_enabled());
1512 EXPECT_EQ(apm_->kNoError, apm_->voice_detection()->Enable(false));
1513 EXPECT_FALSE(apm_->voice_detection()->is_enabled());
1514
1515 // Test that AudioFrame activity is maintained when VAD is disabled.
1516 EXPECT_EQ(apm_->kNoError, apm_->voice_detection()->Enable(false));
1517 AudioFrame::VADActivity activity[] = {
1518 AudioFrame::kVadActive,
1519 AudioFrame::kVadPassive,
1520 AudioFrame::kVadUnknown
1521 };
1522 for (size_t i = 0; i < arraysize(activity); i++) {
1523 frame_->vad_activity_ = activity[i];
1524 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1525 EXPECT_EQ(activity[i], frame_->vad_activity_);
1526 }
1527
1528 // Test that AudioFrame activity is set when VAD is enabled.
1529 EXPECT_EQ(apm_->kNoError, apm_->voice_detection()->Enable(true));
1530 frame_->vad_activity_ = AudioFrame::kVadUnknown;
1531 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1532 EXPECT_NE(AudioFrame::kVadUnknown, frame_->vad_activity_);
1533
1534 // TODO(bjornv): Add tests for streamed voice; stream_has_voice()
1535 }
1536
TEST_F(ApmTest,AllProcessingDisabledByDefault)1537 TEST_F(ApmTest, AllProcessingDisabledByDefault) {
1538 EXPECT_FALSE(apm_->echo_cancellation()->is_enabled());
1539 EXPECT_FALSE(apm_->echo_control_mobile()->is_enabled());
1540 EXPECT_FALSE(apm_->gain_control()->is_enabled());
1541 EXPECT_FALSE(apm_->high_pass_filter()->is_enabled());
1542 EXPECT_FALSE(apm_->level_estimator()->is_enabled());
1543 EXPECT_FALSE(apm_->noise_suppression()->is_enabled());
1544 EXPECT_FALSE(apm_->voice_detection()->is_enabled());
1545 }
1546
TEST_F(ApmTest,NoProcessingWhenAllComponentsDisabled)1547 TEST_F(ApmTest, NoProcessingWhenAllComponentsDisabled) {
1548 for (size_t i = 0; i < arraysize(kSampleRates); i++) {
1549 Init(kSampleRates[i], kSampleRates[i], kSampleRates[i], 2, 2, 2, false);
1550 SetFrameTo(frame_, 1000, 2000);
1551 AudioFrame frame_copy;
1552 frame_copy.CopyFrom(*frame_);
1553 for (int j = 0; j < 1000; j++) {
1554 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1555 EXPECT_TRUE(FrameDataAreEqual(*frame_, frame_copy));
1556 EXPECT_EQ(apm_->kNoError, apm_->ProcessReverseStream(frame_));
1557 EXPECT_TRUE(FrameDataAreEqual(*frame_, frame_copy));
1558 }
1559 }
1560 }
1561
TEST_F(ApmTest,NoProcessingWhenAllComponentsDisabledFloat)1562 TEST_F(ApmTest, NoProcessingWhenAllComponentsDisabledFloat) {
1563 // Test that ProcessStream copies input to output even with no processing.
1564 const size_t kSamples = 80;
1565 const int sample_rate = 8000;
1566 const float src[kSamples] = {
1567 -1.0f, 0.0f, 1.0f
1568 };
1569 float dest[kSamples] = {};
1570
1571 auto src_channels = &src[0];
1572 auto dest_channels = &dest[0];
1573
1574 apm_.reset(AudioProcessing::Create());
1575 EXPECT_NOERR(apm_->ProcessStream(
1576 &src_channels, kSamples, sample_rate, LayoutFromChannels(1),
1577 sample_rate, LayoutFromChannels(1), &dest_channels));
1578
1579 for (size_t i = 0; i < kSamples; ++i) {
1580 EXPECT_EQ(src[i], dest[i]);
1581 }
1582
1583 // Same for ProcessReverseStream.
1584 float rev_dest[kSamples] = {};
1585 auto rev_dest_channels = &rev_dest[0];
1586
1587 StreamConfig input_stream = {sample_rate, 1};
1588 StreamConfig output_stream = {sample_rate, 1};
1589 EXPECT_NOERR(apm_->ProcessReverseStream(&src_channels, input_stream,
1590 output_stream, &rev_dest_channels));
1591
1592 for (size_t i = 0; i < kSamples; ++i) {
1593 EXPECT_EQ(src[i], rev_dest[i]);
1594 }
1595 }
1596
TEST_F(ApmTest,IdenticalInputChannelsResultInIdenticalOutputChannels)1597 TEST_F(ApmTest, IdenticalInputChannelsResultInIdenticalOutputChannels) {
1598 EnableAllComponents();
1599
1600 for (size_t i = 0; i < arraysize(kProcessSampleRates); i++) {
1601 Init(kProcessSampleRates[i],
1602 kProcessSampleRates[i],
1603 kProcessSampleRates[i],
1604 2,
1605 2,
1606 2,
1607 false);
1608 int analog_level = 127;
1609 ASSERT_EQ(0, feof(far_file_));
1610 ASSERT_EQ(0, feof(near_file_));
1611 while (ReadFrame(far_file_, revframe_) && ReadFrame(near_file_, frame_)) {
1612 CopyLeftToRightChannel(revframe_->data_, revframe_->samples_per_channel_);
1613
1614 ASSERT_EQ(kNoErr, apm_->AnalyzeReverseStream(revframe_));
1615
1616 CopyLeftToRightChannel(frame_->data_, frame_->samples_per_channel_);
1617 frame_->vad_activity_ = AudioFrame::kVadUnknown;
1618
1619 ASSERT_EQ(kNoErr, apm_->set_stream_delay_ms(0));
1620 apm_->echo_cancellation()->set_stream_drift_samples(0);
1621 ASSERT_EQ(kNoErr,
1622 apm_->gain_control()->set_stream_analog_level(analog_level));
1623 ASSERT_EQ(kNoErr, apm_->ProcessStream(frame_));
1624 analog_level = apm_->gain_control()->stream_analog_level();
1625
1626 VerifyChannelsAreEqual(frame_->data_, frame_->samples_per_channel_);
1627 }
1628 rewind(far_file_);
1629 rewind(near_file_);
1630 }
1631 }
1632
TEST_F(ApmTest,SplittingFilter)1633 TEST_F(ApmTest, SplittingFilter) {
1634 // Verify the filter is not active through undistorted audio when:
1635 // 1. No components are enabled...
1636 SetFrameTo(frame_, 1000);
1637 AudioFrame frame_copy;
1638 frame_copy.CopyFrom(*frame_);
1639 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1640 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1641 EXPECT_TRUE(FrameDataAreEqual(*frame_, frame_copy));
1642
1643 // 2. Only the level estimator is enabled...
1644 SetFrameTo(frame_, 1000);
1645 frame_copy.CopyFrom(*frame_);
1646 EXPECT_EQ(apm_->kNoError, apm_->level_estimator()->Enable(true));
1647 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1648 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1649 EXPECT_TRUE(FrameDataAreEqual(*frame_, frame_copy));
1650 EXPECT_EQ(apm_->kNoError, apm_->level_estimator()->Enable(false));
1651
1652 // 3. Only VAD is enabled...
1653 SetFrameTo(frame_, 1000);
1654 frame_copy.CopyFrom(*frame_);
1655 EXPECT_EQ(apm_->kNoError, apm_->voice_detection()->Enable(true));
1656 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1657 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1658 EXPECT_TRUE(FrameDataAreEqual(*frame_, frame_copy));
1659 EXPECT_EQ(apm_->kNoError, apm_->voice_detection()->Enable(false));
1660
1661 // 4. Both VAD and the level estimator are enabled...
1662 SetFrameTo(frame_, 1000);
1663 frame_copy.CopyFrom(*frame_);
1664 EXPECT_EQ(apm_->kNoError, apm_->level_estimator()->Enable(true));
1665 EXPECT_EQ(apm_->kNoError, apm_->voice_detection()->Enable(true));
1666 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1667 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1668 EXPECT_TRUE(FrameDataAreEqual(*frame_, frame_copy));
1669 EXPECT_EQ(apm_->kNoError, apm_->level_estimator()->Enable(false));
1670 EXPECT_EQ(apm_->kNoError, apm_->voice_detection()->Enable(false));
1671
1672 // TODO(aluebs): Figure out exactly why the AEC affects the audio on Android.
1673 /*// 5. Not using super-wb.
1674 frame_->samples_per_channel_ = 160;
1675 frame_->num_channels_ = 2;
1676 frame_->sample_rate_hz_ = 16000;
1677 // Enable AEC, which would require the filter in super-wb. We rely on the
1678 // first few frames of data being unaffected by the AEC.
1679 // TODO(andrew): This test, and the one below, rely rather tenuously on the
1680 // behavior of the AEC. Think of something more robust.
1681 EXPECT_EQ(apm_->kNoError, apm_->echo_cancellation()->Enable(true));
1682 // Make sure we have extended filter enabled. This makes sure nothing is
1683 // touched until we have a farend frame.
1684 Config config;
1685 config.Set<ExtendedFilter>(new ExtendedFilter(true));
1686 apm_->SetExtraOptions(config);
1687 SetFrameTo(frame_, 1000);
1688 frame_copy.CopyFrom(*frame_);
1689 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(0));
1690 apm_->echo_cancellation()->set_stream_drift_samples(0);
1691 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1692 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(0));
1693 apm_->echo_cancellation()->set_stream_drift_samples(0);
1694 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1695 EXPECT_TRUE(FrameDataAreEqual(*frame_, frame_copy));
1696
1697 // Check the test is valid. We should have distortion from the filter
1698 // when AEC is enabled (which won't affect the audio).
1699 frame_->samples_per_channel_ = 320;
1700 frame_->num_channels_ = 2;
1701 frame_->sample_rate_hz_ = 32000;
1702 SetFrameTo(frame_, 1000);
1703 frame_copy.CopyFrom(*frame_);
1704 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(0));
1705 apm_->echo_cancellation()->set_stream_drift_samples(0);
1706 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1707 EXPECT_FALSE(FrameDataAreEqual(*frame_, frame_copy));*/
1708 }
1709
1710 #ifdef WEBRTC_AUDIOPROC_DEBUG_DUMP
ProcessDebugDump(const std::string & in_filename,const std::string & out_filename,Format format)1711 void ApmTest::ProcessDebugDump(const std::string& in_filename,
1712 const std::string& out_filename,
1713 Format format) {
1714 FILE* in_file = fopen(in_filename.c_str(), "rb");
1715 ASSERT_TRUE(in_file != NULL);
1716 audioproc::Event event_msg;
1717 bool first_init = true;
1718
1719 while (ReadMessageFromFile(in_file, &event_msg)) {
1720 if (event_msg.type() == audioproc::Event::INIT) {
1721 const audioproc::Init msg = event_msg.init();
1722 int reverse_sample_rate = msg.sample_rate();
1723 if (msg.has_reverse_sample_rate()) {
1724 reverse_sample_rate = msg.reverse_sample_rate();
1725 }
1726 int output_sample_rate = msg.sample_rate();
1727 if (msg.has_output_sample_rate()) {
1728 output_sample_rate = msg.output_sample_rate();
1729 }
1730
1731 Init(msg.sample_rate(),
1732 output_sample_rate,
1733 reverse_sample_rate,
1734 msg.num_input_channels(),
1735 msg.num_output_channels(),
1736 msg.num_reverse_channels(),
1737 false);
1738 if (first_init) {
1739 // StartDebugRecording() writes an additional init message. Don't start
1740 // recording until after the first init to avoid the extra message.
1741 EXPECT_NOERR(apm_->StartDebugRecording(out_filename.c_str()));
1742 first_init = false;
1743 }
1744
1745 } else if (event_msg.type() == audioproc::Event::REVERSE_STREAM) {
1746 const audioproc::ReverseStream msg = event_msg.reverse_stream();
1747
1748 if (msg.channel_size() > 0) {
1749 ASSERT_EQ(revframe_->num_channels_,
1750 static_cast<size_t>(msg.channel_size()));
1751 for (int i = 0; i < msg.channel_size(); ++i) {
1752 memcpy(revfloat_cb_->channels()[i],
1753 msg.channel(i).data(),
1754 msg.channel(i).size());
1755 }
1756 } else {
1757 memcpy(revframe_->data_, msg.data().data(), msg.data().size());
1758 if (format == kFloatFormat) {
1759 // We're using an int16 input file; convert to float.
1760 ConvertToFloat(*revframe_, revfloat_cb_.get());
1761 }
1762 }
1763 AnalyzeReverseStreamChooser(format);
1764
1765 } else if (event_msg.type() == audioproc::Event::STREAM) {
1766 const audioproc::Stream msg = event_msg.stream();
1767 // ProcessStream could have changed this for the output frame.
1768 frame_->num_channels_ = apm_->num_input_channels();
1769
1770 EXPECT_NOERR(apm_->gain_control()->set_stream_analog_level(msg.level()));
1771 EXPECT_NOERR(apm_->set_stream_delay_ms(msg.delay()));
1772 apm_->echo_cancellation()->set_stream_drift_samples(msg.drift());
1773 if (msg.has_keypress()) {
1774 apm_->set_stream_key_pressed(msg.keypress());
1775 } else {
1776 apm_->set_stream_key_pressed(true);
1777 }
1778
1779 if (msg.input_channel_size() > 0) {
1780 ASSERT_EQ(frame_->num_channels_,
1781 static_cast<size_t>(msg.input_channel_size()));
1782 for (int i = 0; i < msg.input_channel_size(); ++i) {
1783 memcpy(float_cb_->channels()[i],
1784 msg.input_channel(i).data(),
1785 msg.input_channel(i).size());
1786 }
1787 } else {
1788 memcpy(frame_->data_, msg.input_data().data(), msg.input_data().size());
1789 if (format == kFloatFormat) {
1790 // We're using an int16 input file; convert to float.
1791 ConvertToFloat(*frame_, float_cb_.get());
1792 }
1793 }
1794 ProcessStreamChooser(format);
1795 }
1796 }
1797 EXPECT_NOERR(apm_->StopDebugRecording());
1798 fclose(in_file);
1799 }
1800
VerifyDebugDumpTest(Format format)1801 void ApmTest::VerifyDebugDumpTest(Format format) {
1802 const std::string in_filename = test::ResourcePath("ref03", "aecdump");
1803 std::string format_string;
1804 switch (format) {
1805 case kIntFormat:
1806 format_string = "_int";
1807 break;
1808 case kFloatFormat:
1809 format_string = "_float";
1810 break;
1811 }
1812 const std::string ref_filename = test::TempFilename(
1813 test::OutputPath(), std::string("ref") + format_string + "_aecdump");
1814 const std::string out_filename = test::TempFilename(
1815 test::OutputPath(), std::string("out") + format_string + "_aecdump");
1816 EnableAllComponents();
1817 ProcessDebugDump(in_filename, ref_filename, format);
1818 ProcessDebugDump(ref_filename, out_filename, format);
1819
1820 FILE* ref_file = fopen(ref_filename.c_str(), "rb");
1821 FILE* out_file = fopen(out_filename.c_str(), "rb");
1822 ASSERT_TRUE(ref_file != NULL);
1823 ASSERT_TRUE(out_file != NULL);
1824 rtc::scoped_ptr<uint8_t[]> ref_bytes;
1825 rtc::scoped_ptr<uint8_t[]> out_bytes;
1826
1827 size_t ref_size = ReadMessageBytesFromFile(ref_file, &ref_bytes);
1828 size_t out_size = ReadMessageBytesFromFile(out_file, &out_bytes);
1829 size_t bytes_read = 0;
1830 while (ref_size > 0 && out_size > 0) {
1831 bytes_read += ref_size;
1832 EXPECT_EQ(ref_size, out_size);
1833 EXPECT_EQ(0, memcmp(ref_bytes.get(), out_bytes.get(), ref_size));
1834 ref_size = ReadMessageBytesFromFile(ref_file, &ref_bytes);
1835 out_size = ReadMessageBytesFromFile(out_file, &out_bytes);
1836 }
1837 EXPECT_GT(bytes_read, 0u);
1838 EXPECT_NE(0, feof(ref_file));
1839 EXPECT_NE(0, feof(out_file));
1840 ASSERT_EQ(0, fclose(ref_file));
1841 ASSERT_EQ(0, fclose(out_file));
1842 remove(ref_filename.c_str());
1843 remove(out_filename.c_str());
1844 }
1845
TEST_F(ApmTest,VerifyDebugDumpInt)1846 TEST_F(ApmTest, VerifyDebugDumpInt) {
1847 VerifyDebugDumpTest(kIntFormat);
1848 }
1849
TEST_F(ApmTest,VerifyDebugDumpFloat)1850 TEST_F(ApmTest, VerifyDebugDumpFloat) {
1851 VerifyDebugDumpTest(kFloatFormat);
1852 }
1853 #endif
1854
1855 // TODO(andrew): expand test to verify output.
TEST_F(ApmTest,DebugDump)1856 TEST_F(ApmTest, DebugDump) {
1857 const std::string filename =
1858 test::TempFilename(test::OutputPath(), "debug_aec");
1859 EXPECT_EQ(apm_->kNullPointerError,
1860 apm_->StartDebugRecording(static_cast<const char*>(NULL)));
1861
1862 #ifdef WEBRTC_AUDIOPROC_DEBUG_DUMP
1863 // Stopping without having started should be OK.
1864 EXPECT_EQ(apm_->kNoError, apm_->StopDebugRecording());
1865
1866 EXPECT_EQ(apm_->kNoError, apm_->StartDebugRecording(filename.c_str()));
1867 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1868 EXPECT_EQ(apm_->kNoError, apm_->AnalyzeReverseStream(revframe_));
1869 EXPECT_EQ(apm_->kNoError, apm_->StopDebugRecording());
1870
1871 // Verify the file has been written.
1872 FILE* fid = fopen(filename.c_str(), "r");
1873 ASSERT_TRUE(fid != NULL);
1874
1875 // Clean it up.
1876 ASSERT_EQ(0, fclose(fid));
1877 ASSERT_EQ(0, remove(filename.c_str()));
1878 #else
1879 EXPECT_EQ(apm_->kUnsupportedFunctionError,
1880 apm_->StartDebugRecording(filename.c_str()));
1881 EXPECT_EQ(apm_->kUnsupportedFunctionError, apm_->StopDebugRecording());
1882
1883 // Verify the file has NOT been written.
1884 ASSERT_TRUE(fopen(filename.c_str(), "r") == NULL);
1885 #endif // WEBRTC_AUDIOPROC_DEBUG_DUMP
1886 }
1887
1888 // TODO(andrew): expand test to verify output.
TEST_F(ApmTest,DebugDumpFromFileHandle)1889 TEST_F(ApmTest, DebugDumpFromFileHandle) {
1890 FILE* fid = NULL;
1891 EXPECT_EQ(apm_->kNullPointerError, apm_->StartDebugRecording(fid));
1892 const std::string filename =
1893 test::TempFilename(test::OutputPath(), "debug_aec");
1894 fid = fopen(filename.c_str(), "w");
1895 ASSERT_TRUE(fid);
1896
1897 #ifdef WEBRTC_AUDIOPROC_DEBUG_DUMP
1898 // Stopping without having started should be OK.
1899 EXPECT_EQ(apm_->kNoError, apm_->StopDebugRecording());
1900
1901 EXPECT_EQ(apm_->kNoError, apm_->StartDebugRecording(fid));
1902 EXPECT_EQ(apm_->kNoError, apm_->AnalyzeReverseStream(revframe_));
1903 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
1904 EXPECT_EQ(apm_->kNoError, apm_->StopDebugRecording());
1905
1906 // Verify the file has been written.
1907 fid = fopen(filename.c_str(), "r");
1908 ASSERT_TRUE(fid != NULL);
1909
1910 // Clean it up.
1911 ASSERT_EQ(0, fclose(fid));
1912 ASSERT_EQ(0, remove(filename.c_str()));
1913 #else
1914 EXPECT_EQ(apm_->kUnsupportedFunctionError,
1915 apm_->StartDebugRecording(fid));
1916 EXPECT_EQ(apm_->kUnsupportedFunctionError, apm_->StopDebugRecording());
1917
1918 ASSERT_EQ(0, fclose(fid));
1919 #endif // WEBRTC_AUDIOPROC_DEBUG_DUMP
1920 }
1921
TEST_F(ApmTest,FloatAndIntInterfacesGiveSimilarResults)1922 TEST_F(ApmTest, FloatAndIntInterfacesGiveSimilarResults) {
1923 audioproc::OutputData ref_data;
1924 OpenFileAndReadMessage(ref_filename_, &ref_data);
1925
1926 Config config;
1927 config.Set<ExperimentalAgc>(new ExperimentalAgc(false));
1928 rtc::scoped_ptr<AudioProcessing> fapm(AudioProcessing::Create(config));
1929 EnableAllComponents();
1930 EnableAllAPComponents(fapm.get());
1931 for (int i = 0; i < ref_data.test_size(); i++) {
1932 printf("Running test %d of %d...\n", i + 1, ref_data.test_size());
1933
1934 audioproc::Test* test = ref_data.mutable_test(i);
1935 // TODO(ajm): Restore downmixing test cases.
1936 if (test->num_input_channels() != test->num_output_channels())
1937 continue;
1938
1939 const size_t num_render_channels =
1940 static_cast<size_t>(test->num_reverse_channels());
1941 const size_t num_input_channels =
1942 static_cast<size_t>(test->num_input_channels());
1943 const size_t num_output_channels =
1944 static_cast<size_t>(test->num_output_channels());
1945 const size_t samples_per_channel = static_cast<size_t>(
1946 test->sample_rate() * AudioProcessing::kChunkSizeMs / 1000);
1947
1948 Init(test->sample_rate(), test->sample_rate(), test->sample_rate(),
1949 num_input_channels, num_output_channels, num_render_channels, true);
1950 Init(fapm.get());
1951
1952 ChannelBuffer<int16_t> output_cb(samples_per_channel, num_input_channels);
1953 ChannelBuffer<int16_t> output_int16(samples_per_channel,
1954 num_input_channels);
1955
1956 int analog_level = 127;
1957 while (ReadFrame(far_file_, revframe_, revfloat_cb_.get()) &&
1958 ReadFrame(near_file_, frame_, float_cb_.get())) {
1959 frame_->vad_activity_ = AudioFrame::kVadUnknown;
1960
1961 EXPECT_NOERR(apm_->AnalyzeReverseStream(revframe_));
1962 EXPECT_NOERR(fapm->AnalyzeReverseStream(
1963 revfloat_cb_->channels(),
1964 samples_per_channel,
1965 test->sample_rate(),
1966 LayoutFromChannels(num_render_channels)));
1967
1968 EXPECT_NOERR(apm_->set_stream_delay_ms(0));
1969 EXPECT_NOERR(fapm->set_stream_delay_ms(0));
1970 apm_->echo_cancellation()->set_stream_drift_samples(0);
1971 fapm->echo_cancellation()->set_stream_drift_samples(0);
1972 EXPECT_NOERR(apm_->gain_control()->set_stream_analog_level(analog_level));
1973 EXPECT_NOERR(fapm->gain_control()->set_stream_analog_level(analog_level));
1974
1975 EXPECT_NOERR(apm_->ProcessStream(frame_));
1976 Deinterleave(frame_->data_, samples_per_channel, num_output_channels,
1977 output_int16.channels());
1978
1979 EXPECT_NOERR(fapm->ProcessStream(
1980 float_cb_->channels(),
1981 samples_per_channel,
1982 test->sample_rate(),
1983 LayoutFromChannels(num_input_channels),
1984 test->sample_rate(),
1985 LayoutFromChannels(num_output_channels),
1986 float_cb_->channels()));
1987 for (size_t j = 0; j < num_output_channels; ++j) {
1988 FloatToS16(float_cb_->channels()[j],
1989 samples_per_channel,
1990 output_cb.channels()[j]);
1991 float variance = 0;
1992 float snr = ComputeSNR(output_int16.channels()[j],
1993 output_cb.channels()[j],
1994 samples_per_channel, &variance);
1995 #if defined(WEBRTC_AUDIOPROC_FIXED_PROFILE)
1996 // There are a few chunks in the fixed-point profile that give low SNR.
1997 // Listening confirmed the difference is acceptable.
1998 const float kVarianceThreshold = 150;
1999 const float kSNRThreshold = 10;
2000 #else
2001 const float kVarianceThreshold = 20;
2002 const float kSNRThreshold = 20;
2003 #endif
2004 // Skip frames with low energy.
2005 if (sqrt(variance) > kVarianceThreshold) {
2006 EXPECT_LT(kSNRThreshold, snr);
2007 }
2008 }
2009
2010 analog_level = fapm->gain_control()->stream_analog_level();
2011 EXPECT_EQ(apm_->gain_control()->stream_analog_level(),
2012 fapm->gain_control()->stream_analog_level());
2013 EXPECT_EQ(apm_->echo_cancellation()->stream_has_echo(),
2014 fapm->echo_cancellation()->stream_has_echo());
2015 EXPECT_NEAR(apm_->noise_suppression()->speech_probability(),
2016 fapm->noise_suppression()->speech_probability(),
2017 0.01);
2018
2019 // Reset in case of downmixing.
2020 frame_->num_channels_ = static_cast<size_t>(test->num_input_channels());
2021 }
2022 rewind(far_file_);
2023 rewind(near_file_);
2024 }
2025 }
2026
2027 // TODO(andrew): Add a test to process a few frames with different combinations
2028 // of enabled components.
2029
TEST_F(ApmTest,Process)2030 TEST_F(ApmTest, Process) {
2031 GOOGLE_PROTOBUF_VERIFY_VERSION;
2032 audioproc::OutputData ref_data;
2033
2034 if (!write_ref_data) {
2035 OpenFileAndReadMessage(ref_filename_, &ref_data);
2036 } else {
2037 // Write the desired tests to the protobuf reference file.
2038 for (size_t i = 0; i < arraysize(kChannels); i++) {
2039 for (size_t j = 0; j < arraysize(kChannels); j++) {
2040 for (size_t l = 0; l < arraysize(kProcessSampleRates); l++) {
2041 audioproc::Test* test = ref_data.add_test();
2042 test->set_num_reverse_channels(kChannels[i]);
2043 test->set_num_input_channels(kChannels[j]);
2044 test->set_num_output_channels(kChannels[j]);
2045 test->set_sample_rate(kProcessSampleRates[l]);
2046 test->set_use_aec_extended_filter(false);
2047 }
2048 }
2049 }
2050 #if defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
2051 // To test the extended filter mode.
2052 audioproc::Test* test = ref_data.add_test();
2053 test->set_num_reverse_channels(2);
2054 test->set_num_input_channels(2);
2055 test->set_num_output_channels(2);
2056 test->set_sample_rate(AudioProcessing::kSampleRate32kHz);
2057 test->set_use_aec_extended_filter(true);
2058 #endif
2059 }
2060
2061 for (int i = 0; i < ref_data.test_size(); i++) {
2062 printf("Running test %d of %d...\n", i + 1, ref_data.test_size());
2063
2064 audioproc::Test* test = ref_data.mutable_test(i);
2065 // TODO(ajm): We no longer allow different input and output channels. Skip
2066 // these tests for now, but they should be removed from the set.
2067 if (test->num_input_channels() != test->num_output_channels())
2068 continue;
2069
2070 Config config;
2071 config.Set<ExperimentalAgc>(new ExperimentalAgc(false));
2072 config.Set<ExtendedFilter>(
2073 new ExtendedFilter(test->use_aec_extended_filter()));
2074 apm_.reset(AudioProcessing::Create(config));
2075
2076 EnableAllComponents();
2077
2078 Init(test->sample_rate(),
2079 test->sample_rate(),
2080 test->sample_rate(),
2081 static_cast<size_t>(test->num_input_channels()),
2082 static_cast<size_t>(test->num_output_channels()),
2083 static_cast<size_t>(test->num_reverse_channels()),
2084 true);
2085
2086 int frame_count = 0;
2087 int has_echo_count = 0;
2088 int has_voice_count = 0;
2089 int is_saturated_count = 0;
2090 int analog_level = 127;
2091 int analog_level_average = 0;
2092 int max_output_average = 0;
2093 float ns_speech_prob_average = 0.0f;
2094
2095 while (ReadFrame(far_file_, revframe_) && ReadFrame(near_file_, frame_)) {
2096 EXPECT_EQ(apm_->kNoError, apm_->AnalyzeReverseStream(revframe_));
2097
2098 frame_->vad_activity_ = AudioFrame::kVadUnknown;
2099
2100 EXPECT_EQ(apm_->kNoError, apm_->set_stream_delay_ms(0));
2101 apm_->echo_cancellation()->set_stream_drift_samples(0);
2102 EXPECT_EQ(apm_->kNoError,
2103 apm_->gain_control()->set_stream_analog_level(analog_level));
2104
2105 EXPECT_EQ(apm_->kNoError, apm_->ProcessStream(frame_));
2106
2107 // Ensure the frame was downmixed properly.
2108 EXPECT_EQ(static_cast<size_t>(test->num_output_channels()),
2109 frame_->num_channels_);
2110
2111 max_output_average += MaxAudioFrame(*frame_);
2112
2113 if (apm_->echo_cancellation()->stream_has_echo()) {
2114 has_echo_count++;
2115 }
2116
2117 analog_level = apm_->gain_control()->stream_analog_level();
2118 analog_level_average += analog_level;
2119 if (apm_->gain_control()->stream_is_saturated()) {
2120 is_saturated_count++;
2121 }
2122 if (apm_->voice_detection()->stream_has_voice()) {
2123 has_voice_count++;
2124 EXPECT_EQ(AudioFrame::kVadActive, frame_->vad_activity_);
2125 } else {
2126 EXPECT_EQ(AudioFrame::kVadPassive, frame_->vad_activity_);
2127 }
2128
2129 ns_speech_prob_average += apm_->noise_suppression()->speech_probability();
2130
2131 size_t frame_size = frame_->samples_per_channel_ * frame_->num_channels_;
2132 size_t write_count = fwrite(frame_->data_,
2133 sizeof(int16_t),
2134 frame_size,
2135 out_file_);
2136 ASSERT_EQ(frame_size, write_count);
2137
2138 // Reset in case of downmixing.
2139 frame_->num_channels_ = static_cast<size_t>(test->num_input_channels());
2140 frame_count++;
2141 }
2142 max_output_average /= frame_count;
2143 analog_level_average /= frame_count;
2144 ns_speech_prob_average /= frame_count;
2145
2146 #if defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
2147 EchoCancellation::Metrics echo_metrics;
2148 EXPECT_EQ(apm_->kNoError,
2149 apm_->echo_cancellation()->GetMetrics(&echo_metrics));
2150 int median = 0;
2151 int std = 0;
2152 float fraction_poor_delays = 0;
2153 EXPECT_EQ(apm_->kNoError,
2154 apm_->echo_cancellation()->GetDelayMetrics(
2155 &median, &std, &fraction_poor_delays));
2156
2157 int rms_level = apm_->level_estimator()->RMS();
2158 EXPECT_LE(0, rms_level);
2159 EXPECT_GE(127, rms_level);
2160 #endif
2161
2162 if (!write_ref_data) {
2163 const int kIntNear = 1;
2164 // When running the test on a N7 we get a {2, 6} difference of
2165 // |has_voice_count| and |max_output_average| is up to 18 higher.
2166 // All numbers being consistently higher on N7 compare to ref_data.
2167 // TODO(bjornv): If we start getting more of these offsets on Android we
2168 // should consider a different approach. Either using one slack for all,
2169 // or generate a separate android reference.
2170 #if defined(WEBRTC_ANDROID)
2171 const int kHasVoiceCountOffset = 3;
2172 const int kHasVoiceCountNear = 3;
2173 const int kMaxOutputAverageOffset = 9;
2174 const int kMaxOutputAverageNear = 9;
2175 #else
2176 const int kHasVoiceCountOffset = 0;
2177 const int kHasVoiceCountNear = kIntNear;
2178 const int kMaxOutputAverageOffset = 0;
2179 const int kMaxOutputAverageNear = kIntNear;
2180 #endif
2181 EXPECT_NEAR(test->has_echo_count(), has_echo_count, kIntNear);
2182 EXPECT_NEAR(test->has_voice_count(),
2183 has_voice_count - kHasVoiceCountOffset,
2184 kHasVoiceCountNear);
2185 EXPECT_NEAR(test->is_saturated_count(), is_saturated_count, kIntNear);
2186
2187 EXPECT_NEAR(test->analog_level_average(), analog_level_average, kIntNear);
2188 EXPECT_NEAR(test->max_output_average(),
2189 max_output_average - kMaxOutputAverageOffset,
2190 kMaxOutputAverageNear);
2191
2192 #if defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
2193 audioproc::Test::EchoMetrics reference = test->echo_metrics();
2194 TestStats(echo_metrics.residual_echo_return_loss,
2195 reference.residual_echo_return_loss());
2196 TestStats(echo_metrics.echo_return_loss,
2197 reference.echo_return_loss());
2198 TestStats(echo_metrics.echo_return_loss_enhancement,
2199 reference.echo_return_loss_enhancement());
2200 TestStats(echo_metrics.a_nlp,
2201 reference.a_nlp());
2202
2203 const double kFloatNear = 0.0005;
2204 audioproc::Test::DelayMetrics reference_delay = test->delay_metrics();
2205 EXPECT_NEAR(reference_delay.median(), median, kIntNear);
2206 EXPECT_NEAR(reference_delay.std(), std, kIntNear);
2207 EXPECT_NEAR(reference_delay.fraction_poor_delays(), fraction_poor_delays,
2208 kFloatNear);
2209
2210 EXPECT_NEAR(test->rms_level(), rms_level, kIntNear);
2211
2212 EXPECT_NEAR(test->ns_speech_probability_average(),
2213 ns_speech_prob_average,
2214 kFloatNear);
2215 #endif
2216 } else {
2217 test->set_has_echo_count(has_echo_count);
2218 test->set_has_voice_count(has_voice_count);
2219 test->set_is_saturated_count(is_saturated_count);
2220
2221 test->set_analog_level_average(analog_level_average);
2222 test->set_max_output_average(max_output_average);
2223
2224 #if defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
2225 audioproc::Test::EchoMetrics* message = test->mutable_echo_metrics();
2226 WriteStatsMessage(echo_metrics.residual_echo_return_loss,
2227 message->mutable_residual_echo_return_loss());
2228 WriteStatsMessage(echo_metrics.echo_return_loss,
2229 message->mutable_echo_return_loss());
2230 WriteStatsMessage(echo_metrics.echo_return_loss_enhancement,
2231 message->mutable_echo_return_loss_enhancement());
2232 WriteStatsMessage(echo_metrics.a_nlp,
2233 message->mutable_a_nlp());
2234
2235 audioproc::Test::DelayMetrics* message_delay =
2236 test->mutable_delay_metrics();
2237 message_delay->set_median(median);
2238 message_delay->set_std(std);
2239 message_delay->set_fraction_poor_delays(fraction_poor_delays);
2240
2241 test->set_rms_level(rms_level);
2242
2243 EXPECT_LE(0.0f, ns_speech_prob_average);
2244 EXPECT_GE(1.0f, ns_speech_prob_average);
2245 test->set_ns_speech_probability_average(ns_speech_prob_average);
2246 #endif
2247 }
2248
2249 rewind(far_file_);
2250 rewind(near_file_);
2251 }
2252
2253 if (write_ref_data) {
2254 OpenFileAndWriteMessage(ref_filename_, ref_data);
2255 }
2256 }
2257
TEST_F(ApmTest,NoErrorsWithKeyboardChannel)2258 TEST_F(ApmTest, NoErrorsWithKeyboardChannel) {
2259 struct ChannelFormat {
2260 AudioProcessing::ChannelLayout in_layout;
2261 AudioProcessing::ChannelLayout out_layout;
2262 };
2263 ChannelFormat cf[] = {
2264 {AudioProcessing::kMonoAndKeyboard, AudioProcessing::kMono},
2265 {AudioProcessing::kStereoAndKeyboard, AudioProcessing::kMono},
2266 {AudioProcessing::kStereoAndKeyboard, AudioProcessing::kStereo},
2267 };
2268
2269 rtc::scoped_ptr<AudioProcessing> ap(AudioProcessing::Create());
2270 // Enable one component just to ensure some processing takes place.
2271 ap->noise_suppression()->Enable(true);
2272 for (size_t i = 0; i < arraysize(cf); ++i) {
2273 const int in_rate = 44100;
2274 const int out_rate = 48000;
2275 ChannelBuffer<float> in_cb(SamplesFromRate(in_rate),
2276 TotalChannelsFromLayout(cf[i].in_layout));
2277 ChannelBuffer<float> out_cb(SamplesFromRate(out_rate),
2278 ChannelsFromLayout(cf[i].out_layout));
2279
2280 // Run over a few chunks.
2281 for (int j = 0; j < 10; ++j) {
2282 EXPECT_NOERR(ap->ProcessStream(
2283 in_cb.channels(),
2284 in_cb.num_frames(),
2285 in_rate,
2286 cf[i].in_layout,
2287 out_rate,
2288 cf[i].out_layout,
2289 out_cb.channels()));
2290 }
2291 }
2292 }
2293
2294 // Compares the reference and test arrays over a region around the expected
2295 // delay. Finds the highest SNR in that region and adds the variance and squared
2296 // error results to the supplied accumulators.
UpdateBestSNR(const float * ref,const float * test,size_t length,int expected_delay,double * variance_acc,double * sq_error_acc)2297 void UpdateBestSNR(const float* ref,
2298 const float* test,
2299 size_t length,
2300 int expected_delay,
2301 double* variance_acc,
2302 double* sq_error_acc) {
2303 double best_snr = std::numeric_limits<double>::min();
2304 double best_variance = 0;
2305 double best_sq_error = 0;
2306 // Search over a region of eight samples around the expected delay.
2307 for (int delay = std::max(expected_delay - 4, 0); delay <= expected_delay + 4;
2308 ++delay) {
2309 double sq_error = 0;
2310 double variance = 0;
2311 for (size_t i = 0; i < length - delay; ++i) {
2312 double error = test[i + delay] - ref[i];
2313 sq_error += error * error;
2314 variance += ref[i] * ref[i];
2315 }
2316
2317 if (sq_error == 0) {
2318 *variance_acc += variance;
2319 return;
2320 }
2321 double snr = variance / sq_error;
2322 if (snr > best_snr) {
2323 best_snr = snr;
2324 best_variance = variance;
2325 best_sq_error = sq_error;
2326 }
2327 }
2328
2329 *variance_acc += best_variance;
2330 *sq_error_acc += best_sq_error;
2331 }
2332
2333 // Used to test a multitude of sample rate and channel combinations. It works
2334 // by first producing a set of reference files (in SetUpTestCase) that are
2335 // assumed to be correct, as the used parameters are verified by other tests
2336 // in this collection. Primarily the reference files are all produced at
2337 // "native" rates which do not involve any resampling.
2338
2339 // Each test pass produces an output file with a particular format. The output
2340 // is matched against the reference file closest to its internal processing
2341 // format. If necessary the output is resampled back to its process format.
2342 // Due to the resampling distortion, we don't expect identical results, but
2343 // enforce SNR thresholds which vary depending on the format. 0 is a special
2344 // case SNR which corresponds to inf, or zero error.
2345 typedef std::tr1::tuple<int, int, int, int, double, double>
2346 AudioProcessingTestData;
2347 class AudioProcessingTest
2348 : public testing::TestWithParam<AudioProcessingTestData> {
2349 public:
AudioProcessingTest()2350 AudioProcessingTest()
2351 : input_rate_(std::tr1::get<0>(GetParam())),
2352 output_rate_(std::tr1::get<1>(GetParam())),
2353 reverse_input_rate_(std::tr1::get<2>(GetParam())),
2354 reverse_output_rate_(std::tr1::get<3>(GetParam())),
2355 expected_snr_(std::tr1::get<4>(GetParam())),
2356 expected_reverse_snr_(std::tr1::get<5>(GetParam())) {}
2357
~AudioProcessingTest()2358 virtual ~AudioProcessingTest() {}
2359
SetUpTestCase()2360 static void SetUpTestCase() {
2361 // Create all needed output reference files.
2362 const int kNativeRates[] = {8000, 16000, 32000, 48000};
2363 const size_t kNumChannels[] = {1, 2};
2364 for (size_t i = 0; i < arraysize(kNativeRates); ++i) {
2365 for (size_t j = 0; j < arraysize(kNumChannels); ++j) {
2366 for (size_t k = 0; k < arraysize(kNumChannels); ++k) {
2367 // The reference files always have matching input and output channels.
2368 ProcessFormat(kNativeRates[i], kNativeRates[i], kNativeRates[i],
2369 kNativeRates[i], kNumChannels[j], kNumChannels[j],
2370 kNumChannels[k], kNumChannels[k], "ref");
2371 }
2372 }
2373 }
2374 }
2375
TearDownTestCase()2376 static void TearDownTestCase() {
2377 ClearTempFiles();
2378 }
2379
2380 // Runs a process pass on files with the given parameters and dumps the output
2381 // to a file specified with |output_file_prefix|. Both forward and reverse
2382 // output streams are dumped.
ProcessFormat(int input_rate,int output_rate,int reverse_input_rate,int reverse_output_rate,size_t num_input_channels,size_t num_output_channels,size_t num_reverse_input_channels,size_t num_reverse_output_channels,std::string output_file_prefix)2383 static void ProcessFormat(int input_rate,
2384 int output_rate,
2385 int reverse_input_rate,
2386 int reverse_output_rate,
2387 size_t num_input_channels,
2388 size_t num_output_channels,
2389 size_t num_reverse_input_channels,
2390 size_t num_reverse_output_channels,
2391 std::string output_file_prefix) {
2392 Config config;
2393 config.Set<ExperimentalAgc>(new ExperimentalAgc(false));
2394 rtc::scoped_ptr<AudioProcessing> ap(AudioProcessing::Create(config));
2395 EnableAllAPComponents(ap.get());
2396
2397 ProcessingConfig processing_config = {
2398 {{input_rate, num_input_channels},
2399 {output_rate, num_output_channels},
2400 {reverse_input_rate, num_reverse_input_channels},
2401 {reverse_output_rate, num_reverse_output_channels}}};
2402 ap->Initialize(processing_config);
2403
2404 FILE* far_file =
2405 fopen(ResourceFilePath("far", reverse_input_rate).c_str(), "rb");
2406 FILE* near_file = fopen(ResourceFilePath("near", input_rate).c_str(), "rb");
2407 FILE* out_file =
2408 fopen(OutputFilePath(output_file_prefix, input_rate, output_rate,
2409 reverse_input_rate, reverse_output_rate,
2410 num_input_channels, num_output_channels,
2411 num_reverse_input_channels,
2412 num_reverse_output_channels, kForward).c_str(),
2413 "wb");
2414 FILE* rev_out_file =
2415 fopen(OutputFilePath(output_file_prefix, input_rate, output_rate,
2416 reverse_input_rate, reverse_output_rate,
2417 num_input_channels, num_output_channels,
2418 num_reverse_input_channels,
2419 num_reverse_output_channels, kReverse).c_str(),
2420 "wb");
2421 ASSERT_TRUE(far_file != NULL);
2422 ASSERT_TRUE(near_file != NULL);
2423 ASSERT_TRUE(out_file != NULL);
2424 ASSERT_TRUE(rev_out_file != NULL);
2425
2426 ChannelBuffer<float> fwd_cb(SamplesFromRate(input_rate),
2427 num_input_channels);
2428 ChannelBuffer<float> rev_cb(SamplesFromRate(reverse_input_rate),
2429 num_reverse_input_channels);
2430 ChannelBuffer<float> out_cb(SamplesFromRate(output_rate),
2431 num_output_channels);
2432 ChannelBuffer<float> rev_out_cb(SamplesFromRate(reverse_output_rate),
2433 num_reverse_output_channels);
2434
2435 // Temporary buffers.
2436 const int max_length =
2437 2 * std::max(std::max(out_cb.num_frames(), rev_out_cb.num_frames()),
2438 std::max(fwd_cb.num_frames(), rev_cb.num_frames()));
2439 rtc::scoped_ptr<float[]> float_data(new float[max_length]);
2440 rtc::scoped_ptr<int16_t[]> int_data(new int16_t[max_length]);
2441
2442 int analog_level = 127;
2443 while (ReadChunk(far_file, int_data.get(), float_data.get(), &rev_cb) &&
2444 ReadChunk(near_file, int_data.get(), float_data.get(), &fwd_cb)) {
2445 EXPECT_NOERR(ap->ProcessReverseStream(
2446 rev_cb.channels(), processing_config.reverse_input_stream(),
2447 processing_config.reverse_output_stream(), rev_out_cb.channels()));
2448
2449 EXPECT_NOERR(ap->set_stream_delay_ms(0));
2450 ap->echo_cancellation()->set_stream_drift_samples(0);
2451 EXPECT_NOERR(ap->gain_control()->set_stream_analog_level(analog_level));
2452
2453 EXPECT_NOERR(ap->ProcessStream(
2454 fwd_cb.channels(),
2455 fwd_cb.num_frames(),
2456 input_rate,
2457 LayoutFromChannels(num_input_channels),
2458 output_rate,
2459 LayoutFromChannels(num_output_channels),
2460 out_cb.channels()));
2461
2462 // Dump forward output to file.
2463 Interleave(out_cb.channels(), out_cb.num_frames(), out_cb.num_channels(),
2464 float_data.get());
2465 size_t out_length = out_cb.num_channels() * out_cb.num_frames();
2466
2467 ASSERT_EQ(out_length,
2468 fwrite(float_data.get(), sizeof(float_data[0]),
2469 out_length, out_file));
2470
2471 // Dump reverse output to file.
2472 Interleave(rev_out_cb.channels(), rev_out_cb.num_frames(),
2473 rev_out_cb.num_channels(), float_data.get());
2474 size_t rev_out_length =
2475 rev_out_cb.num_channels() * rev_out_cb.num_frames();
2476
2477 ASSERT_EQ(rev_out_length,
2478 fwrite(float_data.get(), sizeof(float_data[0]), rev_out_length,
2479 rev_out_file));
2480
2481 analog_level = ap->gain_control()->stream_analog_level();
2482 }
2483 fclose(far_file);
2484 fclose(near_file);
2485 fclose(out_file);
2486 fclose(rev_out_file);
2487 }
2488
2489 protected:
2490 int input_rate_;
2491 int output_rate_;
2492 int reverse_input_rate_;
2493 int reverse_output_rate_;
2494 double expected_snr_;
2495 double expected_reverse_snr_;
2496 };
2497
TEST_P(AudioProcessingTest,Formats)2498 TEST_P(AudioProcessingTest, Formats) {
2499 struct ChannelFormat {
2500 int num_input;
2501 int num_output;
2502 int num_reverse_input;
2503 int num_reverse_output;
2504 };
2505 ChannelFormat cf[] = {
2506 {1, 1, 1, 1},
2507 {1, 1, 2, 1},
2508 {2, 1, 1, 1},
2509 {2, 1, 2, 1},
2510 {2, 2, 1, 1},
2511 {2, 2, 2, 2},
2512 };
2513
2514 for (size_t i = 0; i < arraysize(cf); ++i) {
2515 ProcessFormat(input_rate_, output_rate_, reverse_input_rate_,
2516 reverse_output_rate_, cf[i].num_input, cf[i].num_output,
2517 cf[i].num_reverse_input, cf[i].num_reverse_output, "out");
2518
2519 // Verify output for both directions.
2520 std::vector<StreamDirection> stream_directions;
2521 stream_directions.push_back(kForward);
2522 stream_directions.push_back(kReverse);
2523 for (StreamDirection file_direction : stream_directions) {
2524 const int in_rate = file_direction ? reverse_input_rate_ : input_rate_;
2525 const int out_rate = file_direction ? reverse_output_rate_ : output_rate_;
2526 const int out_num =
2527 file_direction ? cf[i].num_reverse_output : cf[i].num_output;
2528 const double expected_snr =
2529 file_direction ? expected_reverse_snr_ : expected_snr_;
2530
2531 const int min_ref_rate = std::min(in_rate, out_rate);
2532 int ref_rate;
2533
2534 if (min_ref_rate > 32000) {
2535 ref_rate = 48000;
2536 } else if (min_ref_rate > 16000) {
2537 ref_rate = 32000;
2538 } else if (min_ref_rate > 8000) {
2539 ref_rate = 16000;
2540 } else {
2541 ref_rate = 8000;
2542 }
2543 #ifdef WEBRTC_AUDIOPROC_FIXED_PROFILE
2544 if (file_direction == kForward) {
2545 ref_rate = std::min(ref_rate, 16000);
2546 }
2547 #endif
2548 FILE* out_file = fopen(
2549 OutputFilePath("out", input_rate_, output_rate_, reverse_input_rate_,
2550 reverse_output_rate_, cf[i].num_input,
2551 cf[i].num_output, cf[i].num_reverse_input,
2552 cf[i].num_reverse_output, file_direction).c_str(),
2553 "rb");
2554 // The reference files always have matching input and output channels.
2555 FILE* ref_file = fopen(
2556 OutputFilePath("ref", ref_rate, ref_rate, ref_rate, ref_rate,
2557 cf[i].num_output, cf[i].num_output,
2558 cf[i].num_reverse_output, cf[i].num_reverse_output,
2559 file_direction).c_str(),
2560 "rb");
2561 ASSERT_TRUE(out_file != NULL);
2562 ASSERT_TRUE(ref_file != NULL);
2563
2564 const size_t ref_length = SamplesFromRate(ref_rate) * out_num;
2565 const size_t out_length = SamplesFromRate(out_rate) * out_num;
2566 // Data from the reference file.
2567 rtc::scoped_ptr<float[]> ref_data(new float[ref_length]);
2568 // Data from the output file.
2569 rtc::scoped_ptr<float[]> out_data(new float[out_length]);
2570 // Data from the resampled output, in case the reference and output rates
2571 // don't match.
2572 rtc::scoped_ptr<float[]> cmp_data(new float[ref_length]);
2573
2574 PushResampler<float> resampler;
2575 resampler.InitializeIfNeeded(out_rate, ref_rate, out_num);
2576
2577 // Compute the resampling delay of the output relative to the reference,
2578 // to find the region over which we should search for the best SNR.
2579 float expected_delay_sec = 0;
2580 if (in_rate != ref_rate) {
2581 // Input resampling delay.
2582 expected_delay_sec +=
2583 PushSincResampler::AlgorithmicDelaySeconds(in_rate);
2584 }
2585 if (out_rate != ref_rate) {
2586 // Output resampling delay.
2587 expected_delay_sec +=
2588 PushSincResampler::AlgorithmicDelaySeconds(ref_rate);
2589 // Delay of converting the output back to its processing rate for
2590 // testing.
2591 expected_delay_sec +=
2592 PushSincResampler::AlgorithmicDelaySeconds(out_rate);
2593 }
2594 int expected_delay =
2595 floor(expected_delay_sec * ref_rate + 0.5f) * out_num;
2596
2597 double variance = 0;
2598 double sq_error = 0;
2599 while (fread(out_data.get(), sizeof(out_data[0]), out_length, out_file) &&
2600 fread(ref_data.get(), sizeof(ref_data[0]), ref_length, ref_file)) {
2601 float* out_ptr = out_data.get();
2602 if (out_rate != ref_rate) {
2603 // Resample the output back to its internal processing rate if
2604 // necssary.
2605 ASSERT_EQ(ref_length,
2606 static_cast<size_t>(resampler.Resample(
2607 out_ptr, out_length, cmp_data.get(), ref_length)));
2608 out_ptr = cmp_data.get();
2609 }
2610
2611 // Update the |sq_error| and |variance| accumulators with the highest
2612 // SNR of reference vs output.
2613 UpdateBestSNR(ref_data.get(), out_ptr, ref_length, expected_delay,
2614 &variance, &sq_error);
2615 }
2616
2617 std::cout << "(" << input_rate_ << ", " << output_rate_ << ", "
2618 << reverse_input_rate_ << ", " << reverse_output_rate_ << ", "
2619 << cf[i].num_input << ", " << cf[i].num_output << ", "
2620 << cf[i].num_reverse_input << ", " << cf[i].num_reverse_output
2621 << ", " << file_direction << "): ";
2622 if (sq_error > 0) {
2623 double snr = 10 * log10(variance / sq_error);
2624 EXPECT_GE(snr, expected_snr);
2625 EXPECT_NE(0, expected_snr);
2626 std::cout << "SNR=" << snr << " dB" << std::endl;
2627 } else {
2628 EXPECT_EQ(expected_snr, 0);
2629 std::cout << "SNR="
2630 << "inf dB" << std::endl;
2631 }
2632
2633 fclose(out_file);
2634 fclose(ref_file);
2635 }
2636 }
2637 }
2638
2639 #if defined(WEBRTC_AUDIOPROC_FLOAT_PROFILE)
2640 INSTANTIATE_TEST_CASE_P(
2641 CommonFormats,
2642 AudioProcessingTest,
2643 testing::Values(std::tr1::make_tuple(48000, 48000, 48000, 48000, 0, 0),
2644 std::tr1::make_tuple(48000, 48000, 32000, 48000, 40, 30),
2645 std::tr1::make_tuple(48000, 48000, 16000, 48000, 40, 20),
2646 std::tr1::make_tuple(48000, 44100, 48000, 44100, 20, 20),
2647 std::tr1::make_tuple(48000, 44100, 32000, 44100, 20, 15),
2648 std::tr1::make_tuple(48000, 44100, 16000, 44100, 20, 15),
2649 std::tr1::make_tuple(48000, 32000, 48000, 32000, 30, 35),
2650 std::tr1::make_tuple(48000, 32000, 32000, 32000, 30, 0),
2651 std::tr1::make_tuple(48000, 32000, 16000, 32000, 30, 20),
2652 std::tr1::make_tuple(48000, 16000, 48000, 16000, 25, 20),
2653 std::tr1::make_tuple(48000, 16000, 32000, 16000, 25, 20),
2654 std::tr1::make_tuple(48000, 16000, 16000, 16000, 25, 0),
2655
2656 std::tr1::make_tuple(44100, 48000, 48000, 48000, 30, 0),
2657 std::tr1::make_tuple(44100, 48000, 32000, 48000, 30, 30),
2658 std::tr1::make_tuple(44100, 48000, 16000, 48000, 30, 20),
2659 std::tr1::make_tuple(44100, 44100, 48000, 44100, 20, 20),
2660 std::tr1::make_tuple(44100, 44100, 32000, 44100, 20, 15),
2661 std::tr1::make_tuple(44100, 44100, 16000, 44100, 20, 15),
2662 std::tr1::make_tuple(44100, 32000, 48000, 32000, 30, 35),
2663 std::tr1::make_tuple(44100, 32000, 32000, 32000, 30, 0),
2664 std::tr1::make_tuple(44100, 32000, 16000, 32000, 30, 20),
2665 std::tr1::make_tuple(44100, 16000, 48000, 16000, 25, 20),
2666 std::tr1::make_tuple(44100, 16000, 32000, 16000, 25, 20),
2667 std::tr1::make_tuple(44100, 16000, 16000, 16000, 25, 0),
2668
2669 std::tr1::make_tuple(32000, 48000, 48000, 48000, 30, 0),
2670 std::tr1::make_tuple(32000, 48000, 32000, 48000, 35, 30),
2671 std::tr1::make_tuple(32000, 48000, 16000, 48000, 30, 20),
2672 std::tr1::make_tuple(32000, 44100, 48000, 44100, 20, 20),
2673 std::tr1::make_tuple(32000, 44100, 32000, 44100, 20, 15),
2674 std::tr1::make_tuple(32000, 44100, 16000, 44100, 20, 15),
2675 std::tr1::make_tuple(32000, 32000, 48000, 32000, 40, 35),
2676 std::tr1::make_tuple(32000, 32000, 32000, 32000, 0, 0),
2677 std::tr1::make_tuple(32000, 32000, 16000, 32000, 40, 20),
2678 std::tr1::make_tuple(32000, 16000, 48000, 16000, 25, 20),
2679 std::tr1::make_tuple(32000, 16000, 32000, 16000, 25, 20),
2680 std::tr1::make_tuple(32000, 16000, 16000, 16000, 25, 0),
2681
2682 std::tr1::make_tuple(16000, 48000, 48000, 48000, 25, 0),
2683 std::tr1::make_tuple(16000, 48000, 32000, 48000, 25, 30),
2684 std::tr1::make_tuple(16000, 48000, 16000, 48000, 25, 20),
2685 std::tr1::make_tuple(16000, 44100, 48000, 44100, 15, 20),
2686 std::tr1::make_tuple(16000, 44100, 32000, 44100, 15, 15),
2687 std::tr1::make_tuple(16000, 44100, 16000, 44100, 15, 15),
2688 std::tr1::make_tuple(16000, 32000, 48000, 32000, 25, 35),
2689 std::tr1::make_tuple(16000, 32000, 32000, 32000, 25, 0),
2690 std::tr1::make_tuple(16000, 32000, 16000, 32000, 25, 20),
2691 std::tr1::make_tuple(16000, 16000, 48000, 16000, 40, 20),
2692 std::tr1::make_tuple(16000, 16000, 32000, 16000, 50, 20),
2693 std::tr1::make_tuple(16000, 16000, 16000, 16000, 0, 0)));
2694
2695 #elif defined(WEBRTC_AUDIOPROC_FIXED_PROFILE)
2696 INSTANTIATE_TEST_CASE_P(
2697 CommonFormats,
2698 AudioProcessingTest,
2699 testing::Values(std::tr1::make_tuple(48000, 48000, 48000, 48000, 20, 0),
2700 std::tr1::make_tuple(48000, 48000, 32000, 48000, 20, 30),
2701 std::tr1::make_tuple(48000, 48000, 16000, 48000, 20, 20),
2702 std::tr1::make_tuple(48000, 44100, 48000, 44100, 15, 20),
2703 std::tr1::make_tuple(48000, 44100, 32000, 44100, 15, 15),
2704 std::tr1::make_tuple(48000, 44100, 16000, 44100, 15, 15),
2705 std::tr1::make_tuple(48000, 32000, 48000, 32000, 20, 35),
2706 std::tr1::make_tuple(48000, 32000, 32000, 32000, 20, 0),
2707 std::tr1::make_tuple(48000, 32000, 16000, 32000, 20, 20),
2708 std::tr1::make_tuple(48000, 16000, 48000, 16000, 20, 20),
2709 std::tr1::make_tuple(48000, 16000, 32000, 16000, 20, 20),
2710 std::tr1::make_tuple(48000, 16000, 16000, 16000, 20, 0),
2711
2712 std::tr1::make_tuple(44100, 48000, 48000, 48000, 20, 0),
2713 std::tr1::make_tuple(44100, 48000, 32000, 48000, 20, 30),
2714 std::tr1::make_tuple(44100, 48000, 16000, 48000, 20, 20),
2715 std::tr1::make_tuple(44100, 44100, 48000, 44100, 15, 20),
2716 std::tr1::make_tuple(44100, 44100, 32000, 44100, 15, 15),
2717 std::tr1::make_tuple(44100, 44100, 16000, 44100, 15, 15),
2718 std::tr1::make_tuple(44100, 32000, 48000, 32000, 20, 35),
2719 std::tr1::make_tuple(44100, 32000, 32000, 32000, 20, 0),
2720 std::tr1::make_tuple(44100, 32000, 16000, 32000, 20, 20),
2721 std::tr1::make_tuple(44100, 16000, 48000, 16000, 20, 20),
2722 std::tr1::make_tuple(44100, 16000, 32000, 16000, 20, 20),
2723 std::tr1::make_tuple(44100, 16000, 16000, 16000, 20, 0),
2724
2725 std::tr1::make_tuple(32000, 48000, 48000, 48000, 20, 0),
2726 std::tr1::make_tuple(32000, 48000, 32000, 48000, 20, 30),
2727 std::tr1::make_tuple(32000, 48000, 16000, 48000, 20, 20),
2728 std::tr1::make_tuple(32000, 44100, 48000, 44100, 15, 20),
2729 std::tr1::make_tuple(32000, 44100, 32000, 44100, 15, 15),
2730 std::tr1::make_tuple(32000, 44100, 16000, 44100, 15, 15),
2731 std::tr1::make_tuple(32000, 32000, 48000, 32000, 20, 35),
2732 std::tr1::make_tuple(32000, 32000, 32000, 32000, 20, 0),
2733 std::tr1::make_tuple(32000, 32000, 16000, 32000, 20, 20),
2734 std::tr1::make_tuple(32000, 16000, 48000, 16000, 20, 20),
2735 std::tr1::make_tuple(32000, 16000, 32000, 16000, 20, 20),
2736 std::tr1::make_tuple(32000, 16000, 16000, 16000, 20, 0),
2737
2738 std::tr1::make_tuple(16000, 48000, 48000, 48000, 25, 0),
2739 std::tr1::make_tuple(16000, 48000, 32000, 48000, 25, 30),
2740 std::tr1::make_tuple(16000, 48000, 16000, 48000, 25, 20),
2741 std::tr1::make_tuple(16000, 44100, 48000, 44100, 15, 20),
2742 std::tr1::make_tuple(16000, 44100, 32000, 44100, 15, 15),
2743 std::tr1::make_tuple(16000, 44100, 16000, 44100, 15, 15),
2744 std::tr1::make_tuple(16000, 32000, 48000, 32000, 25, 35),
2745 std::tr1::make_tuple(16000, 32000, 32000, 32000, 25, 0),
2746 std::tr1::make_tuple(16000, 32000, 16000, 32000, 25, 20),
2747 std::tr1::make_tuple(16000, 16000, 48000, 16000, 35, 20),
2748 std::tr1::make_tuple(16000, 16000, 32000, 16000, 40, 20),
2749 std::tr1::make_tuple(16000, 16000, 16000, 16000, 0, 0)));
2750 #endif
2751
2752 } // namespace
2753 } // namespace webrtc
2754