cc: Make picture pile base thread safe.
[chromium-blink-merge.git] / content / renderer / media / speech_recognition_audio_sink_unittest.cc
blob4366309f1ed7ac4d52cc5fd80ad28aaea08add27
1 // Copyright 2014 The Chromium Authors. All rights reserved.
2 // Use of this source code is governed by a BSD-style license that can be
3 // found in the LICENSE file.
5 #include "content/renderer/media/speech_recognition_audio_sink.h"
7 #include "base/bind.h"
8 #include "base/strings/utf_string_conversions.h"
9 #include "content/renderer/media/media_stream_audio_source.h"
10 #include "content/renderer/media/mock_media_constraint_factory.h"
11 #include "content/renderer/media/webrtc/webrtc_local_audio_track_adapter.h"
12 #include "content/renderer/media/webrtc_local_audio_track.h"
13 #include "media/audio/audio_parameters.h"
14 #include "media/base/audio_bus.h"
15 #include "testing/gmock/include/gmock/gmock.h"
16 #include "testing/gtest/include/gtest/gtest.h"
17 #include "third_party/WebKit/public/platform/WebMediaStreamTrack.h"
19 namespace {
21 // Supported speech recognition audio parameters.
22 const int kSpeechRecognitionSampleRate = 16000;
23 const int kSpeechRecognitionFramesPerBuffer = 1600;
25 // Input audio format.
26 const media::AudioParameters::Format kInputFormat =
27 media::AudioParameters::AUDIO_PCM_LOW_LATENCY;
28 const media::ChannelLayout kInputChannelLayout = media::CHANNEL_LAYOUT_MONO;
29 const int kInputChannels = 1;
30 const int kInputBitsPerSample = 16;
32 // Output audio format.
33 const media::AudioParameters::Format kOutputFormat =
34 media::AudioParameters::AUDIO_PCM_LOW_LATENCY;
35 const media::ChannelLayout kOutputChannelLayout = media::CHANNEL_LAYOUT_STEREO;
36 const int kOutputChannels = 2;
37 const int kOutputBitsPerSample = 16;
39 // Mocked out sockets used for Send/Receive.
40 // Data is written and read from a shared buffer used as a FIFO and there is
41 // no blocking. |OnSendCB| is used to trigger a |Receive| on the other socket.
42 class MockSyncSocket : public base::SyncSocket {
43 public:
44 // This allows for 2 requests in queue between the |MockSyncSocket|s.
45 static const int kSharedBufferSize = 8;
47 // Buffer to be shared between two |MockSyncSocket|s. Allocated on heap.
48 struct SharedBuffer {
49 SharedBuffer() : data(), start(0), length(0) {}
51 uint8 data[kSharedBufferSize];
52 size_t start;
53 size_t length;
56 // Callback used for pairing an A.Send() with B.Receieve() without blocking.
57 typedef base::Callback<void()> OnSendCB;
59 explicit MockSyncSocket(SharedBuffer* shared_buffer)
60 : buffer_(shared_buffer),
61 in_failure_mode_(false) {}
63 MockSyncSocket(SharedBuffer* shared_buffer, const OnSendCB& on_send_cb)
64 : buffer_(shared_buffer),
65 on_send_cb_(on_send_cb),
66 in_failure_mode_(false) {}
68 size_t Send(const void* buffer, size_t length) override;
69 size_t Receive(void* buffer, size_t length) override;
71 // When |in_failure_mode_| == true, the socket fails to send.
72 void SetFailureMode(bool in_failure_mode) {
73 in_failure_mode_ = in_failure_mode;
76 private:
77 SharedBuffer* buffer_;
78 const OnSendCB on_send_cb_;
79 bool in_failure_mode_;
81 DISALLOW_COPY_AND_ASSIGN(MockSyncSocket);
84 // base::SyncSocket implementation
85 size_t MockSyncSocket::Send(const void* buffer, size_t length) {
86 if (in_failure_mode_)
87 return 0;
89 const uint8* b = static_cast<const uint8*>(buffer);
90 for (size_t i = 0; i < length; ++i, ++buffer_->length)
91 buffer_->data[buffer_->start + buffer_->length] = b[i];
93 on_send_cb_.Run();
94 return length;
97 size_t MockSyncSocket::Receive(void* buffer, size_t length) {
98 uint8* b = static_cast<uint8*>(buffer);
99 for (size_t i = buffer_->start; i < buffer_->length; ++i, ++buffer_->start)
100 b[i] = buffer_->data[buffer_->start];
102 // Since buffer is used sequentially, we can reset the buffer indices here.
103 buffer_->start = buffer_->length = 0;
104 return length;
107 // This fake class is the consumer used to verify behaviour of the producer.
108 // The |Initialize()| method shows what the consumer should be responsible for
109 // in the production code (minus the mocks).
110 class FakeSpeechRecognizer {
111 public:
112 FakeSpeechRecognizer() : is_responsive_(true) {}
114 void Initialize(
115 const blink::WebMediaStreamTrack& track,
116 const media::AudioParameters& sink_params,
117 base::SharedMemoryHandle* foreign_memory_handle) {
118 // Shared memory is allocated, mapped and shared.
119 const uint32 kSharedMemorySize =
120 sizeof(media::AudioInputBufferParameters) +
121 media::AudioBus::CalculateMemorySize(sink_params);
122 shared_memory_.reset(new base::SharedMemory());
123 ASSERT_TRUE(shared_memory_->CreateAndMapAnonymous(kSharedMemorySize));
124 memset(shared_memory_->memory(), 0, kSharedMemorySize);
125 ASSERT_TRUE(shared_memory_->ShareToProcess(base::GetCurrentProcessHandle(),
126 foreign_memory_handle));
128 // Wrap the shared memory for the audio bus.
129 media::AudioInputBuffer* buffer =
130 static_cast<media::AudioInputBuffer*>(shared_memory_->memory());
132 audio_track_bus_ = media::AudioBus::WrapMemory(sink_params, buffer->audio);
133 audio_track_bus_->Zero();
135 // Reference to the counter used to synchronize.
136 buffer->params.size = 0U;
138 // Create a shared buffer for the |MockSyncSocket|s.
139 shared_buffer_.reset(new MockSyncSocket::SharedBuffer());
141 // Local socket will receive signals from the producer.
142 receiving_socket_.reset(new MockSyncSocket(shared_buffer_.get()));
144 // We automatically trigger a Receive when data is sent over the socket.
145 sending_socket_ = new MockSyncSocket(
146 shared_buffer_.get(),
147 base::Bind(&FakeSpeechRecognizer::EmulateReceiveThreadLoopIteration,
148 base::Unretained(this)));
150 // This is usually done to pair the sockets. Here it's not effective.
151 base::SyncSocket::CreatePair(receiving_socket_.get(), sending_socket_);
154 // Emulates a single iteraton of a thread receiving on the socket.
155 // This would normally be done on a receiving thread's task on the browser.
156 void EmulateReceiveThreadLoopIteration() {
157 if (!is_responsive_)
158 return;
160 const int kSize = sizeof(media::AudioInputBufferParameters().size);
161 receiving_socket_->Receive(&(GetAudioInputBuffer()->params.size), kSize);
163 // Notify the producer that the audio buffer has been consumed.
164 GetAudioInputBuffer()->params.size++;
167 // Used to simulate an unresponsive behaviour of the consumer.
168 void SimulateResponsiveness(bool is_responsive) {
169 is_responsive_ = is_responsive;
172 media::AudioInputBuffer * GetAudioInputBuffer() const {
173 return static_cast<media::AudioInputBuffer*>(shared_memory_->memory());
176 MockSyncSocket* sending_socket() { return sending_socket_; }
177 media::AudioBus* audio_bus() const { return audio_track_bus_.get(); }
180 private:
181 bool is_responsive_;
183 // Shared memory for the audio and synchronization.
184 scoped_ptr<base::SharedMemory> shared_memory_;
186 // Fake sockets and their shared buffer.
187 scoped_ptr<MockSyncSocket::SharedBuffer> shared_buffer_;
188 scoped_ptr<MockSyncSocket> receiving_socket_;
189 MockSyncSocket* sending_socket_;
191 // Audio bus wrapping the shared memory from the renderer.
192 scoped_ptr<media::AudioBus> audio_track_bus_;
194 DISALLOW_COPY_AND_ASSIGN(FakeSpeechRecognizer);
197 } // namespace
199 namespace content {
201 class SpeechRecognitionAudioSinkTest : public testing::Test {
202 public:
203 SpeechRecognitionAudioSinkTest() {}
205 ~SpeechRecognitionAudioSinkTest() {}
207 // Initializes the producer and consumer with specified audio parameters.
208 // Returns the minimal number of input audio buffers which need to be captured
209 // before they get sent to the consumer.
210 uint32 Initialize(int input_sample_rate,
211 int input_frames_per_buffer,
212 int output_sample_rate,
213 int output_frames_per_buffer) {
214 // Audio Environment setup.
215 source_params_.Reset(kInputFormat,
216 kInputChannelLayout,
217 kInputChannels,
218 input_sample_rate,
219 kInputBitsPerSample,
220 input_frames_per_buffer);
221 sink_params_.Reset(kOutputFormat,
222 kOutputChannelLayout,
223 kOutputChannels,
224 output_sample_rate,
225 kOutputBitsPerSample,
226 output_frames_per_buffer);
227 source_data_.reset(new int16[input_frames_per_buffer * kInputChannels]{});
229 // Prepare the track and audio source.
230 blink::WebMediaStreamTrack blink_track;
231 PrepareBlinkTrackOfType(MEDIA_DEVICE_AUDIO_CAPTURE, &blink_track);
233 // Get the native track from the blink track and initialize.
234 native_track_ =
235 static_cast<WebRtcLocalAudioTrack*>(blink_track.extraData());
236 native_track_->OnSetFormat(source_params_);
238 // Create and initialize the consumer.
239 recognizer_.reset(new FakeSpeechRecognizer());
240 base::SharedMemoryHandle foreign_memory_handle;
241 recognizer_->Initialize(blink_track, sink_params_, &foreign_memory_handle);
243 // Create the producer.
244 scoped_ptr<base::SyncSocket> sending_socket(recognizer_->sending_socket());
245 speech_audio_sink_.reset(new SpeechRecognitionAudioSink(
246 blink_track, sink_params_, foreign_memory_handle,
247 sending_socket.Pass(),
248 base::Bind(&SpeechRecognitionAudioSinkTest::StoppedCallback,
249 base::Unretained(this))));
251 // Return number of buffers needed to trigger resampling and consumption.
252 return static_cast<uint32>(std::ceil(
253 static_cast<double>(output_frames_per_buffer * input_sample_rate) /
254 (input_frames_per_buffer * output_sample_rate)));
257 // Mock callback expected to be called when the track is stopped.
258 MOCK_METHOD0(StoppedCallback, void());
260 protected:
261 // Prepares a blink track of a given MediaStreamType and attaches the native
262 // track which can be used to capture audio data and pass it to the producer.
263 static void PrepareBlinkTrackOfType(
264 const MediaStreamType device_type,
265 blink::WebMediaStreamTrack* blink_track) {
266 StreamDeviceInfo device_info(device_type, "Mock device",
267 "mock_device_id");
268 MockMediaConstraintFactory constraint_factory;
269 const blink::WebMediaConstraints constraints =
270 constraint_factory.CreateWebMediaConstraints();
271 scoped_refptr<WebRtcAudioCapturer> capturer(
272 WebRtcAudioCapturer::CreateCapturer(-1, device_info, constraints, NULL,
273 NULL));
274 scoped_refptr<WebRtcLocalAudioTrackAdapter> adapter(
275 WebRtcLocalAudioTrackAdapter::Create(std::string(), NULL));
276 scoped_ptr<WebRtcLocalAudioTrack> native_track(
277 new WebRtcLocalAudioTrack(adapter.get(), capturer, NULL));
278 blink::WebMediaStreamSource blink_audio_source;
279 blink_audio_source.initialize(base::UTF8ToUTF16("dummy_source_id"),
280 blink::WebMediaStreamSource::TypeAudio,
281 base::UTF8ToUTF16("dummy_source_name"));
282 MediaStreamSource::SourceStoppedCallback cb;
283 blink_audio_source.setExtraData(
284 new MediaStreamAudioSource(-1, device_info, cb, NULL));
285 blink_track->initialize(blink::WebString::fromUTF8("dummy_track"),
286 blink_audio_source);
287 blink_track->setExtraData(native_track.release());
290 // Emulates an audio capture device capturing data from the source.
291 inline void CaptureAudio(const uint32 buffers) {
292 for (uint32 i = 0; i < buffers; ++i)
293 native_track()->Capture(source_data(),
294 base::TimeDelta::FromMilliseconds(0), 1, false,
295 false, false);
298 // Used to simulate a problem with sockets.
299 void SetFailureModeOnForeignSocket(bool in_failure_mode) {
300 recognizer()->sending_socket()->SetFailureMode(in_failure_mode);
303 // Helper method for verifying captured audio data has been consumed.
304 inline void AssertConsumedBuffers(const uint32 buffer_index) {
305 ASSERT_EQ(buffer_index, recognizer()->GetAudioInputBuffer()->params.size);
308 // Helper method for providing audio data to producer and verifying it was
309 // consumed on the recognizer.
310 inline void CaptureAudioAndAssertConsumedBuffers(const uint32 buffers,
311 const uint32 buffer_index) {
312 CaptureAudio(buffers);
313 AssertConsumedBuffers(buffer_index);
316 // Helper method to capture and assert consumption at different sample rates
317 // and audio buffer sizes.
318 inline void AssertConsumptionForAudioParameters(
319 const int input_sample_rate,
320 const int input_frames_per_buffer,
321 const int output_sample_rate,
322 const int output_frames_per_buffer,
323 const uint32 consumptions) {
324 const uint32 kBuffersPerNotification = Initialize(input_sample_rate,
325 input_frames_per_buffer,
326 output_sample_rate,
327 output_frames_per_buffer);
328 AssertConsumedBuffers(0U);
330 for (uint32 i = 1U; i <= consumptions; ++i) {
331 CaptureAudio(kBuffersPerNotification);
332 ASSERT_EQ(i, recognizer()->GetAudioInputBuffer()->params.size)
333 << "Tested at rates: "
334 << "In(" << input_sample_rate << ", " << input_frames_per_buffer
335 << ") "
336 << "Out(" << output_sample_rate << ", " << output_frames_per_buffer
337 << ")";
341 int16* source_data() { return source_data_.get(); }
343 FakeSpeechRecognizer* recognizer() { return recognizer_.get(); }
345 const media::AudioParameters& sink_params() { return sink_params_; }
347 WebRtcLocalAudioTrack* native_track() { return native_track_; }
349 private:
350 // Producer.
351 scoped_ptr<SpeechRecognitionAudioSink> speech_audio_sink_;
353 // Consumer.
354 scoped_ptr<FakeSpeechRecognizer> recognizer_;
356 // Audio related members.
357 scoped_ptr<int16[]> source_data_;
358 media::AudioParameters source_params_;
359 media::AudioParameters sink_params_;
360 WebRtcLocalAudioTrack* native_track_;
362 DISALLOW_COPY_AND_ASSIGN(SpeechRecognitionAudioSinkTest);
365 // Not all types of tracks are supported. This test checks if that policy is
366 // implemented correctly.
367 TEST_F(SpeechRecognitionAudioSinkTest, CheckIsSupportedAudioTrack) {
368 typedef std::map<MediaStreamType, bool> SupportedTrackPolicy;
370 // This test must be aligned with the policy of supported tracks.
371 SupportedTrackPolicy p;
372 p[MEDIA_NO_SERVICE] = false;
373 p[MEDIA_DEVICE_AUDIO_CAPTURE] = true; // The only one supported for now.
374 p[MEDIA_DEVICE_VIDEO_CAPTURE] = false;
375 p[MEDIA_TAB_AUDIO_CAPTURE] = false;
376 p[MEDIA_TAB_VIDEO_CAPTURE] = false;
377 p[MEDIA_DESKTOP_VIDEO_CAPTURE] = false;
378 p[MEDIA_LOOPBACK_AUDIO_CAPTURE] = false;
379 p[MEDIA_DEVICE_AUDIO_OUTPUT] = false;
381 // Ensure this test gets updated along with |content::MediaStreamType| enum.
382 EXPECT_EQ(NUM_MEDIA_TYPES, p.size());
384 // Check the the entire policy.
385 for (SupportedTrackPolicy::iterator it = p.begin(); it != p.end(); ++it) {
386 blink::WebMediaStreamTrack blink_track;
387 PrepareBlinkTrackOfType(it->first, &blink_track);
388 ASSERT_EQ(
389 it->second,
390 SpeechRecognitionAudioSink::IsSupportedTrack(blink_track));
394 // Checks if the producer can support the listed range of input sample rates
395 // and associated buffer sizes.
396 TEST_F(SpeechRecognitionAudioSinkTest, RecognizerNotifiedOnSocket) {
397 const size_t kNumAudioParamTuples = 24;
398 const int kAudioParams[kNumAudioParamTuples][2] = {
399 {8000, 80}, {8000, 800}, {16000, 160}, {16000, 1600},
400 {24000, 240}, {24000, 2400}, {32000, 320}, {32000, 3200},
401 {44100, 441}, {44100, 4410}, {48000, 480}, {48000, 4800},
402 {96000, 960}, {96000, 9600}, {11025, 111}, {11025, 1103},
403 {22050, 221}, {22050, 2205}, {88200, 882}, {88200, 8820},
404 {176400, 1764}, {176400, 17640}, {192000, 1920}, {192000, 19200}};
406 // Check all listed tuples of input sample rates and buffers sizes.
407 for (size_t i = 0; i < kNumAudioParamTuples; ++i) {
408 AssertConsumptionForAudioParameters(
409 kAudioParams[i][0], kAudioParams[i][1],
410 kSpeechRecognitionSampleRate, kSpeechRecognitionFramesPerBuffer, 3U);
414 // Checks that the input data is getting resampled to the target sample rate.
415 TEST_F(SpeechRecognitionAudioSinkTest, AudioDataIsResampledOnSink) {
416 EXPECT_GE(kInputChannels, 1);
417 EXPECT_GE(kOutputChannels, 1);
419 // Input audio is sampled at 44.1 KHz with data chunks of 10ms. Desired output
420 // is corresponding to the speech recognition engine requirements: 16 KHz with
421 // 100 ms chunks (1600 frames per buffer).
422 const uint32 kBuffersPerNotification = Initialize(44100, 441, 16000, 1600);
423 // Fill audio input frames with 0, 1, 2, 3, ..., 440.
424 const uint32 kSourceDataLength = 441 * kInputChannels;
425 for (uint32 i = 0; i < kSourceDataLength; ++i) {
426 for (int c = 0; c < kInputChannels; ++c)
427 source_data()[i * kInputChannels + c] = i;
430 // Prepare sink audio bus and data for rendering.
431 media::AudioBus* sink_bus = recognizer()->audio_bus();
432 const uint32 kSinkDataLength = 1600 * kOutputChannels;
433 int16 sink_data[kSinkDataLength] = {0};
435 // Render the audio data from the recognizer.
436 sink_bus->ToInterleaved(sink_bus->frames(),
437 sink_params().bits_per_sample() / 8, sink_data);
439 // Checking only a fraction of the sink frames.
440 const uint32 kNumFramesToTest = 12;
442 // Check all channels are zeroed out before we trigger resampling.
443 for (uint32 i = 0; i < kNumFramesToTest; ++i) {
444 for (int c = 0; c < kOutputChannels; ++c)
445 EXPECT_EQ(0, sink_data[i * kOutputChannels + c]);
448 // Trigger the speech sink to resample the input data.
449 AssertConsumedBuffers(0U);
450 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
452 // Render the audio data from the recognizer.
453 sink_bus->ToInterleaved(sink_bus->frames(),
454 sink_params().bits_per_sample() / 8, sink_data);
456 // Resampled data expected frames. Extracted based on |source_data()|.
457 const int16 kExpectedData[kNumFramesToTest] = {0, 2, 5, 8, 11, 13,
458 16, 19, 22, 24, 27, 30};
460 // Check all channels have the same resampled data.
461 for (uint32 i = 0; i < kNumFramesToTest; ++i) {
462 for (int c = 0; c < kOutputChannels; ++c)
463 EXPECT_EQ(kExpectedData[i], sink_data[i * kOutputChannels + c]);
467 // Checks that the producer does not misbehave when a socket failure occurs.
468 TEST_F(SpeechRecognitionAudioSinkTest, SyncSocketFailsSendingData) {
469 const uint32 kBuffersPerNotification = Initialize(44100, 441, 16000, 1600);
470 // Start with no problems on the socket.
471 AssertConsumedBuffers(0U);
472 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
474 // A failure occurs (socket cannot send).
475 SetFailureModeOnForeignSocket(true);
476 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
479 // A very unlikely scenario in which the peer is not synchronizing for a long
480 // time (e.g. 300 ms) which results in dropping cached buffers and restarting.
481 // We check that the FIFO overflow does not occur and that the producer is able
482 // to resume.
483 TEST_F(SpeechRecognitionAudioSinkTest, RepeatedSycnhronizationLag) {
484 const uint32 kBuffersPerNotification = Initialize(44100, 441, 16000, 1600);
486 // Start with no synchronization problems.
487 AssertConsumedBuffers(0U);
488 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
490 // Consumer gets out of sync.
491 recognizer()->SimulateResponsiveness(false);
492 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
493 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
494 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
496 // Consumer recovers.
497 recognizer()->SimulateResponsiveness(true);
498 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 2U);
499 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 3U);
500 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 4U);
503 // Checks that an OnStoppedCallback is issued when the track is stopped.
504 TEST_F(SpeechRecognitionAudioSinkTest, OnReadyStateChangedOccured) {
505 const uint32 kBuffersPerNotification = Initialize(44100, 441, 16000, 1600);
506 AssertConsumedBuffers(0U);
507 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
508 EXPECT_CALL(*this, StoppedCallback()).Times(1);
510 native_track()->Stop();
511 CaptureAudioAndAssertConsumedBuffers(kBuffersPerNotification, 1U);
514 } // namespace content