1 // Copyright (c) 2012 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.
4
5 #include "media/audio/audio_input_controller.h"
6
7 #include "base/bind.h"
8 #include "base/strings/stringprintf.h"
9 #include "base/threading/thread_restrictions.h"
10 #include "base/time/time.h"
11 #include "media/base/limits.h"
12 #include "media/base/scoped_histogram_timer.h"
13 #include "media/base/user_input_monitor.h"
14
15 using base::TimeDelta;
16
17 namespace {
18 const int kMaxInputChannels = 3;
19
20 // TODO(henrika): remove usage of timers and add support for proper
21 // notification of when the input device is removed. This was originally added
22 // to resolve http://crbug.com/79936 for Windows platforms. This then caused
23 // breakage (very hard to repro bugs!) on other platforms: See
24 // http://crbug.com/226327 and http://crbug.com/230972.
25 // See also that the timer has been disabled on Mac now due to
26 // crbug.com/357501.
27 const int kTimerResetIntervalSeconds = 1;
28 // We have received reports that the timer can be too trigger happy on some
29 // Mac devices and the initial timer interval has therefore been increased
30 // from 1 second to 5 seconds.
31 const int kTimerInitialIntervalSeconds = 5;
32
33 #if defined(AUDIO_POWER_MONITORING)
34 // Time constant for AudioPowerMonitor.
35 // The utilized smoothing factor (alpha) in the exponential filter is given
36 // by 1-exp(-1/(fs*ts)), where fs is the sample rate in Hz and ts is the time
37 // constant given by |kPowerMeasurementTimeConstantMilliseconds|.
38 // Example: fs=44100, ts=10e-3 => alpha~0.022420
39 // fs=44100, ts=20e-3 => alpha~0.165903
40 // A large smoothing factor corresponds to a faster filter response to input
41 // changes since y(n)=alpha*x(n)+(1-alpha)*y(n-1), where x(n) is the input
42 // and y(n) is the output.
43 const int kPowerMeasurementTimeConstantMilliseconds = 10;
44
45 // Time in seconds between two successive measurements of audio power levels.
46 const int kPowerMonitorLogIntervalSeconds = 5;
47 #endif
48 }
49
50 namespace media {
51
52 // static
53 AudioInputController::Factory* AudioInputController::factory_ = NULL;
54
AudioInputController(EventHandler * handler,SyncWriter * sync_writer,UserInputMonitor * user_input_monitor)55 AudioInputController::AudioInputController(EventHandler* handler,
56 SyncWriter* sync_writer,
57 UserInputMonitor* user_input_monitor)
58 : creator_task_runner_(base::MessageLoopProxy::current()),
59 handler_(handler),
60 stream_(NULL),
61 data_is_active_(false),
62 state_(CLOSED),
63 sync_writer_(sync_writer),
64 max_volume_(0.0),
65 user_input_monitor_(user_input_monitor),
66 prev_key_down_count_(0) {
67 DCHECK(creator_task_runner_.get());
68 }
69
~AudioInputController()70 AudioInputController::~AudioInputController() {
71 DCHECK_EQ(state_, CLOSED);
72 }
73
74 // static
Create(AudioManager * audio_manager,EventHandler * event_handler,const AudioParameters & params,const std::string & device_id,UserInputMonitor * user_input_monitor)75 scoped_refptr<AudioInputController> AudioInputController::Create(
76 AudioManager* audio_manager,
77 EventHandler* event_handler,
78 const AudioParameters& params,
79 const std::string& device_id,
80 UserInputMonitor* user_input_monitor) {
81 DCHECK(audio_manager);
82
83 if (!params.IsValid() || (params.channels() > kMaxInputChannels))
84 return NULL;
85
86 if (factory_) {
87 return factory_->Create(
88 audio_manager, event_handler, params, user_input_monitor);
89 }
90 scoped_refptr<AudioInputController> controller(
91 new AudioInputController(event_handler, NULL, user_input_monitor));
92
93 controller->task_runner_ = audio_manager->GetTaskRunner();
94
95 // Create and open a new audio input stream from the existing
96 // audio-device thread.
97 if (!controller->task_runner_->PostTask(FROM_HERE,
98 base::Bind(&AudioInputController::DoCreate, controller,
99 base::Unretained(audio_manager), params, device_id))) {
100 controller = NULL;
101 }
102
103 return controller;
104 }
105
106 // static
CreateLowLatency(AudioManager * audio_manager,EventHandler * event_handler,const AudioParameters & params,const std::string & device_id,SyncWriter * sync_writer,UserInputMonitor * user_input_monitor)107 scoped_refptr<AudioInputController> AudioInputController::CreateLowLatency(
108 AudioManager* audio_manager,
109 EventHandler* event_handler,
110 const AudioParameters& params,
111 const std::string& device_id,
112 SyncWriter* sync_writer,
113 UserInputMonitor* user_input_monitor) {
114 DCHECK(audio_manager);
115 DCHECK(sync_writer);
116
117 if (!params.IsValid() || (params.channels() > kMaxInputChannels))
118 return NULL;
119
120 // Create the AudioInputController object and ensure that it runs on
121 // the audio-manager thread.
122 scoped_refptr<AudioInputController> controller(
123 new AudioInputController(event_handler, sync_writer, user_input_monitor));
124 controller->task_runner_ = audio_manager->GetTaskRunner();
125
126 // Create and open a new audio input stream from the existing
127 // audio-device thread. Use the provided audio-input device.
128 if (!controller->task_runner_->PostTask(FROM_HERE,
129 base::Bind(&AudioInputController::DoCreate, controller,
130 base::Unretained(audio_manager), params, device_id))) {
131 controller = NULL;
132 }
133
134 return controller;
135 }
136
137 // static
CreateForStream(const scoped_refptr<base::SingleThreadTaskRunner> & task_runner,EventHandler * event_handler,AudioInputStream * stream,SyncWriter * sync_writer,UserInputMonitor * user_input_monitor)138 scoped_refptr<AudioInputController> AudioInputController::CreateForStream(
139 const scoped_refptr<base::SingleThreadTaskRunner>& task_runner,
140 EventHandler* event_handler,
141 AudioInputStream* stream,
142 SyncWriter* sync_writer,
143 UserInputMonitor* user_input_monitor) {
144 DCHECK(sync_writer);
145 DCHECK(stream);
146
147 // Create the AudioInputController object and ensure that it runs on
148 // the audio-manager thread.
149 scoped_refptr<AudioInputController> controller(
150 new AudioInputController(event_handler, sync_writer, user_input_monitor));
151 controller->task_runner_ = task_runner;
152
153 // TODO(miu): See TODO at top of file. Until that's resolved, we need to
154 // disable the error auto-detection here (since the audio mirroring
155 // implementation will reliably report error and close events). Note, of
156 // course, that we're assuming CreateForStream() has been called for the audio
157 // mirroring use case only.
158 if (!controller->task_runner_->PostTask(
159 FROM_HERE,
160 base::Bind(&AudioInputController::DoCreateForStream, controller,
161 stream, false))) {
162 controller = NULL;
163 }
164
165 return controller;
166 }
167
Record()168 void AudioInputController::Record() {
169 task_runner_->PostTask(FROM_HERE, base::Bind(
170 &AudioInputController::DoRecord, this));
171 }
172
Close(const base::Closure & closed_task)173 void AudioInputController::Close(const base::Closure& closed_task) {
174 DCHECK(!closed_task.is_null());
175 DCHECK(creator_task_runner_->BelongsToCurrentThread());
176
177 task_runner_->PostTaskAndReply(
178 FROM_HERE, base::Bind(&AudioInputController::DoClose, this), closed_task);
179 }
180
SetVolume(double volume)181 void AudioInputController::SetVolume(double volume) {
182 task_runner_->PostTask(FROM_HERE, base::Bind(
183 &AudioInputController::DoSetVolume, this, volume));
184 }
185
SetAutomaticGainControl(bool enabled)186 void AudioInputController::SetAutomaticGainControl(bool enabled) {
187 task_runner_->PostTask(FROM_HERE, base::Bind(
188 &AudioInputController::DoSetAutomaticGainControl, this, enabled));
189 }
190
DoCreate(AudioManager * audio_manager,const AudioParameters & params,const std::string & device_id)191 void AudioInputController::DoCreate(AudioManager* audio_manager,
192 const AudioParameters& params,
193 const std::string& device_id) {
194 DCHECK(task_runner_->BelongsToCurrentThread());
195 SCOPED_UMA_HISTOGRAM_TIMER("Media.AudioInputController.CreateTime");
196
197 #if defined(AUDIO_POWER_MONITORING)
198 // Create the audio (power) level meter given the provided audio parameters.
199 // An AudioBus is also needed to wrap the raw data buffer from the native
200 // layer to match AudioPowerMonitor::Scan().
201 // TODO(henrika): Remove use of extra AudioBus. See http://crbug.com/375155.
202 audio_level_.reset(new media::AudioPowerMonitor(
203 params.sample_rate(),
204 TimeDelta::FromMilliseconds(kPowerMeasurementTimeConstantMilliseconds)));
205 audio_params_ = params;
206 #endif
207
208 // TODO(miu): See TODO at top of file. Until that's resolved, assume all
209 // platform audio input requires the |no_data_timer_| be used to auto-detect
210 // errors. In reality, probably only Windows needs to be treated as
211 // unreliable here.
212 DoCreateForStream(audio_manager->MakeAudioInputStream(params, device_id),
213 true);
214 }
215
DoCreateForStream(AudioInputStream * stream_to_control,bool enable_nodata_timer)216 void AudioInputController::DoCreateForStream(
217 AudioInputStream* stream_to_control, bool enable_nodata_timer) {
218 DCHECK(task_runner_->BelongsToCurrentThread());
219
220 DCHECK(!stream_);
221 stream_ = stream_to_control;
222
223 if (!stream_) {
224 if (handler_)
225 handler_->OnError(this, STREAM_CREATE_ERROR);
226 return;
227 }
228
229 if (stream_ && !stream_->Open()) {
230 stream_->Close();
231 stream_ = NULL;
232 if (handler_)
233 handler_->OnError(this, STREAM_OPEN_ERROR);
234 return;
235 }
236
237 DCHECK(!no_data_timer_.get());
238
239 // The timer is enabled for logging purposes. The NO_DATA_ERROR triggered
240 // from the timer must be ignored by the EventHandler.
241 // TODO(henrika): remove usage of timer when it has been verified on Canary
242 // that we are safe doing so. Goal is to get rid of |no_data_timer_| and
243 // everything that is tied to it. crbug.com/357569.
244 enable_nodata_timer = true;
245
246 if (enable_nodata_timer) {
247 // Create the data timer which will call FirstCheckForNoData(). The timer
248 // is started in DoRecord() and restarted in each DoCheckForNoData()
249 // callback.
250 no_data_timer_.reset(new base::Timer(
251 FROM_HERE, base::TimeDelta::FromSeconds(kTimerInitialIntervalSeconds),
252 base::Bind(&AudioInputController::FirstCheckForNoData,
253 base::Unretained(this)), false));
254 } else {
255 DVLOG(1) << "Disabled: timer check for no data.";
256 }
257
258 state_ = CREATED;
259 if (handler_)
260 handler_->OnCreated(this);
261
262 if (user_input_monitor_) {
263 user_input_monitor_->EnableKeyPressMonitoring();
264 prev_key_down_count_ = user_input_monitor_->GetKeyPressCount();
265 }
266 }
267
DoRecord()268 void AudioInputController::DoRecord() {
269 DCHECK(task_runner_->BelongsToCurrentThread());
270 SCOPED_UMA_HISTOGRAM_TIMER("Media.AudioInputController.RecordTime");
271
272 if (state_ != CREATED)
273 return;
274
275 {
276 base::AutoLock auto_lock(lock_);
277 state_ = RECORDING;
278 }
279
280 if (no_data_timer_) {
281 // Start the data timer. Once |kTimerResetIntervalSeconds| have passed,
282 // a callback to FirstCheckForNoData() is made.
283 no_data_timer_->Reset();
284 }
285
286 stream_->Start(this);
287 if (handler_)
288 handler_->OnRecording(this);
289 }
290
DoClose()291 void AudioInputController::DoClose() {
292 DCHECK(task_runner_->BelongsToCurrentThread());
293 SCOPED_UMA_HISTOGRAM_TIMER("Media.AudioInputController.CloseTime");
294
295 if (state_ == CLOSED)
296 return;
297
298 // Delete the timer on the same thread that created it.
299 no_data_timer_.reset();
300
301 DoStopCloseAndClearStream();
302 SetDataIsActive(false);
303
304 if (SharedMemoryAndSyncSocketMode())
305 sync_writer_->Close();
306
307 if (user_input_monitor_)
308 user_input_monitor_->DisableKeyPressMonitoring();
309
310 state_ = CLOSED;
311 }
312
DoReportError()313 void AudioInputController::DoReportError() {
314 DCHECK(task_runner_->BelongsToCurrentThread());
315 if (handler_)
316 handler_->OnError(this, STREAM_ERROR);
317 }
318
DoSetVolume(double volume)319 void AudioInputController::DoSetVolume(double volume) {
320 DCHECK(task_runner_->BelongsToCurrentThread());
321 DCHECK_GE(volume, 0);
322 DCHECK_LE(volume, 1.0);
323
324 if (state_ != CREATED && state_ != RECORDING)
325 return;
326
327 // Only ask for the maximum volume at first call and use cached value
328 // for remaining function calls.
329 if (!max_volume_) {
330 max_volume_ = stream_->GetMaxVolume();
331 }
332
333 if (max_volume_ == 0.0) {
334 DLOG(WARNING) << "Failed to access input volume control";
335 return;
336 }
337
338 // Set the stream volume and scale to a range matched to the platform.
339 stream_->SetVolume(max_volume_ * volume);
340 }
341
DoSetAutomaticGainControl(bool enabled)342 void AudioInputController::DoSetAutomaticGainControl(bool enabled) {
343 DCHECK(task_runner_->BelongsToCurrentThread());
344 DCHECK_NE(state_, RECORDING);
345
346 // Ensure that the AGC state only can be modified before streaming starts.
347 if (state_ != CREATED)
348 return;
349
350 stream_->SetAutomaticGainControl(enabled);
351 }
352
FirstCheckForNoData()353 void AudioInputController::FirstCheckForNoData() {
354 DCHECK(task_runner_->BelongsToCurrentThread());
355 UMA_HISTOGRAM_BOOLEAN("Media.AudioInputControllerCaptureStartupSuccess",
356 GetDataIsActive());
357 DoCheckForNoData();
358 }
359
DoCheckForNoData()360 void AudioInputController::DoCheckForNoData() {
361 DCHECK(task_runner_->BelongsToCurrentThread());
362
363 if (!GetDataIsActive()) {
364 // The data-is-active marker will be false only if it has been more than
365 // one second since a data packet was recorded. This can happen if a
366 // capture device has been removed or disabled.
367 if (handler_)
368 handler_->OnError(this, NO_DATA_ERROR);
369 }
370
371 // Mark data as non-active. The flag will be re-enabled in OnData() each
372 // time a data packet is received. Hence, under normal conditions, the
373 // flag will only be disabled during a very short period.
374 SetDataIsActive(false);
375
376 // Restart the timer to ensure that we check the flag again in
377 // |kTimerResetIntervalSeconds|.
378 no_data_timer_->Start(
379 FROM_HERE, base::TimeDelta::FromSeconds(kTimerResetIntervalSeconds),
380 base::Bind(&AudioInputController::DoCheckForNoData,
381 base::Unretained(this)));
382 }
383
OnData(AudioInputStream * stream,const AudioBus * source,uint32 hardware_delay_bytes,double volume)384 void AudioInputController::OnData(AudioInputStream* stream,
385 const AudioBus* source,
386 uint32 hardware_delay_bytes,
387 double volume) {
388 // Mark data as active to ensure that the periodic calls to
389 // DoCheckForNoData() does not report an error to the event handler.
390 SetDataIsActive(true);
391
392 {
393 base::AutoLock auto_lock(lock_);
394 if (state_ != RECORDING)
395 return;
396 }
397
398 bool key_pressed = false;
399 if (user_input_monitor_) {
400 size_t current_count = user_input_monitor_->GetKeyPressCount();
401 key_pressed = current_count != prev_key_down_count_;
402 prev_key_down_count_ = current_count;
403 DVLOG_IF(6, key_pressed) << "Detected keypress.";
404 }
405
406 // Use SharedMemory and SyncSocket if the client has created a SyncWriter.
407 // Used by all low-latency clients except WebSpeech.
408 if (SharedMemoryAndSyncSocketMode()) {
409 sync_writer_->Write(source, volume, key_pressed);
410 sync_writer_->UpdateRecordedBytes(hardware_delay_bytes);
411
412 #if defined(AUDIO_POWER_MONITORING)
413 // Only do power-level measurements if an AudioPowerMonitor object has
414 // been created. Done in DoCreate() but not DoCreateForStream(), hence
415 // logging will mainly be done for WebRTC and WebSpeech clients.
416 if (!audio_level_)
417 return;
418
419 // Perform periodic audio (power) level measurements.
420 if ((base::TimeTicks::Now() - last_audio_level_log_time_).InSeconds() >
421 kPowerMonitorLogIntervalSeconds) {
422 // Wrap data into an AudioBus to match AudioPowerMonitor::Scan.
423 // TODO(henrika): remove this section when capture side uses AudioBus.
424 // See http://crbug.com/375155 for details.
425 audio_level_->Scan(*source, source->frames());
426
427 // Get current average power level and add it to the log.
428 // Possible range is given by [-inf, 0] dBFS.
429 std::pair<float, bool> result = audio_level_->ReadCurrentPowerAndClip();
430
431 // Use event handler on the audio thread to relay a message to the ARIH
432 // in content which does the actual logging on the IO thread.
433 task_runner_->PostTask(
434 FROM_HERE,
435 base::Bind(
436 &AudioInputController::DoLogAudioLevel, this, result.first));
437
438 last_audio_level_log_time_ = base::TimeTicks::Now();
439
440 // Reset the average power level (since we don't log continuously).
441 audio_level_->Reset();
442 }
443 #endif
444 return;
445 }
446
447 // TODO(henrika): Investigate if we can avoid the extra copy here.
448 // (see http://crbug.com/249316 for details). AFAIK, this scope is only
449 // active for WebSpeech clients.
450 scoped_ptr<AudioBus> audio_data =
451 AudioBus::Create(source->channels(), source->frames());
452 source->CopyTo(audio_data.get());
453
454 // Ownership of the audio buffer will be with the callback until it is run,
455 // when ownership is passed to the callback function.
456 task_runner_->PostTask(
457 FROM_HERE,
458 base::Bind(
459 &AudioInputController::DoOnData, this, base::Passed(&audio_data)));
460 }
461
DoOnData(scoped_ptr<AudioBus> data)462 void AudioInputController::DoOnData(scoped_ptr<AudioBus> data) {
463 DCHECK(task_runner_->BelongsToCurrentThread());
464 if (handler_)
465 handler_->OnData(this, data.get());
466 }
467
DoLogAudioLevel(float level_dbfs)468 void AudioInputController::DoLogAudioLevel(float level_dbfs) {
469 #if defined(AUDIO_POWER_MONITORING)
470 DCHECK(task_runner_->BelongsToCurrentThread());
471 if (!handler_)
472 return;
473
474 std::string log_string = base::StringPrintf(
475 "AIC::OnData: average audio level=%.2f dBFS", level_dbfs);
476 static const float kSilenceThresholdDBFS = -72.24719896f;
477 if (level_dbfs < kSilenceThresholdDBFS)
478 log_string += " <=> no audio input!";
479
480 handler_->OnLog(this, log_string);
481 #endif
482 }
483
OnError(AudioInputStream * stream)484 void AudioInputController::OnError(AudioInputStream* stream) {
485 // Handle error on the audio-manager thread.
486 task_runner_->PostTask(FROM_HERE, base::Bind(
487 &AudioInputController::DoReportError, this));
488 }
489
DoStopCloseAndClearStream()490 void AudioInputController::DoStopCloseAndClearStream() {
491 DCHECK(task_runner_->BelongsToCurrentThread());
492
493 // Allow calling unconditionally and bail if we don't have a stream to close.
494 if (stream_ != NULL) {
495 stream_->Stop();
496 stream_->Close();
497 stream_ = NULL;
498 }
499
500 // The event handler should not be touched after the stream has been closed.
501 handler_ = NULL;
502 }
503
SetDataIsActive(bool enabled)504 void AudioInputController::SetDataIsActive(bool enabled) {
505 base::subtle::Release_Store(&data_is_active_, enabled);
506 }
507
GetDataIsActive()508 bool AudioInputController::GetDataIsActive() {
509 return (base::subtle::Acquire_Load(&data_is_active_) != false);
510 }
511
512 } // namespace media
513