xref: /aosp_15_r20/external/webrtc/sdk/android/src/jni/audio_device/opensles_recorder.cc (revision d9f758449e529ab9291ac668be2861e7a55c2422)
1 /*
2  *  Copyright (c) 2016 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 "sdk/android/src/jni/audio_device/opensles_recorder.h"
12 
13 #include <android/log.h>
14 
15 #include <memory>
16 
17 #include "api/array_view.h"
18 #include "modules/audio_device/fine_audio_buffer.h"
19 #include "rtc_base/arraysize.h"
20 #include "rtc_base/checks.h"
21 #include "rtc_base/platform_thread.h"
22 #include "rtc_base/time_utils.h"
23 #include "sdk/android/src/jni/audio_device/audio_common.h"
24 
25 #define TAG "OpenSLESRecorder"
26 #define ALOGV(...) __android_log_print(ANDROID_LOG_VERBOSE, TAG, __VA_ARGS__)
27 #define ALOGD(...) __android_log_print(ANDROID_LOG_DEBUG, TAG, __VA_ARGS__)
28 #define ALOGE(...) __android_log_print(ANDROID_LOG_ERROR, TAG, __VA_ARGS__)
29 #define ALOGW(...) __android_log_print(ANDROID_LOG_WARN, TAG, __VA_ARGS__)
30 #define ALOGI(...) __android_log_print(ANDROID_LOG_INFO, TAG, __VA_ARGS__)
31 
32 #define LOG_ON_ERROR(op)                                    \
33   [](SLresult err) {                                        \
34     if (err != SL_RESULT_SUCCESS) {                         \
35       ALOGE("%s:%d %s failed: %s", __FILE__, __LINE__, #op, \
36             GetSLErrorString(err));                         \
37       return true;                                          \
38     }                                                       \
39     return false;                                           \
40   }(op)
41 
42 namespace webrtc {
43 
44 namespace jni {
45 
OpenSLESRecorder(const AudioParameters & audio_parameters,rtc::scoped_refptr<OpenSLEngineManager> engine_manager)46 OpenSLESRecorder::OpenSLESRecorder(
47     const AudioParameters& audio_parameters,
48     rtc::scoped_refptr<OpenSLEngineManager> engine_manager)
49     : audio_parameters_(audio_parameters),
50       audio_device_buffer_(nullptr),
51       initialized_(false),
52       recording_(false),
53       engine_manager_(std::move(engine_manager)),
54       engine_(nullptr),
55       recorder_(nullptr),
56       simple_buffer_queue_(nullptr),
57       buffer_index_(0),
58       last_rec_time_(0) {
59   ALOGD("ctor[tid=%d]", rtc::CurrentThreadId());
60   // Detach from this thread since we want to use the checker to verify calls
61   // from the internal  audio thread.
62   thread_checker_opensles_.Detach();
63   // Use native audio output parameters provided by the audio manager and
64   // define the PCM format structure.
65   pcm_format_ = CreatePCMConfiguration(audio_parameters_.channels(),
66                                        audio_parameters_.sample_rate(),
67                                        audio_parameters_.bits_per_sample());
68 }
69 
~OpenSLESRecorder()70 OpenSLESRecorder::~OpenSLESRecorder() {
71   ALOGD("dtor[tid=%d]", rtc::CurrentThreadId());
72   RTC_DCHECK(thread_checker_.IsCurrent());
73   Terminate();
74   DestroyAudioRecorder();
75   engine_ = nullptr;
76   RTC_DCHECK(!engine_);
77   RTC_DCHECK(!recorder_);
78   RTC_DCHECK(!simple_buffer_queue_);
79 }
80 
Init()81 int OpenSLESRecorder::Init() {
82   ALOGD("Init[tid=%d]", rtc::CurrentThreadId());
83   RTC_DCHECK(thread_checker_.IsCurrent());
84   if (audio_parameters_.channels() == 2) {
85     ALOGD("Stereo mode is enabled");
86   }
87   return 0;
88 }
89 
Terminate()90 int OpenSLESRecorder::Terminate() {
91   ALOGD("Terminate[tid=%d]", rtc::CurrentThreadId());
92   RTC_DCHECK(thread_checker_.IsCurrent());
93   StopRecording();
94   return 0;
95 }
96 
InitRecording()97 int OpenSLESRecorder::InitRecording() {
98   ALOGD("InitRecording[tid=%d]", rtc::CurrentThreadId());
99   RTC_DCHECK(thread_checker_.IsCurrent());
100   RTC_DCHECK(!initialized_);
101   RTC_DCHECK(!recording_);
102   if (!ObtainEngineInterface()) {
103     ALOGE("Failed to obtain SL Engine interface");
104     return -1;
105   }
106   CreateAudioRecorder();
107   initialized_ = true;
108   buffer_index_ = 0;
109   return 0;
110 }
111 
RecordingIsInitialized() const112 bool OpenSLESRecorder::RecordingIsInitialized() const {
113   return initialized_;
114 }
115 
StartRecording()116 int OpenSLESRecorder::StartRecording() {
117   ALOGD("StartRecording[tid=%d]", rtc::CurrentThreadId());
118   RTC_DCHECK(thread_checker_.IsCurrent());
119   RTC_DCHECK(initialized_);
120   RTC_DCHECK(!recording_);
121   if (fine_audio_buffer_) {
122     fine_audio_buffer_->ResetRecord();
123   }
124   // Add buffers to the queue before changing state to SL_RECORDSTATE_RECORDING
125   // to ensure that recording starts as soon as the state is modified. On some
126   // devices, SLAndroidSimpleBufferQueue::Clear() used in Stop() does not flush
127   // the buffers as intended and we therefore check the number of buffers
128   // already queued first. Enqueue() can return SL_RESULT_BUFFER_INSUFFICIENT
129   // otherwise.
130   int num_buffers_in_queue = GetBufferCount();
131   for (int i = 0; i < kNumOfOpenSLESBuffers - num_buffers_in_queue; ++i) {
132     if (!EnqueueAudioBuffer()) {
133       recording_ = false;
134       return -1;
135     }
136   }
137   num_buffers_in_queue = GetBufferCount();
138   RTC_DCHECK_EQ(num_buffers_in_queue, kNumOfOpenSLESBuffers);
139   LogBufferState();
140   // Start audio recording by changing the state to SL_RECORDSTATE_RECORDING.
141   // Given that buffers are already enqueued, recording should start at once.
142   // The macro returns -1 if recording fails to start.
143   last_rec_time_ = rtc::Time();
144   if (LOG_ON_ERROR(
145           (*recorder_)->SetRecordState(recorder_, SL_RECORDSTATE_RECORDING))) {
146     return -1;
147   }
148   recording_ = (GetRecordState() == SL_RECORDSTATE_RECORDING);
149   RTC_DCHECK(recording_);
150   return 0;
151 }
152 
StopRecording()153 int OpenSLESRecorder::StopRecording() {
154   ALOGD("StopRecording[tid=%d]", rtc::CurrentThreadId());
155   RTC_DCHECK(thread_checker_.IsCurrent());
156   if (!initialized_ || !recording_) {
157     return 0;
158   }
159   // Stop recording by setting the record state to SL_RECORDSTATE_STOPPED.
160   if (LOG_ON_ERROR(
161           (*recorder_)->SetRecordState(recorder_, SL_RECORDSTATE_STOPPED))) {
162     return -1;
163   }
164   // Clear the buffer queue to get rid of old data when resuming recording.
165   if (LOG_ON_ERROR((*simple_buffer_queue_)->Clear(simple_buffer_queue_))) {
166     return -1;
167   }
168   thread_checker_opensles_.Detach();
169   initialized_ = false;
170   recording_ = false;
171   return 0;
172 }
173 
Recording() const174 bool OpenSLESRecorder::Recording() const {
175   return recording_;
176 }
177 
AttachAudioBuffer(AudioDeviceBuffer * audio_buffer)178 void OpenSLESRecorder::AttachAudioBuffer(AudioDeviceBuffer* audio_buffer) {
179   ALOGD("AttachAudioBuffer");
180   RTC_DCHECK(thread_checker_.IsCurrent());
181   RTC_CHECK(audio_buffer);
182   audio_device_buffer_ = audio_buffer;
183   // Ensure that the audio device buffer is informed about the native sample
184   // rate used on the recording side.
185   const int sample_rate_hz = audio_parameters_.sample_rate();
186   ALOGD("SetRecordingSampleRate(%d)", sample_rate_hz);
187   audio_device_buffer_->SetRecordingSampleRate(sample_rate_hz);
188   // Ensure that the audio device buffer is informed about the number of
189   // channels preferred by the OS on the recording side.
190   const size_t channels = audio_parameters_.channels();
191   ALOGD("SetRecordingChannels(%zu)", channels);
192   audio_device_buffer_->SetRecordingChannels(channels);
193   // Allocated memory for internal data buffers given existing audio parameters.
194   AllocateDataBuffers();
195 }
196 
IsAcousticEchoCancelerSupported() const197 bool OpenSLESRecorder::IsAcousticEchoCancelerSupported() const {
198   return false;
199 }
200 
IsNoiseSuppressorSupported() const201 bool OpenSLESRecorder::IsNoiseSuppressorSupported() const {
202   return false;
203 }
204 
EnableBuiltInAEC(bool enable)205 int OpenSLESRecorder::EnableBuiltInAEC(bool enable) {
206   ALOGD("EnableBuiltInAEC(%d)", enable);
207   RTC_DCHECK(thread_checker_.IsCurrent());
208   ALOGE("Not implemented");
209   return 0;
210 }
211 
EnableBuiltInNS(bool enable)212 int OpenSLESRecorder::EnableBuiltInNS(bool enable) {
213   ALOGD("EnableBuiltInNS(%d)", enable);
214   RTC_DCHECK(thread_checker_.IsCurrent());
215   ALOGE("Not implemented");
216   return 0;
217 }
218 
ObtainEngineInterface()219 bool OpenSLESRecorder::ObtainEngineInterface() {
220   ALOGD("ObtainEngineInterface");
221   RTC_DCHECK(thread_checker_.IsCurrent());
222   if (engine_)
223     return true;
224   // Get access to (or create if not already existing) the global OpenSL Engine
225   // object.
226   SLObjectItf engine_object = engine_manager_->GetOpenSLEngine();
227   if (engine_object == nullptr) {
228     ALOGE("Failed to access the global OpenSL engine");
229     return false;
230   }
231   // Get the SL Engine Interface which is implicit.
232   if (LOG_ON_ERROR(
233           (*engine_object)
234               ->GetInterface(engine_object, SL_IID_ENGINE, &engine_))) {
235     return false;
236   }
237   return true;
238 }
239 
CreateAudioRecorder()240 bool OpenSLESRecorder::CreateAudioRecorder() {
241   ALOGD("CreateAudioRecorder");
242   RTC_DCHECK(thread_checker_.IsCurrent());
243   if (recorder_object_.Get())
244     return true;
245   RTC_DCHECK(!recorder_);
246   RTC_DCHECK(!simple_buffer_queue_);
247 
248   // Audio source configuration.
249   SLDataLocator_IODevice mic_locator = {SL_DATALOCATOR_IODEVICE,
250                                         SL_IODEVICE_AUDIOINPUT,
251                                         SL_DEFAULTDEVICEID_AUDIOINPUT, NULL};
252   SLDataSource audio_source = {&mic_locator, NULL};
253 
254   // Audio sink configuration.
255   SLDataLocator_AndroidSimpleBufferQueue buffer_queue = {
256       SL_DATALOCATOR_ANDROIDSIMPLEBUFFERQUEUE,
257       static_cast<SLuint32>(kNumOfOpenSLESBuffers)};
258   SLDataSink audio_sink = {&buffer_queue, &pcm_format_};
259 
260   // Create the audio recorder object (requires the RECORD_AUDIO permission).
261   // Do not realize the recorder yet. Set the configuration first.
262   const SLInterfaceID interface_id[] = {SL_IID_ANDROIDSIMPLEBUFFERQUEUE,
263                                         SL_IID_ANDROIDCONFIGURATION};
264   const SLboolean interface_required[] = {SL_BOOLEAN_TRUE, SL_BOOLEAN_TRUE};
265   if (LOG_ON_ERROR((*engine_)->CreateAudioRecorder(
266           engine_, recorder_object_.Receive(), &audio_source, &audio_sink,
267           arraysize(interface_id), interface_id, interface_required))) {
268     return false;
269   }
270 
271   // Configure the audio recorder (before it is realized).
272   SLAndroidConfigurationItf recorder_config;
273   if (LOG_ON_ERROR((recorder_object_->GetInterface(recorder_object_.Get(),
274                                                    SL_IID_ANDROIDCONFIGURATION,
275                                                    &recorder_config)))) {
276     return false;
277   }
278 
279   // Uses the default microphone tuned for audio communication.
280   // Note that, SL_ANDROID_RECORDING_PRESET_VOICE_RECOGNITION leads to a fast
281   // track but also excludes usage of required effects like AEC, AGC and NS.
282   // SL_ANDROID_RECORDING_PRESET_VOICE_COMMUNICATION
283   SLint32 stream_type = SL_ANDROID_RECORDING_PRESET_VOICE_COMMUNICATION;
284   if (LOG_ON_ERROR(((*recorder_config)
285                         ->SetConfiguration(recorder_config,
286                                            SL_ANDROID_KEY_RECORDING_PRESET,
287                                            &stream_type, sizeof(SLint32))))) {
288     return false;
289   }
290 
291   // The audio recorder can now be realized (in synchronous mode).
292   if (LOG_ON_ERROR((recorder_object_->Realize(recorder_object_.Get(),
293                                               SL_BOOLEAN_FALSE)))) {
294     return false;
295   }
296 
297   // Get the implicit recorder interface (SL_IID_RECORD).
298   if (LOG_ON_ERROR((recorder_object_->GetInterface(
299           recorder_object_.Get(), SL_IID_RECORD, &recorder_)))) {
300     return false;
301   }
302 
303   // Get the simple buffer queue interface (SL_IID_ANDROIDSIMPLEBUFFERQUEUE).
304   // It was explicitly requested.
305   if (LOG_ON_ERROR((recorder_object_->GetInterface(
306           recorder_object_.Get(), SL_IID_ANDROIDSIMPLEBUFFERQUEUE,
307           &simple_buffer_queue_)))) {
308     return false;
309   }
310 
311   // Register the input callback for the simple buffer queue.
312   // This callback will be called when receiving new data from the device.
313   if (LOG_ON_ERROR(((*simple_buffer_queue_)
314                         ->RegisterCallback(simple_buffer_queue_,
315                                            SimpleBufferQueueCallback, this)))) {
316     return false;
317   }
318   return true;
319 }
320 
DestroyAudioRecorder()321 void OpenSLESRecorder::DestroyAudioRecorder() {
322   ALOGD("DestroyAudioRecorder");
323   RTC_DCHECK(thread_checker_.IsCurrent());
324   if (!recorder_object_.Get())
325     return;
326   (*simple_buffer_queue_)
327       ->RegisterCallback(simple_buffer_queue_, nullptr, nullptr);
328   recorder_object_.Reset();
329   recorder_ = nullptr;
330   simple_buffer_queue_ = nullptr;
331 }
332 
SimpleBufferQueueCallback(SLAndroidSimpleBufferQueueItf buffer_queue,void * context)333 void OpenSLESRecorder::SimpleBufferQueueCallback(
334     SLAndroidSimpleBufferQueueItf buffer_queue,
335     void* context) {
336   OpenSLESRecorder* stream = static_cast<OpenSLESRecorder*>(context);
337   stream->ReadBufferQueue();
338 }
339 
AllocateDataBuffers()340 void OpenSLESRecorder::AllocateDataBuffers() {
341   ALOGD("AllocateDataBuffers");
342   RTC_DCHECK(thread_checker_.IsCurrent());
343   RTC_DCHECK(!simple_buffer_queue_);
344   RTC_CHECK(audio_device_buffer_);
345   // Create a modified audio buffer class which allows us to deliver any number
346   // of samples (and not only multiple of 10ms) to match the native audio unit
347   // buffer size.
348   ALOGD("frames per native buffer: %zu", audio_parameters_.frames_per_buffer());
349   ALOGD("frames per 10ms buffer: %zu",
350         audio_parameters_.frames_per_10ms_buffer());
351   ALOGD("bytes per native buffer: %zu", audio_parameters_.GetBytesPerBuffer());
352   ALOGD("native sample rate: %d", audio_parameters_.sample_rate());
353   RTC_DCHECK(audio_device_buffer_);
354   fine_audio_buffer_ = std::make_unique<FineAudioBuffer>(audio_device_buffer_);
355   // Allocate queue of audio buffers that stores recorded audio samples.
356   const int buffer_size_samples =
357       audio_parameters_.frames_per_buffer() * audio_parameters_.channels();
358   audio_buffers_.reset(new std::unique_ptr<SLint16[]>[kNumOfOpenSLESBuffers]);
359   for (int i = 0; i < kNumOfOpenSLESBuffers; ++i) {
360     audio_buffers_[i].reset(new SLint16[buffer_size_samples]);
361   }
362 }
363 
ReadBufferQueue()364 void OpenSLESRecorder::ReadBufferQueue() {
365   RTC_DCHECK(thread_checker_opensles_.IsCurrent());
366   SLuint32 state = GetRecordState();
367   if (state != SL_RECORDSTATE_RECORDING) {
368     ALOGW("Buffer callback in non-recording state!");
369     return;
370   }
371   // Check delta time between two successive callbacks and provide a warning
372   // if it becomes very large.
373   // TODO(henrika): using 150ms as upper limit but this value is rather random.
374   const uint32_t current_time = rtc::Time();
375   const uint32_t diff = current_time - last_rec_time_;
376   if (diff > 150) {
377     ALOGW("Bad OpenSL ES record timing, dT=%u [ms]", diff);
378   }
379   last_rec_time_ = current_time;
380   // Send recorded audio data to the WebRTC sink.
381   // TODO(henrika): fix delay estimates. It is OK to use fixed values for now
382   // since there is no support to turn off built-in EC in combination with
383   // OpenSL ES anyhow. Hence, as is, the WebRTC based AEC (which would use
384   // these estimates) will never be active.
385   fine_audio_buffer_->DeliverRecordedData(
386       rtc::ArrayView<const int16_t>(
387           audio_buffers_[buffer_index_].get(),
388           audio_parameters_.frames_per_buffer() * audio_parameters_.channels()),
389       25);
390   // Enqueue the utilized audio buffer and use if for recording again.
391   EnqueueAudioBuffer();
392 }
393 
EnqueueAudioBuffer()394 bool OpenSLESRecorder::EnqueueAudioBuffer() {
395   SLresult err =
396       (*simple_buffer_queue_)
397           ->Enqueue(
398               simple_buffer_queue_,
399               reinterpret_cast<SLint8*>(audio_buffers_[buffer_index_].get()),
400               audio_parameters_.GetBytesPerBuffer());
401   if (SL_RESULT_SUCCESS != err) {
402     ALOGE("Enqueue failed: %s", GetSLErrorString(err));
403     return false;
404   }
405   buffer_index_ = (buffer_index_ + 1) % kNumOfOpenSLESBuffers;
406   return true;
407 }
408 
GetRecordState() const409 SLuint32 OpenSLESRecorder::GetRecordState() const {
410   RTC_DCHECK(recorder_);
411   SLuint32 state;
412   SLresult err = (*recorder_)->GetRecordState(recorder_, &state);
413   if (SL_RESULT_SUCCESS != err) {
414     ALOGE("GetRecordState failed: %s", GetSLErrorString(err));
415   }
416   return state;
417 }
418 
GetBufferQueueState() const419 SLAndroidSimpleBufferQueueState OpenSLESRecorder::GetBufferQueueState() const {
420   RTC_DCHECK(simple_buffer_queue_);
421   // state.count: Number of buffers currently in the queue.
422   // state.index: Index of the currently filling buffer. This is a linear index
423   // that keeps a cumulative count of the number of buffers recorded.
424   SLAndroidSimpleBufferQueueState state;
425   SLresult err =
426       (*simple_buffer_queue_)->GetState(simple_buffer_queue_, &state);
427   if (SL_RESULT_SUCCESS != err) {
428     ALOGE("GetState failed: %s", GetSLErrorString(err));
429   }
430   return state;
431 }
432 
LogBufferState() const433 void OpenSLESRecorder::LogBufferState() const {
434   SLAndroidSimpleBufferQueueState state = GetBufferQueueState();
435   ALOGD("state.count:%d state.index:%d", state.count, state.index);
436 }
437 
GetBufferCount()438 SLuint32 OpenSLESRecorder::GetBufferCount() {
439   SLAndroidSimpleBufferQueueState state = GetBufferQueueState();
440   return state.count;
441 }
442 
443 }  // namespace jni
444 
445 }  // namespace webrtc
446