def decodeStream(self, session: DialogSession) -> Optional[ASRResult]:
super().decodeStream(session)
recorder = Recorder(self._timeout, session.user, session.siteId)
self.ASRManager.addRecorder(session.siteId, recorder)
self._recorder = recorder
with Stopwatch() as processingTime:
with recorder as stream:
audioStream = stream.audioStream()
# noinspection PyUnresolvedReferences
try:
requests = (types.StreamingRecognizeRequest(
audio_content=content) for content in audioStream)
responses = self._client.streaming_recognize(
self._streamingConfig, requests)
result = self._checkResponses(session, responses)
except:
self.logWarning('Failed ASR request')
self.end()
return ASRResult(
text=result[0],
session=session,
likelihood=result[1],
processingTime=processingTime.time) if result else None
def decodeStream(self, session: DialogSession) -> Optional[ASRResult]: super().decodeStream(session) result = None with Stopwatch() as processingTime: with Recorder(self._timeout) as recorder: self.ASRManager.addRecorder(session.siteId, recorder) self._decoder.start_utt() inSpeech = False for chunk in recorder: if self._timeout.isSet(): break self._decoder.process_raw(chunk, False, False) if self._decoder.get_in_speech() != inSpeech: inSpeech = self._decoder.get_in_speech() if not inSpeech: self._decoder.end_utt() result = self._decoder.hyp() if self._decoder.hyp() else None break self.end(recorder, session) return ASRResult( text=result.hypstr.strip(), session=session, likelihood=self._decoder.hyp().prob, processingTime=processingTime.time ) if result else None
def decodeStream(self, session: DialogSession) -> Optional[ASRResult]:
super().decodeStream(session)
result = None
with Stopwatch() as processingTime:
with Recorder(self._timeout) as recorder:
self.ASRManager.addRecorder(session.siteId, recorder)
self._recorder = recorder
streamContext = self._model.createStream()
for chunk in recorder:
if not chunk:
break
self._model.feedAudioContent(
streamContext, np.frombuffer(chunk, np.int16))
result = self._model.intermediateDecode(streamContext)
self.partialTextCaptured(session=session,
text=result,
likelihood=1,
seconds=0)
text = self._model.finishStream(streamContext)
self.end(session)
return ASRResult(
text=text,
session=session,
likelihood=1.0,
processingTime=processingTime.time) if result else None
def decodeStream(self, session: DialogSession) -> Optional[ASRResult]:
super().decodeStream(session)
result = None
previous = ''
with Stopwatch() as processingTime:
with Recorder(self._timeout, session.user,
session.deviceUid) as recorder:
self.ASRManager.addRecorder(session.deviceUid, recorder)
self._recorder = recorder
streamContext = self._model.createStream()
for chunk in recorder:
if not chunk:
break
streamContext.feedAudioContent(
np.frombuffer(chunk, np.int16))
result = streamContext.intermediateDecode()
if result and result != previous:
previous = result
self.partialTextCaptured(session=session,
text=result,
likelihood=1,
seconds=0)
text = streamContext.finishStream()
self._triggerFlag.clear()
self.end()
return ASRResult(
text=text,
session=session,
likelihood=1.0,
processingTime=processingTime.time) if result else None
def __init__(self, restartHandler: callable):
Singleton.__init__(self, self.NAME)
self._logger = Logger(prepend='[Project Alice]')
self._logger.logInfo('Starting Alice main unit')
self._booted = False
self._isUpdating = False
self._shuttingDown = False
self._restart = False
self._restartHandler = restartHandler
if not self.checkDependencies():
self._restart = True
self._restartHandler()
else:
with Stopwatch() as stopWatch:
self._superManager = SuperManager(self)
self._superManager.initManagers()
self._superManager.onStart()
if self._superManager.configManager.getAliceConfigByName(
'useHLC'):
self._superManager.commons.runRootSystemCommand(
['systemctl', 'start', 'hermesledcontrol'])
self._superManager.onBooted()
self._logger.logInfo(f'Started in {stopWatch} seconds')
self._booted = True
def nluTrainingThread(self, datasetFile: Path):
try:
with Stopwatch() as stopWatch:
self.logInfo('Begin training...')
self._timer = self.ThreadManager.newTimer(
interval=0.25, func=self.trainingStatus)
tempTrainingData = Path('/tmp/snipsNLU')
if tempTrainingData.exists():
shutil.rmtree(tempTrainingData)
training: CompletedProcess = self.Commons.runSystemCommand([
f'./venv/bin/snips-nlu', 'train',
str(datasetFile),
str(tempTrainingData)
])
if training.returncode != 0:
self.logError(
f'Error while training Snips NLU: {training.stderr.decode()}'
)
assistantPath = Path(
self.Commons.rootDir(),
f'trained/assistants/{self.LanguageManager.activeLanguage}/nlu_engine'
)
if not tempTrainingData.exists():
self.trainingFailed()
if not assistantPath.exists():
self.logFatal('No NLU engine found, cannot start')
self._timer.cancel()
return
if assistantPath.exists():
shutil.rmtree(assistantPath)
shutil.move(tempTrainingData, assistantPath)
self._timer.cancel()
self.MqttManager.publish(constants.TOPIC_NLU_TRAINING_STATUS,
payload={'status': 'done'})
self.WebUINotificationManager.newNotification(
typ=UINotificationType.INFO,
notification='nluTrainingDone',
key='nluTraining')
self.ThreadManager.getEvent('TrainAssistant').clear()
self.logInfo(f'Snips NLU trained in {stopWatch} seconds')
self.broadcast(method=constants.EVENT_NLU_TRAINED,
exceptions=[constants.DUMMY],
propagateToSkills=True)
self.NluManager.restartEngine()
except:
self.trainingFailed()
finally:
self.NluManager.training = False
def nluTrainingThread(self, datasetFile: Path):
with Stopwatch() as stopWatch:
self.logInfo('Begin training...')
self._timer = self.ThreadManager.newTimer(interval=10,
func=self.trainingStatus)
tempTrainingData = Path('/tmp/snipsNLU')
if tempTrainingData.exists():
shutil.rmtree(tempTrainingData)
training: CompletedProcess = self.Commons.runSystemCommand([
f'./venv/bin/snips-nlu', 'train',
str(datasetFile),
str(tempTrainingData)
])
if training.returncode != 0:
self.logError(
f'Error while training Snips NLU: {training.stderr.decode()}'
)
assistantPath = Path(
self.Commons.rootDir(),
f'trained/assistants/assistant_{self.LanguageManager.activeLanguage}/nlu_engine'
)
if not tempTrainingData.exists():
self.logError('Snips NLU training failed')
if not assistantPath.exists():
self.logFatal('No NLU engine found, cannot start')
self._timer.cancel()
return
if assistantPath.exists():
shutil.rmtree(assistantPath)
tempTrainingData.rename(assistantPath)
self.broadcast(method=constants.EVENT_NLU_TRAINED,
exceptions=[constants.DUMMY],
propagateToSkills=True)
self.SnipsServicesManager.runCmd(cmd='restart',
services=['snips-nlu'])
self._timer.cancel()
self.ThreadManager.getEvent('TrainAssistant').clear()
self.logInfo(f'Snips NLU trained in {stopWatch} seconds')
def __init__(self, restartHandler: callable):
Singleton.__init__(self, self.NAME)
self.logInfo('Starting up Project Alice')
self._booted = False
with Stopwatch() as stopWatch:
self._restart = False
self._restartHandler = restartHandler
self._superManager = SuperManager(self)
self._superManager.initManagers()
self._superManager.onStart()
if self._superManager.configManager.getAliceConfigByName('useSLC'):
subprocess.run(
['sudo', 'systemctl', 'start', 'snipsledcontrol'])
self._superManager.onBooted()
self.logInfo(f'- Started Project Alice in {stopWatch} seconds')
self._booted = True
def decodeStream(self, session: DialogSession) -> Optional[ASRResult]:
super().decodeStream(session)
result = None
counter = 0
with Stopwatch() as processingTime:
with Recorder(self._timeout, session.user,
session.deviceUid) as recorder:
self.ASRManager.addRecorder(session.deviceUid, recorder)
self._recorder = recorder
self._decoder.start_utt()
inSpeech = False
for chunk in recorder:
if self._timeout.isSet():
break
self._decoder.process_raw(chunk, False, False)
hypothesis = self._decoder.hyp()
if hypothesis:
counter += 1
if counter == 10:
self.partialTextCaptured(session,
hypothesis.hypstr,
hypothesis.prob,
processingTime.time)
counter = 0
if self._decoder.get_in_speech() != inSpeech:
inSpeech = self._decoder.get_in_speech()
if not inSpeech:
self._decoder.end_utt()
result = self._decoder.hyp() if self._decoder.hyp(
) else None
break
self.end()
return ASRResult(
text=result.hypstr.strip(),
session=session,
likelihood=self._decoder.hyp().prob,
processingTime=processingTime.time) if result else None
def nluTrainingThread(self, datasetFile: Path):
with Stopwatch() as stopWatch:
self.logInfo('Begin training...')
tempTrainingData = Path('/tmp/snipsNLU')
if tempTrainingData.exists():
shutil.rmtree(tempTrainingData)
self.Commons.runSystemCommand([
f'./venv/bin/snips-nlu', 'train',
str(datasetFile),
str(tempTrainingData)
])
assistantPath = Path(
self.Commons.rootDir(),
f'trained/assistants/assistant_{self.LanguageManager.activeLanguage}/nlu_engine'
)
if not tempTrainingData.exists():
self.logError('Snips NLU training failed')
if not assistantPath.exists():
self.logFatal('No NLU engine found, cannot start')
return
if assistantPath.exists():
shutil.rmtree(assistantPath)
tempTrainingData.rename(assistantPath)
self.broadcast(method=constants.EVENT_NLU_TRAINED,
exceptions=[constants.DUMMY],
propagateToSkills=True)
self.SnipsServicesManager.runCmd(cmd='restart',
services=['snips-nlu'])
self.logInfo(f'Snips NLU trained in {stopWatch} seconds')
