async def recognize(websocket, path):
rec = None
audio = b''
sample_rate = model.GetSampleFrequecy() # get default sample frequency
metadata = worker.METADATA
while True:
try:
data = await websocket.recv()
# Load configuration if provided
if isinstance(data, str) and 'config' in data:
jobj = json.loads(data)['config']
if 'sample_rate' in jobj:
sample_rate = float(jobj['sample_rate'])
if 'metadata' in jobj:
metadata = bool(jobj['metadata'])
continue
# Create the recognizer, word list is temporary disabled since not every model supports it
if not rec:
rec = KaldiRecognizer(model, spkModel, sample_rate,
worker.ONLINE)
if not isinstance(data, str):
audio = audio + data
response, stop = process_chunk(rec, data)
await websocket.send(response)
if stop:
if metadata:
obj = rec.GetMetadata()
data = json.loads(obj)
response = worker.process_metadata(data, audio)
await websocket.send(response)
break
except Exception as e:
break
def transcribe():
try:
worker.log.info('[%s] New user entry on /transcribe' %
(strftime("%d/%b/%d %H:%M:%S", gmtime())))
is_metadata = False
# get response content type
if request.headers.get('accept').lower() == 'application/json':
is_metadata = True
elif request.headers.get('accept').lower() == 'text/plain':
is_metadata = False
else:
raise ValueError('Not accepted header')
# get input file
if 'file' in request.files.keys():
file = request.files['file']
worker.getAudio(file)
rec = KaldiRecognizer(model, spkModel, worker.rate, worker.ONLINE)
rec.AcceptWaveform(worker.data)
data_ = rec.FinalResult()
confidence = rec.uttConfidence()
if is_metadata:
data_ = rec.GetMetadata()
data = worker.get_response(data_, confidence, is_metadata)
worker.clean()
else:
raise ValueError('No audio file was uploaded')
return data, 200
except ValueError as error:
return str(error), 400
except Exception as e:
worker.log.error(e)
return 'Server Error', 500
