def init_module(args):
log('Initialize ASR ...')
global MODEL, transcriber
MODEL = args.model
asr = create_asr(MODEL)
frame_generator, vad = create_vad()
transcriber = Transcriber(asr, vad, frame_generator)
log('ASR worker initialized!')
def create_worker(model, hostname, port, master_address, recordings_saver_address):
poller = create_poller("tcp://0.0.0.0:5678")
heartbeat = create_heartbeat(model, "tcp://%s:%s" % (hostname, port), master_address)
asr = create_asr()
audio = AudioUtils()
saver = RemoteSaver(create_recordings_saver_socket(recordings_saver_address), model)
vad = create_vad()
id_generator = lambda: uuid.uuid4().int
run_forever = lambda: True
return Worker(poller, heartbeat, asr, audio, saver, vad, id_generator, run_forever)
def test_vad(self):
vad = create_vad(16000)
utterances = 0
for original_chunk, resampled_chunk in self.chunks():
_, change, _, _ = vad.decide(original_chunk, resampled_chunk)
if change == 'speech':
utterances += 1
self.assertEqual(3, utterances)
def test_vad(self):
vad = create_vad()
utterances = 0
for original_chunk, resampled_chunk in self.chunks():
_, change, _, _ = vad.decide(original_chunk, resampled_chunk)
if change == 'speech':
utterances += 1
self.assertEqual(3, utterances)
def create_worker(model, hostname, port, master_address,
recordings_saver_address):
poller = create_poller("tcp://0.0.0.0:5678")
heartbeat = create_heartbeat(model, "tcp://%s:%s" % (hostname, port),
master_address)
asr = create_asr()
audio = AudioUtils()
saver = RemoteSaver(
create_recordings_saver_socket(recordings_saver_address), model)
vad = create_vad()
id_generator = lambda: uuid.uuid4().int
run_forever = lambda: True
return Worker(poller, heartbeat, asr, audio, saver, vad, id_generator,
run_forever)
def callback(nbest, end_of_segment=False, end_of_stream=False, alignment=None, alignment_with_confidence=None, start=0.0):
if end_of_segment:
for (word, utterance_start, duration, confidence) in alignment_with_confidence:
f.write(u" ".join(["%.2f" % (start + utterance_start), "%.2f" % duration, word, "%.2f" % confidence, "\n"]))
return callback
if __name__ == '__main__':
if len(sys.argv) != 3:
print("Usage: transcribe.py model wav")
sys.exit(-1)
model = sys.argv[1]
wav = sys.argv[2]
asr = create_asr(model)
frame_generator, vad = create_vad()
transcriber = Transcriber(asr, vad, frame_generator)
start = time.time()
transcriber.run(chunks(wav), print_ctm(sys.stdout))
end = time.time()
processing_time = end - start
duration = wav_duration(wav)
print("%.2f" % duration, "%.2f" % processing_time,
"%.2f" % (processing_time / duration), file=sys.stderr)