def run(self, model, lmbin, trie):
model = Model(model, BEAM_WIDTH)
if lmbin:
model.enableDecoderWithLM(lmbin, trie, LM_ALPHA, LM_BETA)
stream = None
while True:
# Try to get the next command from our queue, use a timeout to check
# periodically for a quit signal so the application doesn't hang on
# exit.
try:
cmd, *data = self._in_queue.get(timeout=0.3)
except queue.Empty:
if self._should_quit:
break
# If we haven't received a quit signal just continue trying to
# get a command from the queue indefinitely
continue
if cmd == 'start':
# 'start' means create a new stream
stream = model.createStream()
elif cmd == 'data':
# 'data' means we received more audio data from the recorder
if stream:
model.feedAudioContent(
stream, np.frombuffer(data[0].data(), np.int16))
elif cmd == 'finish':
# 'finish' means the caller wants the result of the current stream
transcript = model.finishStream(stream)
self.finished.emit(transcript)
stream = None
class InferenceThread(QObject):
finished = Signal(str)
def __init__(self):
super(InferenceThread, self).__init__()
self.in_queue = queue.Queue()
self.should_quit = False
self.worker = threading.Thread(target=self.run)
def send_cmd(self, cmd):
''' Insert command in queue to be processed by the thread '''
self.in_queue.put(cmd)
def setQuit(self):
''' Signal to the thread that it should stop running '''
self.should_quit = True
def start(self):
self.worker.start()
def run(self):
# Creating the model
self.model = Model( os.path.join(os.path.dirname(__file__),
"deepspeech-0.6.1-models/output_graph.pbmm"),
BEAM_WIDTH)
self.model.enableDecoderWithLM( os.path.join(os.path.dirname(__file__),
"deepspeech-0.6.1-models/lm.binary"),
os.path.join(os.path.dirname(__file__),
"deepspeech-0.6.1-models/trie"), LM_ALPHA, LM_BETA)
stream = None
while True:
# Try to get the next command from our queue, use a timeout to check
# periodically for a quit signal so the application doesn't hang on
# exit.
try:
cmd, *data = self.in_queue.get(timeout=0.3)
except queue.Empty:
if self.should_quit:
break
# If we haven't received a quit signal just continue trying to
# get a command from the queue indefinitely
continue
if cmd == "start":
# "start" means create a new stream
stream = self.model.createStream()
logging.debug("Starts to process sound")
elif cmd == "data":
# "data" means we received more audio data from the recorder
if stream:
self.model.feedAudioContent(stream,
np.frombuffer(data[0].data(), np.int16))
elif cmd == "finish":
# "finish" means the caller wants the result of the current stream
transcript = self.model.finishStream(stream)
self.finished.emit(transcript)
stream = None
logging.debug("Finishes to process sound")
def main():
parser = argparse.ArgumentParser(
description='Running DeepSpeech inference.')
parser.add_argument('--model',
required=True,
help='Path to the model (protocol buffer binary file)')
parser.add_argument(
'--alphabet',
required=True,
help=
'Path to the configuration file specifying the alphabet used by the network'
)
parser.add_argument('--lm',
nargs='?',
help='Path to the language model binary file')
parser.add_argument(
'--trie',
nargs='?',
help=
'Path to the language model trie file created with native_client/generate_trie'
)
parser.add_argument('--audio1',
required=True,
help='First audio file to use in interleaved streams')
parser.add_argument('--audio2',
required=True,
help='Second audio file to use in interleaved streams')
args = parser.parse_args()
ds = Model(args.model, N_FEATURES, N_CONTEXT, args.alphabet, BEAM_WIDTH)
if args.lm and args.trie:
ds.enableDecoderWithLM(args.alphabet, args.lm, args.trie, LM_ALPHA,
LM_BETA)
fin = wave.open(args.audio1, 'rb')
fs1 = fin.getframerate()
audio1 = np.frombuffer(fin.readframes(fin.getnframes()), np.int16)
fin.close()
fin = wave.open(args.audio2, 'rb')
fs2 = fin.getframerate()
audio2 = np.frombuffer(fin.readframes(fin.getnframes()), np.int16)
fin.close()
stream1 = ds.setupStream(sample_rate=fs1)
stream2 = ds.setupStream(sample_rate=fs2)
splits1 = np.array_split(audio1, 10)
splits2 = np.array_split(audio2, 10)
for part1, part2 in zip(splits1, splits2):
ds.feedAudioContent(stream1, part1)
ds.feedAudioContent(stream2, part2)
print(ds.finishStream(stream1))
print(ds.finishStream(stream2))
