def main():
def _abs_path(rel_path):
return os.path.join(os.path.dirname(__file__), '../..', rel_path)
parser = argparse.ArgumentParser()
parser.add_argument('--library_path',
help="absolute path to Cheetah's dynamic library",
type=str,
default=_abs_path('lib/linux/x86_64/libpv_cheetah.so'))
parser.add_argument('--acoustic_model_path',
help='absolute path to acoustic model parameter file',
type=str,
default=_abs_path('lib/common/acoustic_model.pv'))
parser.add_argument('--language_model_path',
help='absolute path to language model parameter file',
type=str,
default=_abs_path('lib/common/language_model.pv'))
parser.add_argument(
'--license_path',
help='absolute path to license file',
type=str,
default=_abs_path('resources/license/cheetah_eval_linux_public.lic'))
parser.add_argument(
'--audio_paths',
help='comma-separated absolute paths to audio files to be transcribed',
type=str,
required=True)
args = parser.parse_args()
cheetah = Cheetah(library_path=args.library_path,
acoustic_model_path=args.acoustic_model_path,
language_model_path=args.language_model_path,
license_path=args.license_path)
for audio_path in [
os.path.expanduser(x.strip()) for x in args.audio_paths.split(',')
]:
audio, sample_rate = soundfile.read(audio_path, dtype='int16')
if sample_rate != cheetah.sample_rate:
raise ValueError(
'Cheetah can only process audio data with sample rate of %d' %
cheetah.sample_rate)
num_frames = len(audio) // cheetah.frame_length
transcript = ''
for i in range(num_frames):
frame = audio[i * cheetah.frame_length:(i + 1) *
cheetah.frame_length]
partial_transcript, _ = cheetah.process(frame)
transcript += partial_transcript
transcript += cheetah.flush()
print(transcript)
def test_transcribe(self):
def abs_path(rel_path):
return os.path.join(os.path.dirname(__file__), '../..', rel_path)
cheetah = Cheetah(
library_path=abs_path('/home/nithin/project1/libpv_cheetah.so'),
acoustic_model_path=abs_path(
'/home/nithin/project1/acoustic_model.pv'),
language_model_path=abs_path(
'/home/nithin/project1/language_model.pv'),
license_path=abs_path(
'/home/nithin/project1/cheetah_eval_linux_public.lic'))
audio, sample_rate = soundfile.read(
abs_path('/home/nithin/project1/audio_samples/test.wav'),
dtype='int16')
assert sample_rate == cheetah.sample_rate
transcript = ''
num_frames = len(audio) // cheetah.frame_length
for i in range(num_frames):
frame = audio[i * cheetah.frame_length:(i + 1) *
cheetah.frame_length]
partial_transcript, _ = cheetah.process(frame)
transcript += partial_transcript
final_transcript = cheetah.flush()
transcript += final_transcript
self.assertEqual(
transcript,
"MISTER QUILTER IS THE APOSTLE OF THE MIDDLE CLASSES AND WE ARE GLAD TO WELCOME HIS GOSPEL"
)
def run(self):
cheetah = None
pa = None
audio_stream = None
try:
cheetah = Cheetah(
library_path=self._library_path,
acoustic_model_path=self._acoustic_model_path,
language_model_path=self._language_model_path,
license_path=self._license_path,
endpoint_duration_sec=1)
pa = pyaudio.PyAudio()
audio_stream = pa.open(
rate=cheetah.sample_rate,
channels=1,
format=pyaudio.paInt16,
input=True,
frames_per_buffer=cheetah.frame_length,
input_device_index=self._input_device_index)
while True:
pcm = audio_stream.read(cheetah.frame_length)
pcm = struct.unpack_from("h" * cheetah.frame_length, pcm)
if self._output_path is not None:
self._recorded_frames.append(pcm)
partial_transcript, is_endpoint = cheetah.process(pcm)
print(partial_transcript, end='', flush=True)
if is_endpoint:
print(cheetah.flush())
except Exception as e:
print(e)
finally:
if cheetah is not None:
print(cheetah.flush())
cheetah.delete()
if audio_stream is not None:
audio_stream.close()
if pa is not None:
pa.terminate()
if self._output_path is not None and len(self._recorded_frames) > 0:
recorded_audio = np.concatenate(self._recorded_frames, axis=0).astype(np.int16)
soundfile.write(self._output_path, recorded_audio, samplerate=cheetah.sample_rate, subtype='PCM_16')
parser.add_argument(
'--audio_paths',
help='comma-separated absolute paths to audio files to be transcribed',
type=str,
required=True)
args = parser.parse_args()
cheetah = Cheetah(library_path=args.library_path,
acoustic_model_file_path=args.acoustic_model_file_path,
language_model_file_path=args.language_model_file_path,
license_file_path=args.license_file_path)
for audio_path in [
os.path.expanduser(x.strip()) for x in args.audio_paths.split(',')
]:
audio, sample_rate = soundfile.read(audio_path, dtype='int16')
if sample_rate != cheetah.sample_rate:
raise ValueError(
'Cheetah can only process audio data with sample rate of %d' %
cheetah.sample_rate)
num_frames = len(audio) // cheetah.frame_length
for i in range(num_frames):
frame = audio[i * cheetah.frame_length:(i + 1) *
cheetah.frame_length]
cheetah.process(frame)
print(cheetah.transcribe())
type=str,
required=True)
args = parser.parse_args()
cheetah = Cheetah(library_path=args.library_path,
acoustic_model_path=args.acoustic_model_path,
language_model_path=args.language_model_path,
license_path=args.license_path)
for audio_path in [
os.path.expanduser(x.strip()) for x in args.audio_paths.split(',')
]:
audio, sample_rate = soundfile.read(audio_path, dtype='int16')
if sample_rate != cheetah.sample_rate:
raise ValueError(
'Cheetah can only process audio data with sample rate of %d' %
cheetah.sample_rate)
num_frames = len(audio) // cheetah.frame_length
transcript = ''
for i in range(num_frames):
frame = audio[i * cheetah.frame_length:(i + 1) *
cheetah.frame_length]
partial_transcript, _ = cheetah.process(frame)
transcript += partial_transcript
transcript += cheetah.flush()
print(transcript)