def _init_vad_model(self):
model_path = self._cfg.diarizer.vad.model_path
if model_path.endswith('.nemo'):
self._vad_model = EncDecClassificationModel.restore_from(model_path)
logging.info("VAD model loaded locally from {}".format(model_path))
else:
if model_path not in get_available_model_names(EncDecClassificationModel):
logging.warning(
"requested {} model name not available in pretrained models, instead".format(model_path)
)
model_path = "vad_telephony_marblenet"
logging.info("Loading pretrained {} model from NGC".format(model_path))
self._vad_model = EncDecClassificationModel.from_pretrained(model_name=model_path)
self._vad_window_length_in_sec = self._cfg.diarizer.vad.window_length_in_sec
self._vad_shift_length_in_sec = self._cfg.diarizer.vad.shift_length_in_sec
self.has_vad_model_to_save = True
self.has_vad_model = True
def _init_vad_model(self):
if self._cfg.diarizer.vad.model_path.endswith('.nemo'):
self._vad_model = EncDecClassificationModel.restore_from(self._cfg.diarizer.vad.model_path)
self._vad_window_length_in_sec = self._cfg.diarizer.vad.window_length_in_sec
self._vad_shift_length_in_sec = self._cfg.diarizer.vad.shift_length_in_sec
self.has_vad_model_to_save = True
self.has_vad_model = True
else:
raise ValueError("vad.model_path should be a .json file or .nemo or a .ckpt model file")