def __init__(self, model_file, ken_lm_file, context_length=10): self.listener = Listener(sample_rate=8000) self.model = torch.jit.load(model_file) self.model.eval().to('cpu') #run on cpu self.featurizer = get_featurizer(8000) self.audio_q = list() self.hidden = (torch.zeros(1, 1, 1024), torch.zeros(1, 1, 1024)) self.beam_results = "" self.out_args = None self.beam_search = CTCBeamDecoder(beam_size=100, kenlm_path=ken_lm_file) self.context_length = context_length * 50 # multiply by 50 because each 50 from output frame is 1 second self.start = False
def __init__(self, model_file): self.listener = Listener(sample_rate=8000, record_seconds=2) self.model = torch.jit.load(model_file) self.model.eval().to('cpu') #run on cpu self.featurizer = get_featurizer(sample_rate=8000) self.audio_q = list()