def read_recognizer(inference_config):
model_path = Path(__file__).parent / 'pretrained' / inference_config.model
if inference_config.model == 'latest' and not model_path.exists():
download_model(inference_config)
assert model_path.exists(
), f"{inference_config.model} is not a valid model"
# create pm (pm stands for preprocess model: audio -> feature etc..)
pm = read_pm(model_path, inference_config)
# create am (acoustic model: feature -> logits )
am = read_am(model_path, inference_config)
# create lm (language model: logits -> phone)
lm = read_lm(model_path, inference_config)
return Recognizer(pm, am, lm, inference_config)
parser.add_argument('-i',
'--input',
type=str,
required=True,
help='specify your input wav file')
args = parser.parse_args()
# check file format
assert args.input.endswith(
'.wav'
), " Error: Please use a wav file. other audio files can be converted to wav by sox"
# download specified model automatically if no model exists
if len(get_all_models()) == 0:
download_model('latest')
# resolve model's name
model_name = resolve_model_name(args.model)
if model_name == "none":
print(
"Model ", model_name,
" does not exist. Please download this model or use an existing model in list_model"
)
exit(0)
args.model = model_name
# create recognizer
recognizer = read_recognizer(args)