def inference(self,
test_path: str,
model_path: str,
cuda: bool,
precision: int,
lm_config: LMConfig):
# Select one file from our test manifest to run inference
if os.path.isdir(test_path):
file_path = next(Path(test_path).rglob('*.wav'))
else:
with open(test_path) as f:
# select a file to use for inference test
manifest = json.load(f)
file_name = manifest['samples'][0]['wav_path']
directory = manifest['root_path']
file_path = os.path.join(directory, file_name)
transcribe_cfg = TranscribeConfig(
model=ModelConfig(
cuda=cuda,
model_path=model_path,
precision=precision
),
lm=lm_config,
audio_path=file_path
)
transcribe(transcribe_cfg)
def inference(self, test_manifest: str, model_path: str, cuda: bool,
use_half: bool, lm_config: LMConfig):
# Select one file from our test manifest to run inference
with open(test_manifest) as f:
file_path = next(f).strip().split(',')[0]
transcribe_cfg = TranscribeConfig(model=ModelConfig(
cuda=cuda, model_path=model_path, use_half=use_half),
lm=lm_config,
audio_path=file_path)
transcribe(transcribe_cfg)
def hydra_main(cfg: TranscribeConfig):
transcribe(cfg=cfg)
def hydra_main(cfg: TranscribeConfig):
transcript, meta = transcribe(cfg=cfg)
print("\n Output transcript : \033[94m", transcript, "\033[0m")
def hydra_main(cfg: TranscribeConfig):
transcript, transcribeGreedy, meta = transcribe(cfg=cfg)
print("\n Output transcript with LM : \033[94m", transcript, "\033[0m")
print("\n Output transcript with greedy : \033[94m", transcribeGreedy,
"\033[0m")
help='Audio file to predict on')
arg_parser.add_argument('--offsets',
dest='offsets',
action='store_true',
help='Returns time offset information')
arg_parser = add_decoder_args(arg_parser)
args = arg_parser.parse_args()
device = torch.device("cuda" if args.cuda else "cpu")
model = load_model(device, args.model_path, args.half)
decoder = load_decoder(decoder_type=args.decoder,
labels=model.labels,
lm_path=args.lm_path,
alpha=args.alpha,
beta=args.beta,
cutoff_top_n=args.cutoff_top_n,
cutoff_prob=args.cutoff_prob,
beam_width=args.beam_width,
lm_workers=args.lm_workers)
spect_parser = SpectrogramParser(audio_conf=model.audio_conf,
normalize=True)
decoded_output, decoded_offsets = transcribe(audio_path=args.audio_path,
spect_parser=spect_parser,
model=model,
decoder=decoder,
device=device,
use_half=args.half)
print(json.dumps(decode_results(decoded_output, decoded_offsets)))