def run_transcribe(audio_path: str, spect_parser: SpectrogramParser,
model: DeepSpeech, decoder: Decoder, device: torch.device,
use_half: bool):
# audio_path
# try:
# # inTranscript = audio_path.replace("wav", "txt")
# # print(inTranscript)
# # getTranscript(inTranscript)
# pass
# except Exception as asd:
# print(asd)
# pass
spect = spect_parser.parse_audio(audio_path).contiguous()
spect = spect.view(1, 1, spect.size(0), spect.size(1))
spect = spect.to(device)
if use_half:
spect = spect.half()
input_sizes = torch.IntTensor([spect.size(3)]).int()
out, output_sizes = model(spect, input_sizes)
decoded_output, decoded_offsets = decoder.decode(out, output_sizes)
#Thêm vào greedy
decoder2 = GreedyDecoder(labels=model.labels,
blank_index=model.labels.index('_'))
decoded_output2, decoded_offsets2 = decoder2.decode(out, output_sizes)
return decoded_output, decoded_output2, decoded_offsets, decoded_offsets2
def run_transcribe(audio_path: str, spect_parser: SpectrogramParser,
model: DeepSpeech, decoder: Decoder, device: torch.device,
precision: int):
spect = spect_parser.parse_audio(audio_path).contiguous()
spect = spect.view(1, 1, spect.size(0), spect.size(1))
spect = spect.to(device)
input_sizes = torch.IntTensor([spect.size(3)]).int()
with autocast(enabled=precision == 16):
out, output_sizes = model(spect, input_sizes)
decoded_output, decoded_offsets = decoder.decode(out, output_sizes)
return decoded_output, decoded_offsets
def run_transcribe(audio_path: str, spect_parser: SpectrogramParser,
model: DeepSpeech, decoder: Decoder, device: torch.device,
use_half: bool):
spect = spect_parser.parse_audio(audio_path).contiguous()
spect = spect.view(1, 1, spect.size(0), spect.size(1))
spect = spect.to(device)
if use_half:
spect = spect.half()
input_sizes = torch.IntTensor([spect.size(3)]).int()
out, output_sizes = model(spect, input_sizes)
decoded_output, decoded_offsets = decoder.decode(out, output_sizes)
return decoded_output, decoded_offsets