Spaces:
Runtime error
Runtime error
File size: 984 Bytes
1554ec8 daeb975 1554ec8 daeb975 1554ec8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 |
import torch
import torchaudio
#fixes second prediction bug
torch._C._jit_override_can_fuse_on_cpu(False)
torch._C._jit_override_can_fuse_on_gpu(False)
torch._C._jit_set_texpr_fuser_enabled(False)
torch._C._jit_set_nvfuser_enabled(False)
loader = torch.jit.load("audio_loader.pt")
model = torch.jit.load('QuartzNet_thunderspeech_3.pt')
vocab = model.text_transform.vocab.itos
vocab[-1] = ''
def convert_probs(probs):
ids = probs.argmax(1)[0]
s = []
if vocab[ids[0]]: s.append(vocab[ids[0]])
for i in range(1,len(ids)):
if ids[i-1] != ids[i]:
new = vocab[ids[i]]
if new: s.append(new)
#return '.'.join(s)
return s
def predict(path):
audio = loader(path)
probs = model(audio, torch.tensor(audio.shape[0] * [audio.shape[-1]], device=audio.device))[0]
return convert_probs(probs)
gr.Interface(fn=predict,
inputs=[gr.inputs.Audio(source='microphone', type='filepath', optional=True)],
outputs= 'text').launch(debug=Tru |