eschmidbauer/DL-Art-School
1
0
Fork 0
forked from mrq/DL-Art-School
Code Issues Pull Requests Packages Projects Releases Wiki Activity

w2v: new inference function

Browse Source
This commit is contained in:
James Betker 2022-02-21 19:13:03 -07:00
parent 896ac029ae
commit ba7f54c162
1 changed files with 5 additions and 0 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
codes/models/asr/w2v_wrapper.py
View File

@ -129,6 +129,11 @@ class Wav2VecWrapper(nn.Module):
pred = logits.argmax(dim=-1) pred = logits.argmax(dim=-1)
return [self.decode_ctc(p) for p in pred] return [self.decode_ctc(p) for p in pred]
def inference_logits(self, audio):
audio_norm = (audio - audio.mean()) / torch.sqrt(audio.var() + 1e-7)
logits = self.w2v(input_values=audio_norm.squeeze(1)).logits
return logits
@register_model @register_model
def register_wav2vec_feature_extractor(opt_net, opt): def register_wav2vec_feature_extractor(opt_net, opt):