Doubts in Inference and Training

sooftware / conformer

[Unofficial] PyTorch implementation of "Conformer: Convolution-augmented Transformer for Speech Recognition" (INTERSPEECH 2020)

Apache License 2.0

958 stars 175 forks source link

Doubts in Inference and Training #29

Closed codemaster-22 closed 3 years ago

codemaster-22 commented 3 years ago

Hey why are we not maintaining something like blank token during Inference for each time step , and during training also won't there be some alignment problems ?

sooftware commented 3 years ago

Sorry for late response. I handle blank token when converts ids to tokens. like this.