search

hirofumi0810 / neural_sp

End-to-end ASR/LM implementation with PyTorch
Apache License 2.0
596 stars 141 forks source link

quetion about the loss and grad of "mbr" #322

Open Cescfangs opened 3 years ago

Cescfangs commented 3 years ago

https://github.com/hirofumi0810/neural_sp/blob/2b10b9cc4bdecb5180ecc45575c0ef410fb09aa3/neural_sp/models/seq2seq/decoders/las.py#L535-L548 I don't know much about mbr, according to these lines, it looks like a mWER loss and gradient to me

hirofumi0810 commented 3 years ago

@Cescfangs yes

Cescfangs commented 3 years ago

@Cescfangs yes

Thanks for the reply, and I'm curious about the improvement of this mWER tuning, say 5% relative wer reduction?

Cescfangs commented 3 years ago

https://github.com/hirofumi0810/neural_sp/blob/2b10b9cc4bdecb5180ecc45575c0ef410fb09aa3/neural_sp/models/criterion.py#L12-L39

Also, I am a little confused about the “mbr” loss, the inputs are not used in backward function, how does the grad flow to model params?