Somshubra Majumdar

I'm on vacation so I just saw this. Does Wav2Vec do better for your case without training ? What domain of speech are you trying to train / eval on?

@borisfom could you take a look ?

You might need to use a smaller batch sizes of 4 and use grad accumulation instead. RNNT model takes much more memory than CTC

@borisfom

@borisfom please check

We have not yet added support for LM or confidence score in asr model using Hypothesis framework yet, they are placeholders for future support.

It's not straightforward to setup word confidence, especially for RNNT models. I don't have a particular reference at the moment to compute word confidence. Fyi @GNroy

@ericharper for approval

Please do not upload json or text data files

Please check the export section in Nemo documentation.