I am trying to replicate the LibriSpeech recipe before making some changes.
I trained the Tokenzier successfully and moved to the LM part.
There is a note in the README.md file saying:
“Training an LM takes a lot of time. In our case, it takes 3/4 weeks on 4 TESLA V100. Use the pre-trained model to avoid training it from scratch”
I couldn’t find a way of using the pre-trained models that are shared via Google drive.
Where should I place the files?
the RNNLM.yaml (1k BPE) , in drive there is a folder named “1234”
and inside that directory, there are many files and a sub-directory named save.
Which file do I need?
In addition, the next step is training the ASR itself. How do I pick the desired LM pretrained model
to be used with the ASR engine?