Reason behind increasing dep_q from 8 to 16, compared to Moshi #57
Description
Due diligence
- I have done my due diligence in trying to find the answer myself.
Topic
The PyTorch implementation
Question
Hi, I was going through your codebase and wanted to understand why the change of predicting more codebooks done in Personaplex? Is it for modelling the user audio, so during training is it trained to predict both the assistant and user turns?