Agreed, and that's where it seems having lots of experience working with speech data helps more than trying to brute-force it with just larger CTC models and "m

Agreed, and that's where it seems having lots of experience working with speech data helps more than trying to brute-force it with just larger CTC models and "more" data of dubious quality.