I think what he's basically saying is that priors (i.e. domain knowledge + custom, domain-inspired models) help when you're data limited or when your data is ve

I think what he's basically saying is that priors (i.e. domain knowledge + custom, domain-inspired models) help when you're data limited or when your data is very biased, but once that's not the case (e.g. we have an infinite supply of voice samples), model capacity is usually all that matters.