Low latency, conversational voice foundation model

It's odd to use the word "voice" to describe a conversational text model.