Still in development but Here is the article I read. I'd imagine the output would be similar to a speech recognition device, stilted and electronic sounding with no context or intonation.