These are models specially tuned and sized for near real-time, instant translation. It would be naive to think that there aren't technical creatives building and training models tuned for expressiveness and nuance in a more controlled environment.