ive found that it performs basically the same for conversational use cases without the audio input. the main benefit has just been the steerability of the voices and their realism compared to 11labs and everyone else