I feel all of these models when running

[dapper hazel]

Have you tried playing with the prompt to make the conversation feel more natural? Perhaps prompting the agent to use filler words like uhm and ah? Wondering if this helps!

[graceful osprey]

I think that's exactly it. I've been fiddling with various levels and degrees and using the cues etc...but all of it is telling. Even the 'breath' at times is incorrect. And it breaks the immersion 😉

It's close, but it's too much "covering" of the latency with these types of 'tricks'