GPT voice mode is not working

Maybe they cranked up the sampling or use a digital post-processor (like the API speed setting which doesn’t change the model).

To address a slow, plodding, answer style originally?