Having a hard time getting a consistent speech style. I instruct the realtime api to answer with hesitations and giggles and this works for the first few questions, but after that it starts talking without this instruction…
Any tips on having continued speaking styles?