Thanks @Mason. I have tried

model-output

and as far as I can tell these events are not received at the same time as the bot speaks them (rather, all the monologue parts are received before the bot starts speaking). Am I missing something?