just tested the new full duplex speech to speech model on my AI receptionist and the interruption handling alone is wild clip: https://youtube.com/shorts/6wk7j70dWVg?si=Bc9llSBp8XIwNOGd why it feels different from every voice bot you’ve tried: cascade bots chain three models: speech to text, LLM, text to speech. each step adds delay, and the bot has to guess when you’re done talking. that’s where the dead air and talking over you come from. turn-based speech to speech fixed a lot. one model, audio in and out, faster and more natural. but it still waits for you to finish before it speaks. cut in and it stumbles. full duplex doesn’t wait. it hears you while it’s talking, so a “wait, actually” or a quick “mhm” changes what it does in real time. nobody is waiting for a turn to end. that matters most for receptionists, because real callers ramble, interrupt, and change their minds mid-sentence. anyone else building on this yet? what are you using it for?