Saw a live demo of OpenAI's new real-time voice API at a meetup in Chicago last night

Tbh I wasn't expecting much, just another slide deck. But they showed a guy talking to an AI that interrupted him mid sentence and it responded with actual emotion in its voice. The latency was basically zero, like talking to another person on the phone. Anyone else mess with voice AI yet or am I late to this?

2 comments

2 Comments

oliver_mitchell29d ago

That real-time interruption thing is wild when you experience it firsthand. I messed with a similar voice API from ElevenLabs a few months back for a side project, and what got me was how it handled natural pauses and filler words like "um" and "uh" without having to restart. The emotion part is the clincher though, especially if it matched the context of what you were saying. What stuck with me is how much faster it felt compared to stuff like Siri or Alexa, like you didn't have to wait for it to process every single sentence.

christopher_flores4629d ago

Actually the ElevenLabs voice API doesn't handle interruptions natively, that's more of a custom thing you'd build on top of it. Their realtime stuff is fast but you'd still need your own logic to detect when someone cuts in.