EOU is the voice-AI term for end of utterance, the moment a speaker has finished their turn. EOUM.com reads as EOU model: the part of a voice system that decides when that moment has arrived. It’s four letters on a .com, and anyone building voice agents gets the meaning right away.
It’s also the hardest part of making a voice agent feel human. Answer too soon and the agent talks over people mid-thought. Wait too long and every exchange has an awkward gap. Simple silence timers get this wrong all the time, because people pause to think or say “um” right before the important part.
A small model with a big job
The product is a small, fast turn detector. It listens to the audio and the words together and predicts whether the speaker is done, in milliseconds and ideally on the device itself, so the agent can reply the instant a turn ends. A browser demo with a live microphone sells it in ten seconds: talk, pause mid-sentence, keep going, and watch it wait.
Voice agents are one of the hottest categories in AI. They answer phones for clinics and restaurants, handle customer support, take orders and tutor students. Every one of them lives or dies on turn-taking, and the infrastructure companies in the space have started shipping turn-detection models of their own. That’s how you know it’s core.
The buyer could be a voice-AI startup, a speech-recognition company adding turn detection to its API, a contact-center software vendor or a platform that wants a model brand separate from its main product. EOUM is short and exact, the kind of name a model gets called by in code.
Knowing when the other person has finished is half of conversation. Machines are only now learning it.