"Code LeverageX team on AssemblyAI - Voice Agent Hackathon Hackathon"
Team Idea
Voice agents cut people off. Not you, me, or anyone like us. But people who stutter, people who talk on cell phones, people who don't speak English as a first language. Anyone who pauses to find a word, perhaps. The root cause is dull. The default turn detection from AssemblyAI waits 1.28 seconds of silence before deciding you've ended. A stutter block lasts two or three. So the agent starts talking, the person loses place, and the stutter almost always worsens from there. Earshot sits within the streaming session and calculates how each caller speaks -- how long they pause, how fast they speak, and how often they restart right after an endpoint. About twenty seconds in, it begins sending UpdateConfiguration and is pushing max_turn_silence near that person's p95 pause. People who talk quickly get snappier endpointing not slower. Per-caller, not deployment.
Team Members
theLeverager1705
theleverager1705742
Submission
Team Leader hasn't made a submission yet