Clarity

Clarity-1 is aimed at a very specific headache: live calls where your voice agent can’t tell the caller from the room around them. The model strips out background noise and other people talking, then sends the cleaned audio onward in real time. It’s pitched as target speaker extraction, which sounds technical because it is, but the promise is simple enough: the agent hears the right person, not the whole scene.

The timing matters. Clarity-1 streams as audio arrives, so it’s meant for live conversations rather than after-the-fact cleanup. That puts it squarely in the plumbing layer for voice agents, where a few messy seconds can ruin the whole exchange. And because it works on live calls, the “before/after” demo lives on the company’s site instead of in a neat marketing diagram.

Product Hunt says this is the second launch from KugelAudio. That’s a small detail, but it hints that the team is still building out a focused stack around real-time voice tools rather than chasing a broad product line. The launch page frames Clarity-1 as free, which should help it get tested quickly by people building on top of call flows and assistants.

There’s a lot of flashy talk around voice AI, but cleaning up the signal is where the work actually is. If a model can’t hear the caller properly, nothing upstream matters much. This is the sort of unglamorous feature that quietly earns its keep.