Hush came from the Weya AI team watching good voice agents fall apart in the real world, not because the AI was wrong but because a fan, a hallway, or a second voice in the room garbled the audio before the model ever heard it, so they built an open-source model that isolates the main speaker and strips everything else in real time, on a CPU, in any language.
🔥 Our Take: Hush is free, open-source noise cancellation that runs on a normal CPU, and it's built for people making voice AI agents, where one fan or one hallway in the background can wreck the whole thing. The company to think about here is Krisp, which built a paid business doing more or less this. Hush just gives it away. If it's actually as good as they claim, they say top five on Hugging Face's audio leaderboard, then cleaning up audio stops being something you pay for and turns into a free piece you drop into your stack.