Hush

What it does

Hush is an open-source noise suppression model developed by weya AI. It removes background noise, competing voices, and audio interference from real-time calls, producing clean speech for voice AI agents. The model processes each 10 ms audio frame in under 1 ms on standard CPUs, enabling real-time performance without GPUs. It isolates the main caller's voice and suppresses distractions like traffic, office buzz, and street sounds.

Who it is for

Hush is designed for developers and teams building voice AI agents, bots, and compliance systems that rely on clear audio input. The website lists logos of financial services companies (e.g., Hitachi, Kotak, Simpl, Cars24) as trusted users, suggesting adoption in BFSI (banking, financial services, and insurance) for call center and voice-based applications.

Why it matters

Most voice AI failures stem from poor audio quality rather than model accuracy. By cleaning the call signal at the source, Hush improves ASR (automatic speech recognition) accuracy, reduces misunderstandings, and enhances overall agent performance. The model is lightweight (8 MB) and can be deployed in the cloud or on-premises, making it accessible for real-world noisy environments.

Launch signal

Hush launched as a top-5 speech-enhancement model on Hugging Face's Audio-to-Audio leaderboard. It was trained on over 10,000 hours of real-world noisy audio, including overlapping speakers and challenging environments. The model is available on Hugging Face and GitHub under an open-source license.

Brand and naming

The name "Hush" evokes quieting noise, aligning with its purpose of silencing distractions. The branding is straightforward and memorable, positioning the product as a practical tool for developers. The open-source approach and focus on real-time CPU performance differentiate it from heavier, GPU-dependent alternatives.