What is Hush?

Hush is an open-source noise suppression model developed by weya AI that strips background noise, competing voices, and audio interference from real-time calls. Designed specifically for Voice AI pipelines, Hush cleans chaotic phone audio at the source, turning noisy input into clean, ASR-ready speech. It processes each 10 ms audio frame in under 1 ms on standard CPUs, keeping conversations flowing without latency. At just 8 MB, the model is lightweight enough to deploy in your own cloud or data center, and it ranked in the top 5 speech-enhancement models on Hugging Face’s Audio-to-Audio leaderboard at launch.

Who it's for

• Voice AI developers — building agents or bots that need reliable speech recognition from noisy real-world calls.
• Contact center teams — looking to improve agent and bot comprehension in environments with traffic, office buzz, or overlapping speakers.
• Compliance and quality assurance managers — who require clean audio recordings for accurate transcription and analysis.

Key features

Real-time CPU processing

Hush processes each 10 ms audio frame in under 1 ms on standard CPUs, so calls stay fast with no GPUs needed. This makes it practical to run at scale without expensive hardware upgrades.

Focuses on the right voice

The model isolates the main caller and pushes background talk, TV noise, and other competing voices aside. ASR systems hear the person who matters, reducing errors from overlapping speech.

Built for noisy places

Hush handles traffic, office buzz, fans, and street sounds, keeping calls understandable even in the worst everyday noise. It was trained on over 10,000 hours of real-world noisy audio, including overlapping speakers and tough environments.

What stands out

Hush fixes the call signal at the source, turning chaotic phone audio into clean, ASR-ready speech.