Skip to content
hushAI
  • Listen
  • How it's built
  • Features
  • Compare
  • Benchmarks
  • FAQ
  • Docs
GitHubDownload
Documentation

Everything about Hush

Plain guides for everyone, and the full engineering story for the curious: how audio flows, how it was measured, and how to build it yourself.

Start with the user guide Source on GitHub

Start here

  • OverviewWhat Hush AI is, how to install it and how it works, on one page.12 min read
  • User guideInstall, set up, choose Hush AI Microphone in your apps, and fix problems.10 min read

How it works

  • ArchitectureHow audio flows through Hush: threads, queues, clock drift and the delay budget.8 min read
  • Engine (hush-core)The Rust engine on its own: stream a signal, live devices and examples.2 min read
  • Hush AI MicrophoneThe Hush AI Microphone drivers for macOS and Windows, and how they were tested.8 min read

Measurements

  • Quality comparisonDeepFilterNet 3, NVIDIA RE-USE and RNNoise on VoiceBank-DEMAND: the method, every metric and the caveats.10 min read
  • PerformanceCPU time per frame, threads, priority under load, long runs and delay.4 min read
  • AccuracyHush's CPU output against NVIDIA's GPU reference, and the regression gates in CI.3 min read

Project

  • Build and releaseDevelop, test and release the app, the engine and the drivers.3 min read
  • ContributingWays to help, and what a pull request needs.1 min read
  • SecurityHow to report a security problem, and what Hush does with your system.1 min read
  • WebsiteHow this website is built and previewed.1 min read
hushAI

Free, open-source noise cancellation for live calls and recordings. It runs on your own computer.

Star on GitHub

Product

  • Hear the difference
  • Features
  • Compare
  • Enhance a recording
  • Live Translate
  • Google Calendar
  • Benchmarks
  • How it's built
  • Download

Docs

  • All docs
  • Overview
  • User guide
  • Architecture
  • Engine (hush-core)
  • Hush AI Microphone
  • Quality comparison
  • Performance
  • Accuracy

Project

  • GitHub
  • Releases
  • Models on Hugging Face
  • Report an issue
  • Contributing
  • Security
  • Roadmap
  • FAQ

The Hush AI app and engine are licensed under Apache-2.0. DeepFilterNet 3 is MIT or Apache-2.0. NVIDIA RE-USE model license. Demo samples are credited next to the player.

Hush AI is an independent open-source project. Zoom, Google Meet, Google Calendar, Microsoft Teams, Slack, Discord, Webex, OBS Studio, WhatsApp, Telegram, Krisp, Adobe Podcast and VB-CABLE are trademarks of their owners, who do not endorse Hush AI. Their names and logos are used only to show compatibility or to compare.