Krasp

Krasp is a simple noise-cancellation app for your microphone on macOS.

It works like a lightweight, open-source Krisp-style microphone filter: choose your real microphone in Krasp, turn noise cancellation on, and then select Krasp Microphone in Zoom, Discord, Google Meet, OBS, QuickTime, or any other app.

Krasp is built for a focused idea:

Your microphone
      |
      v
Krasp removes background noise
      |
      v
Apps hear Krasp Microphone

The processing runs locally on your Mac. Krasp does not send microphone audio to a cloud service.

What It Does

• Removes background noise from your microphone before other apps hear it.
• Creates a virtual microphone named Krasp Microphone.
• Lives quietly in the macOS menu bar.
• Lets you choose the physical input microphone.
• Lets you adjust how strong the noise cancellation should be.
• Shows simple input and reduction meters while it is running.
• Lets you listen through headphones and compare the original microphone with the cleaned output.

To listen, enable noise cancellation and click the headphone button. Switch between Cleaned and Original to hear the difference. This comparison only changes what you hear; other apps still receive the cleaned microphone. Click the headphone button again to stop. Playback uses your current macOS sound output, so use headphones to avoid feedback.

Why

Calls, streams, and recordings often pick up fans, keyboard noise, room echo, and other background sound. Krasp is meant to be a small macOS utility that sits between your real microphone and your apps, cleaning the audio path without needing a full audio-routing setup.

It is not trying to be a studio suite. It is a practical microphone noise-cancellation switch for everyday calls and recordings.

Current Status

Krasp is early-stage macOS audio software. The source is ready for public development, and GitHub Actions builds a macOS installer package when app-affecting files change on master.

The current release artifacts are unsigned developer builds. A polished public release still needs Developer ID signing and Apple notarization.

How It Works

Krasp has two parts:

• Krasp.app: the menu-bar app that captures your selected microphone and applies noise cancellation.
• KraspHAL.driver: the virtual microphone driver that exposes the cleaned audio as Krasp Microphone.

Krasp uses DPDFNet for neural speech enhancement at 48 kHz. If the model or runtime cannot load, the app shows an error and stops processing.

Suppression uses a decibel scale: 25% allows up to 15 dB reduction, 50% up to 30 dB, and 75% up to 45 dB. At 100%, you hear the fully processed signal. These are suppression limits, not guaranteed noise reduction. At 0%, the original audio passes through with the same processing delay.

More detail is in Architecture.

Requirements

• macOS 14 or newer.
• Xcode command-line tools with Swift 6.1 support.
• Internet access for the first neural build, which downloads checksum-pinned DPDFNet and sherpa-onnx assets.

Build

Build the app:

make app

Run it:

make run

Build a macOS installer package:

make dist

Artifacts are written to dist/.

Install The Virtual Microphone

1. Build and launch Krasp with make run.
2. Click Install in the Virtual Microphone row.
3. Approve the macOS administrator prompt.
4. Select Krasp Microphone as the input device in your call, recording, or streaming app.

The app installs its embedded HAL driver to:

/Library/Audio/Plug-Ins/HAL/KraspHAL.driver

and restarts CoreAudio so macOS can discover the virtual microphone.

Release

GitHub Actions builds and publishes a GitHub prerelease when app-affecting files change on master. Pushing a tag like v2026.1.1 also creates a tagged prerelease with the .pkg and checksum attached.

Release workflow notes are in Release.

Third-Party Components

• DPDFNet model by Ceva, Apache-2.0.
• sherpa-onnx runtime and C API, Apache-2.0.
• ONNX Runtime, MIT.

See Third Party Notices.

License

Krasp source code is released under the MIT License. See LICENSE.