obs-cleanstream

CleanStream is an OBS plugin that cleans live audio streams from unwanted words and utterances using real-time local AI.

If this free plugin has been valuable to you consider adding a ⭐ to this GH repo, rating it on OBS, subscribing to my YouTube channel where I post updates, and supporting my work on GitHub or Patreon 🙏

Check out our other plugins:

• Background Removal remove background (virtual green screen) from video
• Detect will detect and track >80 types of objects in any OBS source
• URL/API Source fetch API data and display it on screen as a video source
• LocalVocal speech AI assistant plugin for real-time transcription (captions), translation and more language functions
• PolyGlot a realtime local translation service based on AI.

Usage

• Add the plugin to any audio-generating source
• Adjust the settings

Download

Check out the latest releases for downloads and install instructions.

Method

This video walkthrough (YouTube) will explain various parts of the code if you're looking to learn from what I've discovered.

Audio processing

The filter is running Whisper in real-time to detect words in small chunks of the incoming audio. For each chunck it produces a decision which then determines if the audio rendering will play the original audio or e.g. a beep or silence. The processing happens in a separate thread and therefore there's a built-in lag/delay mechanism to make sure the audio decision (play, beep, silence) is in-sync with the actual audio playback based on the timestamp. The built-in delay is adaptive since some systems (e.g. with CUDA) can make faster decisions.

Here is an illustration of the process:

Requirements

• OBS version 30+ for plugin versions 0.0.4+
• OBS version 29 for plugin versions 0.0.2+
• OBS version 28 for plugin versions 0.0.1

We do not support older versions of OBS since the plugin is using newer APIs.

Introduction

CleanStream is an OBS plugin that cleans live audio streams from unwanted words and utterances, such as "uh"s and "um"s, and other words that you can configure, like profanity.

See our resource on the OBS Forums for additional information.

It is using a neural network (OpenAI Whisper) to predict in real time the speech and remove the unwanted words.

It's using the Whisper.cpp project from ggerganov to run the Whisper network in a very efficient way.

But it is working and you can try it out. Please report any issues you find. 🙏 (submit an issue or meet us on https://discord.gg/KbjGU2vvUz)

We're working on improving the plugin and adding more features. If you have any ideas or suggestions, please open an issue.

Building

The plugin was built and tested on Mac OSX, Windows and Ubuntu Linux. Help is appreciated in building on other OSs and packages.

The building pipelines in CI take care of the heavy lifting. Use them in order to build the plugin locally.

Start by cloning this repo to a directory of your choice.

Mac OSX

Using the CI pipeline scripts, locally you would just call the zsh script.

$ ./.github/scripts/build-macos.zsh -c Release -t macos-x86_64

Install

The above script should succeed and the plugin files will reside in the ./release folder off of the root. Copy the files to the OBS directory e.g. /Users/you/Library/Application Support/obs-studio/obs-plugins.

To get .pkg installer file, run

$ ./.github/scripts/package-macos.zsh -c Release -t macos-x86_64

(Note that maybe the outputs in the e.g. build_x86_64 will be in the Release folder and not the install folder like pakage-macos.zsh expects, so you will need to rename the folder from build_x86_64/Release to build_x86_64/install)

Linux (Ubuntu-ish)

Use the CI scripts again

$ ./.github/scripts/build-linux.sh

Windows

Use the CI scripts again, for example:

> .github/scripts/Build-Windows.ps1 -Configuration Release

The build should exist in the ./release folder off the root. You can manually install the files in the OBS directory.

> Copy-Item -Recurse -Force "release\Release\*" -Destination "C:\Program Files\obs-studio\"

Building with CUDA support on Windows

CleanStream will now build with CUDA support automatically through a prebuilt binary of Whisper.cpp from https://github.com/occ-ai/occ-ai-dep-whispercpp. The CMake scripts will download all necessary files.

To build with cuda add CPU_OR_CUDA as an environment variable (with cpu, 12.2.0 or 11.8.0) and build regularly

> $env:CPU_OR_CUDA="12.2.0"
> .github/scripts/Build-Windows.ps1 -Configuration Release

Star History