How feasible is a lightweight AI-based noise gate plugin for OBS?

[hoodwink]

Hi everyone, I'm relatively new to OBS development but have some experience with Python and C++. I’ve been experimenting with OBS for personal streaming and noticed that while the built-in noise suppression filters are great, they sometimes cut off softer speech or let in background noise in unpredictable ways.

geometry dash​

I’ve been toying with the idea of building a lightweight plugin that uses a small, local AI model (possibly a distilled TensorFlow Lite or ONNX model) to intelligently detect human voice and apply a dynamic noise gate based on voice intent rather than just decibel level.
Before I dive too deep into this, I wanted to ask:

• Has anything similar already been attempted in the OBS plugin ecosystem?
• Are there performance concerns I should be aware of when embedding a small AI model inside an OBS plugin?
• Would it make more sense to develop this as a separate audio virtual device, or directly as an OBS filter?
• Finally, is there any interest from the community for something like this?

I’d love to contribute something genuinely useful, and this idea combines my love for audio tech and machine learning. Open to feedback, warnings, ideas, or anyone who might want to collaborate!

Thanks in advance!

 

[Fenrir]

Trying to use AI to solve a problem that is not only already solved, but is based on hard math for a reason, and has several plugins readily available in OBS sounds like waste of time. If the noise gate itself isn't working for you, I'd recommend making a post asking for assistance in configuring it, as it sounds like you might have it configured improperly. If you want to filter out background noise as well, there's several filter options for that.