CleanStream - Remove profanity and fillers in your live stream or recording with AI!

CleanStream - Remove profanity and fillers in your live stream or recording with AI! v0.2.0

Go to download
Overview Updates (9) History Discussion

v0.2.0

Maintenance release to fix compatibility with OBS v32​

This is not an indication that development is resuming on this plugin
Changes ported over from https://github.com/royshil/obs-localvocal, so they come with bonus acceleration improvements too

Note: AMD variants seem to be currently not working when using hipblas, and MacOS releases may have an issue with Metal on some systems

What's Changed​

  • Fix OBS 32 compatibility and improve acceleration by @Tabby in #34
    • Replace use of circlebuf with deque (renamed in OBS)
    • Update Whispercpp to v1.8.2
    • Update ONNX runtime to v1.20.1
    • Acceleration improvements for Linux and Windows
      • generic variants - Add Vulkan and OpenBLAS support (and OpenCL on Linux)
      • nvidia variants - Same as generic but also with CUDA support (and using CUDA OpenCL library on Linux)
      • amd variants - Same as generic but also with ROCm support via hipblas
    • Dynamic whisper backend loading to make maximum use of available CPU features and to support older CPUs
    • GPU selection options to choose backend to use (defaults to CPU only)


Download: https://github.com/royshil/obs-cleanstream/releases/tag/0.2.0

v0.1.3 - Upgrade Whisper, add fine-tuned models

Download

v0.1.2 - Bump Whisper ASR version

In this release:
  • Bump Whisper.cpp version
  • Bump onnxruntime version
  • Minor UI change
Download

v0.1.0 Bump OBS dependency to support v30.2

Bumping to OBS v30.1+! Fixing a number of bugs related to outdated OBS dependency.

Download

v0.0.9 Bump Whisper + Bugs and Fixes all around!

  • Bumping Whisper.cpp version and resolving a critical bug.
  • Fixing the adaptive buffer to allow any system (even without CUDA) and any model selection (even big ones) to still hit the words and blank them out in real-time.
Download

v0.0.6 - Complete overhaul! Robust, accurate, more HW support, external sounds...

Complete overhaul! It's like a new plugin - lean and mean
Reworked the architecture, made it more robust and accurate
Ability to determine the "replace sound" even with external files - makes for really fun and funny effects!

Enjoy!

  • Like
Reactions: lcalder

v0.0.4 - we are back.

I'm reviving this plugin. Please help me test it!
This release should bump the quality and stability by a whole lot. But it still needs work.
If you have any questions or problems, please reach out to me https://discord.gg/BaKh7qv7xH

v0.0.3 - adding support for ~100 languages!

100 Languages!

I've added the ability to transcribe in almost 100 languages tracking the language support from github.com/ggerganov/whisper.cpp
Also added a model-downloading UI so the plugin is small but any models can be pulled from the internet on-the-fly and used. Adding the bigger models too (base and small).

Enjoy! and give feedback if you can...

What's Changed​

  • Add model selection, downloader and language by @royshil in #8
Full Changelog: 0.0.2...0.0.3

v0.0.2 - new build system and new v of Whisper.cpp

This release introduces a new build system tracking the plugin template, as well as a new version of Whisper.cpp

If this plugin has provided value to you or you're exited about its potential, consider supporting my work: https://github.com/sponsors/royshil

What's Changed​

Top