Your Favorite Character's Karaoke Career Starts Here

What it does

AICoverGen is a browser-based pipeline—plus a CLI for the stubborn—that takes a YouTube link or local audio file, isolates vocals with MDXNET, swaps in an RVC v2-trained AI voice, and remixes the result with the original instrumental. It is essentially a convenience wrapper around vocal separation, pitch extraction, and voice conversion, aimed at developers who want singing chatbots or fans who want to hear fictional characters cover pop songs.

The interesting bit

Instead of treating voice conversion as a black box, the project exposes a surprising amount of audio-engineering detail inside its WebUI. You can tune how much of the AI’s accent bleeds through, preserve breath and voiceless consonants, adjust reverb as if you were in a virtual room, and even shift the entire track’s key like a karaoke machine.

Key highlights

• Accepts YouTube URLs or local audio files and outputs finished covers in minutes, depending on your GPU.
• Built-in WebUI tabs for downloading public RVC models or uploading locally trained ones.
• Granular mixing controls: independent volume sliders for main vocals, backup vocals, and instrumentals; optional reverb; and pitch-shifting for the full mix.
• Choice of pitch detection algorithms (rmvpe for clarity, mangio-crepe for smoother vocals) with adjustable hop length.
• Can keep intermediate stems—isolated AI vocals and instrumentals—for remixing in a real DAW later.

Caveats

• The README explicitly warns that the WebUI is “under constant development and testing,” so the interface may shift or break.
• Requires Python 3.9 exactly; the README notes that other versions risk dependency conflicts.
• You must bring your own RVC v2 models; the tool is a pipeline, not a voice trainer.

Verdict

Worth a look if you already have RVC models and want a faster path from YouTube link to finished cover, or if you are building a VTuber and need a singing backend. Not what you want if you are hoping to train new voices from scratch.