roformer

Production-ready UVR5 CLI & Docker image. Run SOTA separation models (Roformer, SCNet, MDX, Demucs, VR Architecture) on headless GPU servers without dependency hell.

MLX-native audio stem separation for Apple Silicon (Demucs, MDX, MDXC, VR, Roformer).

This repository contains all public UVR models, with possibly a few private ones included.

Clean audiobooks from background music and noise using MDX-Net/Roformer AI models. Handles large files with automatic chunking.

群友让我优化一下UVR5的4个常用的模型的推理速度，以及运行时环境的大小

Split any song into vocals and instrumental, entirely on your own computer. Free, open source, works offline.

Remove vocals from any song: karaoke instrumental and isolated vocals. Demucs + FastAPI + Svelte.

Remove the electric guitar from any song to make a backing track (MelBand-Roformer)

Turn any track into a DJ stem file (.stem.mp4) for Mixxx with state-of-the-art open separation models

Local Stem AI — local GPU audio separation studio

VocalDrop is a free AI vocal isolator and remover for macOS, Windows and Linux that runs entirely on your own hardware. No upload, no watermark, no subscription.

Split a song into 13 true-level stems (vocals, drum-kit pieces, bass, keys, guitar) that sum back to the mix.

Separazione stem (Demucs/Roformer) + mixer di pratica, testi e accordatore, per Windows. | Stem separation + practice mixer, lyrics & tuner for Windows.

AI-powered desktop DAW - stem separation, chord detection, karaoke

Senior Year Capstone Project

Pull songs apart into vocals, drums and more on your own device, with the best available separation models

A from-scratch study of Rotary Position Embedding in Transformers, including implementation, ablation experiments, language modeling evaluation, and comparison with the original RoFormer paper.

Free, local tool that removes music and background noise from videos and keeps only the voice (BS-Roformer + Mel-Roformer DeNoise).

Local AI stem separation for music producers. Private processing, batch queue, WAV preview and DAW drag-out. Linux-first.