ZFTurbo/Music-Source-Separation-Training

Health v2 · maintenance only

Adoption not part of the score

1512 stars · 219 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A Python training framework for music source separation models, supporting many architectures such as MDX23C, Demucs, Band Split RoFormer, and Mel-Band RoFormer. It provides configurable training and inference code for separating audio into stems like vocals, drums, and bass.

Use cases

• train a model to separate vocals from music
• split a song into stems like drums bass and vocals
• train a custom music source separation model
• remove vocals from a track with a deep learning model
• experiment with different audio separation architectures
• create karaoke versions of songs
• fine-tune a stem separation model on my own dataset

When to choose

• you want to train or fine-tune source separation models with many architecture options
• you need a flexible, experiment-friendly training codebase for audio separation
• you want to run inference with pretrained separation models like Mel-Band RoFormer

When to avoid

• you only need a ready-made GUI app to separate audio without training
• you need real-time separation in production rather than offline processing
• you work outside audio, e.g. image or text separation tasks

Facets

library · maturity active

machine-learning deep-learning audio-processing llm-training machine-learning deep-learning media python cross-platform music-source-separation audio-separation stem-separation pytorch model-training demucs roformer karaoke audio gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="ZFTurbo/Music-Source-Separation-Training")