Spleeter
Deezer's open-source library for splitting audio into stems, 100x real-time on GPU.
About
Spleeter is Deezer's open-source audio source separation library with pretrained TensorFlow models. It splits a track into 2 stems (vocals/accompaniment), 4 stems (vocals, drums, bass, other), or 5 stems (adding piano), and separates 4-stem audio about 100x faster than real-time on a GPU.
It ships as a Python library, CLI, Docker image, and even a Google Colab notebook for trying it without installing anything. With over 28,000 GitHub stars it is the reference tool for vocal removal, karaoke tracks, remixing, and sample extraction. Separation quality is strong on the musdb benchmark but artifacts creep in on dense or unusual mixes, so results need a listen before release.
What it can do
- 2, 4, or 5-stem separation with pretrained models
- 100x faster than real-time on GPU
- CLI, Python library, and Docker image
- Free Colab notebook for no-install use
- Trainable on custom isolated-source datasets
Setup
pip install spleeter, or use the Docker image or Colab notebook. Requires ffmpeg and libsndfile.
Prerequisites
Python with TensorFlow, plus ffmpeg and libsndfile installed.
Use cases
Works with
Trust score — 7/10
Scored 7/10 (metrics tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 85; Trending score 74/100 from researcher assessment.
★ 28,130 GitHub stars
Sources
Every fact above was checked against these sources. Where a claim comes from the provider and couldn't be independently verified, the text says so.