Spleeter: A Fast And State-of-the Art Music Source Separation Tool With Pre-trained Models

We present and release a new tool for music source separation with pre-trained models called Spleeter.Spleeter was designed with ease of use, separation performance and speed in mind. Spleeter is based onTensorflow [1] and makes it possible to:•separate audio files into2,4or5stems with a single command line using pre-trained models.•train source separation models or fine-tune pre-trained ones with Tensorflow (provided you have a dataset of isolated sources).The performance of the pre-trained models are very close to the published state of the art and is, to the authors knowledge, the best performing4stems separation model on the common musdb18 benchmark [6]to be publicly released. Spleeter is also very fast as it can separate a mix audio file into4stems100timesfaster than real-time1on a single Graphics Processing Unit (GPU) using the pre-trained4-stems model. Spleeter is packaged within Docker which makes it usable as is on various platforms.

PDF Abstract

Datasets


Results from the Paper


Ranked #18 on Music Source Separation on MUSDB18 (using extra training data)

     Get a GitHub badge
Task Dataset Model Metric Name Metric Value Global Rank Uses Extra
Training Data
Benchmark
Music Source Separation MUSDB18 Spleeter (MWF) SDR (vocals) 6.86 # 17
SDR (drums) 6.71 # 17
SDR (other) 4.02 # 22
SDR (bass) 5.51 # 19
SDR (avg) 5.91 # 18

Methods