Actually Spleeter is MIT-Licensed so you are really free to use it in any way you want. If you are a music hacker and want to build something awesome using Spleeter, then go ahead. Well, you should probably reconsider and try Spleeter. If you’re a researcher working on Music Information Retrieval and have always considered that source separation artifacts made it unsuitable as a pre-processing step in your pipeline. If you are running the GPU version you can expect separating 100x faster than real-time which makes it a good option to process large datasets. That’s how we know that Spleeter performances match those of the best proposed algorithms.Īdditionally, Spleeter is very fast. To keep track, people have been comparing their algorithm in international evaluation campaigns. The pace of progress has recently made some giant leaps, mainly due to advances in machine learning methods. If you’re interested in this fascinating journey you should go read this literature overview, or this one. The challenge is thus to approximate them the best we can, that is to say as close as possible to the originals without creating too much distortions.įor years, a lot of strategies have been explored, by dozens of brilliant research teams from all over the world. In many cases, it may not be possible to exactly recover the individual tracks that have been mixed together. Yet that’s not really separation, you still hear all the other parts.
Just focus on one of the instrument of this track (say the lead vocal for instance) and you will be able to hear it quite distinctively from the others. Interestingly, our brain is very good at isolating instruments.
The task of music source separation is: given a mix can we recover these separate tracks (sometimes called stems)? This has many potential applications: think remixes, upmixing, active listening, educational purposes, but also pre-processing for other tasks such as transcription.įrom a Mix of many instruments, a source separation engine like Spleeter outputs a set of individual tracks or stems. It starts from a simple observation: music recordings are usually a mix of several individual instrument tracks (lead vocal, drums, bass, piano etc.). While not a broadly known topic, the problem of source separation has interested a large community of music signal researchers for a couple of decades now. Spleeter will be presented and live-demoed at the 2019 ISMIR conference in Delft. It comes in the form of a Python Library based on Tensorflow, with pretrained models for 2, 4 and 5 stems separation. We are releasing Spleeter to help the research community in Music Information Retrieval (MIR) leverage the power of a state-of-the-art source separation algorithm.