Spleeter is an open-source music source-separation library released by the streaming service Deezer in November 2019. It takes a finished recording and splits it into stems - the separate instrument and vocal tracks - using pretrained deep neural networks built on TensorFlow. Deezer shipped three ready-to-use models: a 2-stem split into vocals and accompaniment, a 4-stem split into vocals, drums, bass, and other, and a 5-stem split that also isolates piano.
Spleeter’s appeal was speed and accessibility. Deezer reported it could separate a track into 4 stems about 100 times faster than real time on a GPU, and it installed with a single pip command and downloaded its models automatically. That combination made high-quality separation, which had previously required research expertise, available to any hobbyist, and the project rapidly gained tens of thousands of stars on GitHub.
Why business readers should care: Spleeter was a turning point for stem separation as a commodity. By giving away a fast, good-enough tool, Deezer seeded a wave of karaoke apps, remix tools, and DJ software, and demonstrated how a streaming company could shape an ecosystem by open-sourcing internal research.