Using AI to separate songs into stems

There is already artificial intelligence that separates songs by tracks!

Audio source separation extracts components of a recording, such as vocals, drums, bass or piano. The available stems depend on the model: it is not always possible to isolate every instrument separately.

How does it work?

Anaconda is a Python distribution and a tool for managing environments; it does not separate instruments by itself. Separation requires a model and dedicated software, such as Deezer’s Spleeter, which offers two-, four- and five-stem models. This can help music students listen to individual parts or practise with an accompaniment.

This way it will be easier to hear what each musician is playing and it would be as if you were rehearsing with the rest of the musicians and you were playing your instrument. Another use to take into account is that of samples, since this way you can sample only the guitar of a song or the drums and exclude the rest of the instruments that are not wanted.

As a negative aspect, mention that the program is not 100% polished and some impurities are heard in the sound when separating the instruments and it is something that in the future we hope will be better optimized. Despite this, we are not going to detract from the work of the program since it is still really useful.

Another option mentioned in 2022 was to use subscription music libraries offering separate stems. These are different services from source-separation models, and their catalogues, licences and prices can change.

Let the music play!

What do you think? Perhaps you know someone learning an instrument or creating backing tracks. Although separation can introduce artefacts and does not necessarily recover the original studio tracks, it opens up interesting ways to listen and practise.

To learn how these tools work, visit our website and explore our artificial-intelligence courses.

References

Altozano, J. (Dec 6, 2019) We tested an Artificial Intelligence that separates a song BY TRACK (amazing!). https://www.youtube.com/watch?v=4_l31Vucrmo

Epidemicsound (March 2022). https://www.epidemicsound.com/

GitHub (March 2022). https://github.com/enterprise

Written by

Aitor Cossio

Saturdays.AI

Saturdays.AI is an impact-focused organization on a mission to empower diverse individuals to learn Artificial Intelligence in a collaborative and project-based way, beyond the conventional path of traditional education