Stem Extraction

Samples: Stem extraction

As mentioned in The Process, stem extraction is the first step that is done before anything in terms of samples.

If you are lucky and your song's artist has graciously offered you stems through one way or another, congratulations! You can move forward to Stem Processing. If not, read below to learn how you can create your own, high quality stems for your projects.

To extract stems from a song, we will be using an application called Ultimate Vocal Remover 5, or UVR5 for short. UVR5 uses machine learning models such as Demucs and MDX to analyse the waveform of a sound and separate the individual components.

When you open up UVR5 for the first time, you'll need to download the models before you can start extracting. Head over to Settings → Download Centre → Demucs. The model I (and many other Launchpad Artists) use is "v4 | htdemucs_ft". I've personally found this model to be the most consistent and cleanest in-built model, but do try out other models to see which suits your usecase (or song genre) best.

To get the best performance out of UVR5, do enable "GPU Conversion", however reductions in processing time may vary from device to device. Point the input to your downloaded song, and set the output to a specific folder, as you will be extracting multiple stems. Once you are happy with your configuration, hit "Start Processing" and wait until UVR5 is done with extraction.

If your computer is on the slower side, you can use this web-based alternative called melody.ml. Open up this website on an incognito tab, enter in your email address (or a random email address, it does not matter), upload your song, and wait for it to finish processing. You will be able to download the stems shortly after.

Next section: Stem Processing