LALAL.AI has joined the music source separation game just recently but there’s no time to twiddle thumbs! We are ready to take user experience up a notch and improve the stem splitting results with the implementation of a new filter for audio processing.

Why It’s Needed

Music recordings are usually a mixture of several audio tracks (also known as stems), most commonly vocal and instrumental tracks. The main task of Lalal.ai and any other source isolation service is to precisely extract each stem. It’s quite challenging as mixed tracks often contain various sound engineering peculiarities and mastering errors that don’t become apparent until the stems are separated.

The usage of next-gen technologies (like neural networks) alone is not enough to accurately recover vocal and instrumental tracks. In order to reduce the amount of audible distortions, the application of different sound processing techniques is also required.

Since sound perception is relative — the same track may sound differently depending on a device and listener. The same person may even have contrasting experiences listening to the same track from different headphones or speakers.

Due to these factors, we couldn’t enhance the precision of individual stems produced by Lalal.ai with only internal changes. We needed to give our users the freedom to choose the sound processing filter that makes for the cleanest source separation as seen (or rather heard) from their own perspective.

How It Works

The new Processing filter has three levels: Mild, Normal, and Aggressive.

Mild is the minimum processing level. You can hear both the errors of the original track and the interpenetration of the separated tracks.

Normal is the default processing level, with moderate filtration applied. Most of the errors are corrected but some tracks may contain artificial elements and high frequencies blurring.

Aggressive is the maximum processing level. The filtering algorithms react to possible errors in the most sensitive way. It may result in some tracks’ natural mixing peculiarities being perceived as errors and thus removed. This level of processing can lead to even more noticeable artifacts.

Take control over the sound processing the following way:

Head over to Lalal.ai.
Select the filtration level at the Processing section.
Upload a track from your computer.
Wait a few moments until the track is processed.
Playback or download the separated stems.

Select another processing filter option if needed until you get the cleanest results. Enjoy!