So you want to sample something. You’ve got a digital file or maybe a record, because the Isley Brothers haven’t gotten back to you yet when you asked to borrow their masters for your sweet house banger.

What you did before was you put on your studio headphones and you wrote down time codes of what your ears told you were the “cleanest” potential samples in your source material. Sometimes you got unlucky and what sounded clear to you had some bleed from another instrument when you chopped it up. (Sometimes you also got sued.)

LALAL.AI Splitter is not the first app that can help with this. There are a few others (iZotope makes a good one) that can strip out dialog from background noise and smooth out imperfections. Originally licensed to Hollywood for movies, it turns out this works for sampling from old (and new) records too.

LALAL.AI has two advantages when they broke into the mainstream a year or so ago:

1. They made a user-friendly web app front-end for their technology, and
2. They made it so anyone who wrote about it could get paid.

Tons of websites posted basically company copy when LALAL.AI debuted, taking a small slice of referrals by sending them traffic. It was obvious by the bullet points repeated nearly verbatim by different “reviewers” that many never used the product much, or at all.

It’s a shame that many did not try the product themselves, because it is really easy and I would even say it’s fun to use this. That user-friendly web app at invites you to try their separation technology yourself and see how you like it. It automatically IDs elements in the extract you upload, breaking it down by type and letting you listen to the horn, guitar or vocal, by itself and download stems for each or all of these. The job it does is impressive.

To test it out I uploaded a series of different media files, some recorded from very poor source material. A heavily compressed mp3 in bad quality still gave a good result, although it didn’t magically make the sample sound better. Interesting side note: it interpreted some of the audio distortion as an element by itself — “noise” — which was kind of interesting to play with later. LALAL.AI also extracts audio from mp4, AVI and mkv files — crucial for sampling from video or even in some cases from captured streaming audio — in addition to mp3, ogg, wav, aiff, flac and aac audio.

More like this? Get it in your inbox:

Other than the web app, the price is a major factor here over iZotope’s software too. LALAL.AI is sold as SAAS, software as a service, meaning you don’t really own it. You can “rent” it for free just to try. Saving 90 minutes of audio is $15, $30 for 300.

It may be wise to use LALAL.AI while you still can. In a submission to the Office of the US Trade Representative, the RIAA condemned the rise of web-based “AI based extractors/mixers”:

To the extent these services, or their partners, are training their AI models using our members’ music, that use is unauthorized and infringes our members’ rights by making unauthorized copies of our members works. In any event, the files these services disseminate are either unauthorized copies or unauthorized derivative works of our members’ music.


Ominously, the rest of the document deals with straight-forward piracy and online infringement; apparently the RIAA considers them all the same sort of threat. There’s little legal precedent for this, which means the RIAA may be preparing to establish one.

Terry Matthew contributed to this report.


Comments are closed.