Projects

LALAL.AI

ISD Team

Abstract

LALAL.AI transforms any audio or video file into isolated stems — vocals, drums, bass, guitar, and more. Powered by transformer-based AI and available via browser, desktop app, mobile, API, and VST plugin, it delivers professional stem separation to musicians, producers, companies, and content creators globally.

Description

Audio source separation was once the domain of professional studios equipped with expensive software and trained engineers. LALAL.AI was built on a different premise: that anyone (a bedroom producer, a podcaster, a music teacher, a YouTube blogger, or even an enterprise) should be able to extract controllable stems from any recording, instantly and without technical expertise.

The project launched in 2020 as a browser-based tool powered by Rocknet, a proprietary neural network trained on 20TB of audio data. The initial product was straightforward: upload a file, get vocals and instrumentals back, separately; clean and fast. But the core challenge in stem separation is not just separation, it’s quality.

Early tools produced audible artifacts: the “watery” sound, bleed between stems, lost high frequencies. LALAL.AI addressed this through continuous neural network development, releasing six successive generations of models — Rocknet, Cassiopeia, Phoenix, Orion, Perseus, and Andromeda — each trained on larger datasets and incorporating advances in deep learning architecture.

The latest, Andromeda, uses transformer technology and processes tracks up to 40% faster than its predecessor. The product itself evolved in parallel. What began as a two-stem web tool expanded into a full audio processing ecosystem (like a Swiss army knife for audio processing): up to 10-stem separation (vocals, drums, bass, guitar, piano, synths and more), noise and echo removal, voice modification, mobile apps for iOS and Android, API access for business integration, and in 2026 — a VST plugin enabling stem separation directly inside any DAW, offline, without file uploads.

Today, LALAL.AI is trusted by nearly 7 million musicians, producers, sound engineers, video editors, content creators worldwide, including companies, processing tens of millions of stems per year. In 2026, the platform won the Webby People’s Voice Award for Best Use of AI & Machine Learning, selected by its own user community over competitors, including Adobe.