LALAL.AI
NewSplit songs into vocal, drum, bass, guitar, piano and other stems, clean voice recordings and remove echo or reverb, using web, desktop and mobile apps, local VST3 plugins or an API.
LALAL.AI
Explore features, practical uses and pricing below.
LALAL.AI is a stem separation service. It takes a finished mix and pulls it apart into estimated parts: vocals, instrumental, drums, bass, guitars, piano, synth, strings and wind instruments. The same engine drives speech tools for cleaning dialogue, removing echo and reverb, and changing or cloning a voice. It runs in the browser, in desktop apps for Windows, macOS and Linux, on iOS and Android, as VST3 plugins and through an API.
What LALAL.AI separates, and with which models
You upload an audio or video file and choose a stem type. The FAQ lists Vocal and Instrumental, Drums, Bass, Electric Guitar, Acoustic Guitar, Piano, Synthesizer, Voice and Noise, String Instruments and Wind Instruments. Inputs include MP3, WAV, FLAC and AIFF audio as well as MP4, MOV and MKV video.
The work is done by named neural networks. According to the changelog, Perseus grew during 2024 and 2025 from vocals to drums, bass, guitars and piano, and also powers the Lead/Back Splitter for separating lead from backing vocals. A newer network, Andromeda, arrived in November 2025 for vocal isolation and gained drum and bass stems in March 2026. Where a stem offers more than one network, you can pick, which is worth trying on difficult material.
Speech cleanup, voice tools and the DAW plugins
- Voice Cleaner strips background music, mic rumble, plosives and other noise from spoken recordings.
- Echo & Reverb Remover reduces room sound on vocals and dialogue. It also appears as a De-echo setting inside the vocal and voice modes.
- Voice Changer and Voice Cloner reshape a voice or build a model from your own recordings.
The VST plugin family brings three jobs into a DAW: Stem Separator, Speech Cleaner and Voice Dereverb. LALAL.AI says plugin processing happens locally, so audio isn't uploaded, though the plugins check your subscription online from time to time. VST3 runs on Windows, macOS and Linux, AU is in beta on macOS, and AAX isn't supported yet.
Who gets the most from it
Remixers, DJs, cover bands, music teachers and karaoke makers are the obvious users, since each needs one part of a song without the rest. Video editors and podcasters may get more from the Voice Cleaner and dereverb tools when music or room tone sits under dialogue. Developers can build separation into a product using the API manual. If you want to compose new music, look elsewhere, because LALAL.AI only works on audio you already have.
Pulling an acapella for a remix
- Upload a WAV or FLAC of the track rather than a compressed download.
- Run Vocal and Instrumental and check the vocal stem on headphones for reverb tails and cymbal bleed.
- If the lead is tangled with harmonies, run the Lead/Back Splitter.
- If artifacts distract, try the other available network and compare.
- Import the better stem into your DAW and check its timing against the original.
Where separation falls short
Stems are estimates. Dense mixes, distorted guitars in the vocal range and long reverb can leave ghosting or watery artifacts, so expect to clean the output with EQ or gating. Usage is counted per stem type, so five stems from one song use five times its length, according to the pricing page. Separating a commercial track doesn't give you the right to release the result.
Plans, queues and minutes
The pricing page lists a free Starter plan meant for checking output quality, with 10 minutes in the Relaxed queue, a 200 MB upload limit and no result downloads. Paid Lite and Pro subscriptions add unlimited Relaxed-queue processing, a monthly Fast-queue allowance and 2 GB uploads, and Pro includes the VST plugins and API access. The company says both queues give the same quality; Fast jobs simply get priority. One-time top-ups add Fast minutes, and unused Fast minutes don't roll over.
Questions about LALAL.AI
Can I download stems on the free plan?
The pricing table marks result downloads as not included on Starter, so treat it as a quality preview.
How does it differ from AudioShake?
Both split finished mixes. AudioShake also handles dialogue and background splits for video and timed lyrics, while LALAL.AI adds local DAW plugins and more single-purpose voice tools.