LALAL.AI: The World’s Most-Used AI Audio Separator
Built for Music Producers, DJs, Podcasters, Video Editors, and Everyone in Between
LALAL.AI launched in 2020 with a focus on making stem separation fast, accurate, and accessible. Six years later, the platform processed nearly 64 million audio splits in 2025 alone, serving over 6.8 million registered users, which is nearly double the count from the year before.
Stem splitting put LALAL.AI on the map, but the platform has since grown into a full suite of audio processing tools, used by music producers, DJs, podcasters, video editors, vocalists, and karaoke creators worldwide. LALAL.AI was the first service to surpass iZotope RX 9, Spleeter by Deezer, and Moises in both accessibility and stem splitting accuracy, with published comparison results to back it up.
Six Generations of Neural Networks
Since 2020, LALAL.AI has developed six generations of neural networks: Rocknet, Cassiopeia, Phoenix, Orion, Perseus, and Andromeda. Each generation measurably outperformed the last.
Perseus (released in 2024) introduced transformer architecture to audio source separation, among the first applications of this model family in the field, the same technology behind large language models like ChatGPT. It delivered a significant improvement in vocal extraction clarity and precision.
Andromeda (2025) is the current flagship cloud model. Trained on four times more data than Perseus, it processes audio up to 40% faster and scores 10% higher on SDR (Signal-to-Distortion Ratio) benchmarks. It consolidates the previous two Enhanced Processing modes into a single unified network, removing the trade-off between detail extraction and cross-bleed prevention. Andromeda is the default model across Vocal Remover, Stem Splitter, Lead & Back Vocal Splitter, and Echo & Reverb Remover.
In 2026, LALAL.AI also introduced Lynx, a specialized cloud model built for voice isolation and noise removal.
Lynx (2026) is built on a proprietary architecture using a U-Net design with custom blocks and a non-local attention mechanism, optimized for embedded use cases and roughly six times smaller than Andromeda. It separates voice from noise more consistently than earlier models, and is the default model in Voice Cleaner. It also handles the Voice and Noise stem in Stem Splitter and Vocal Remover. Developers can access it via API to integrate voice isolation into their own products.
Alongside the cloud model lineage, LALAL.AI also developed Lyra, a model built specifically for on-device, offline processing.
Lyra (2026) runs on the user’s own hardware, with no file uploads, no internet connection required, and no processing minutes deducted. Seven stems are supported: Vocals, Instrumental, Drums, Bass, Piano, Acoustic Guitar, and Electric Guitar. Lyra is available in both the LALAL.AI desktop app and the VST plugin, a natural fit for DAW-integrated workflows, live settings, and situations where keeping audio files on-device is a hard requirement.
What the Platform Does
Seven products, built around audio separation and voice processing:
Stem Splitter separates audio and video files into up to 10 stems: Vocals, Instrumental, Drums, Bass, Piano, Acoustic Guitar, Electric Guitar, Synthesizer, Strings, and Wind.
Vocal Remover isolates the voice from the rest of a track across multiple stem types.
Voice Cleaner removes background noise, vocal plosives, and mic rumble from voice and vocal recordings.
Voice Changer modifies voices in audio recordings and video files, with controls for accent and tonality.
Voice Cloner builds a personalized AI voice model from a user’s own recordings and audio samples. Used for audiobooks, video voiceovers, ads, and vocal experiments.
Echo & Reverb Remover cleans up echo and reverberation from vocals, voice recordings, songs, and video files.
Lead & Back Vocal Splitter separates lead and backing vocals from songs, outputting four stems: lead vocals, backing vocals, instrumental, and instrumental with backing vocals.
All products support popular audio and video formats: MP3, OGG, WAV, FLAC, AIFF, AAC, M4A, AVI, MP4, MKV, MOV, and M4V.
By the Numbers
Users:
6,793,709
registered as of end-2025, up from 3.8 million in 2024. That’s 79% year-over-year growth.
Volume:
63,811,258
splits processed in a single year, totaling 14,830,175 hours of audio.
What people split most:
32.3M
Vocals
11.5M
Voice
5.5M
Drums
2.5M
Bass
2.1M
Electric Guitar
Most used upload formats:
19.1M files
MP3
3.1M
MP4
2.1M
M4A
For Developers and Studios
API makes LALAL.AI’s audio separation and voice processing available as a service, letting developers integrate stem splitting, noise removal, and voice processing directly into their own SaaS products, media platforms, or production pipelines.
VST Plugin brings Lyra into any compatible DAW as a native plugin, running entirely on the user’s hardware with no file uploads and no subscription minutes deducted. It fits naturally into existing production workflows, and the license activates on up to three workstations.
Enterprise plans are tailored for media companies and large-scale production environments that need custom solutions, higher processing volumes, or dedicated support.
LALAL.AI is also accessible across web, desktop (Windows, macOS, and Ubuntu), iOS, and Android.
30th Annual Webby Awards, 2026 — People’s Voice Winner
Best Use of AI & Machine Learning, Apps, Software & Immersive
The People’s Voice Award is decided by public vote, with millions of ballots cast each year. LALAL.AI won against a shortlist that included Adobe Premiere Object Mask, Beeble AI, Captions AI, and Cosmos, in a competition with more than 13,000 entries from over 70 countries. The platform also received a Webby Honoree distinction in Creator, Creative & Media Tools the same year. The Webby Awards have been called “the Internet’s highest honor” by The New York Times.
About LALAL.AI
LALAL.AI is an audio processing platform used by more than 6 million music producers, engineers, DJs, podcasters, and media professionals worldwide. The company is headquartered in Zug, Switzerland, and operated by OmniSale GmbH. Through continuous neural network research, it has grown from a single stem-splitting tool into a full suite of audio processing products available across web, desktop, mobile, API, and DAW environments.
Contact: marketing@lalal.ai
