# About Us

We are a team of specialists in the fields of artificial intelligence, machine learning, mathematical optimization, and digital signal processing. **Our goal is to make working with audio and video easier** for musicians, sound producers, music engineers, video bloggers, streamers, transcribers, translators, journalists, and many other professionals and creatives.

In 2020, we developed a unique neural network called **Rocknet** using 20TB of training data to extract instrumentals and voice tracks from songs. In 2021, we created Cassiopeia, a next-generation solution superior to Rocknet that allowed improved splitting results with significantly fewer audio artifacts.

Starting as a 2-stem splitter, LALAL.AI has grown significantly during 2021. In addition to **vocal and instrumental**, the service was enhanced with the capability to extract musical instruments – **drums**, **bass**, **acoustic guitar**, **electric guitar**, **piano**, and **synthesizer**. As a result of this upgrade, LALAL.AI became the [world’s first 8-stem splitter](/content/blog/lalal-ai-adds-the-8th-stem-for-separation-synthesizer/index.html). In the same year, we also presented [business solutions](/content/business-solutions/index.html), enabling owners of sites, services and applications to integrate our stem-splitting technology into their environments via API.

Only available in English prior to 2021, LALAL.AI was translated into 7 other languages – Chinese, French, German, Italian, Japanese, Korean, and Spanish. Furthermore, we added new payment methods to make LALAL.AI easier to acquire and more accessible to people worldwide.

In 2022, we created and released [Phoenix](/content/blog/phoenix-neural-network-vocal-separation/index.html), a state-of-the-art audio source separation technology. In terms of stem-splitting accuracy, it surpassed not only our previous neural networks but also all other solutions on the market.

Although Phoenix exclusively handled vocal/instrumental isolation at first, its powerful technology allowed us to continually introduce new stems on a regular basis. Throughout the year we trained Phoenix to extract all musical instruments that Cassiopeia supported.

We also added two brand new stems, **wind** and **string instruments**, which no other service offered. With that update, LALAL.AI broke the record again and became the [world’s first 10-stem splitter](/content/blog/wind-string-instruments/index.html).

LALAL.AI’s innovative technologies are used not only for stem splitting. In July of 2022, we introduced [Voice Cleaner](/content/blog/voice-cleaner/index.html), a noise cancellation solution that removes background music, mic rumble, vocal plosives, and many other types of extraneous noises from video and audio recordings.

At the end of 2022, we created a [desktop version of LALAL.AI](/content/blog/lalalai-desktop-app/index.html). The application enabled users to split audio and video files into stems in one convenient place on their Windows, macOS and Linux computers.

In 2023, we developed [mobile applications for iOS and Android](/content/apps-and-plugins/index.html), making LALAL.AI and all its features readily available on the go. We also trained and released [Orion](/content/blog/orion-new-neural-network/index.html), our fourth-generation neural network that takes a unique approach to audio separation. The network not only extracts stems from the mix but also enhances them in the process, eliminating errors and audible imperfections for the best possible quality results.

In 2024, LALAL.AI continued the trajectory of innovation with several significant updates. We introduced [Enhanced Processing and Noise Canceling Level](/content/blog/enhanced-processing-noise-canceling/index.html) features in March, allowing users greater control over audio separation for vocals and instruments while providing customizable noise reduction options. In April, we launched the [Voice Changer](/content/voice-changer/index.html), enabling users to transform songs and vocal recordings with various artist and character voice clones. August brought the introduction of the [Echo & Reverb Remover](/content/echo-reverb-remover/index.html), a tool designed to enhance audio clarity by effectively removing unwanted echoes and reverberations from recordings.

By September 2024, we released the highly anticipated [Lead & Back Vocal Splitter](/content/lead-back-vocals-remover/index.html), simplifying lead and backing vocal isolation. The launch of the [Perseus](/content/blog/perseus-ai-lalalai-transformer-neural-network/index.html) neural network shortly thereafter marked a significant leap in vocal extraction technology. Utilizing advanced transformer models similar to those behind OpenAI's ChatGPT, Perseus offers unprecedented clarity in vocal isolation while minimizing artifacts. Throughout 2024, Perseus was updated to support not only vocal and instrumental isolation but also the extraction of drums and bass.

In 2025, LALAL.AI reached a new milestone in growth and product development, with the **number of registered users increasing from 3.8 million in 2024 to 6.79 million**, marking the strongest year-on-year gains in the company’s history. The Perseus neural network was expanded to handle **Acoustic Guitar**, **Electric Guitar**, and **Piano** stems. [LALAL.AI Voice Cloner](/content/voice-cloning/index.html) was introduced as a separate product that creates a reusable copy of a user’s voice from uploaded speech samples, so the same voice can be used in audiobooks, voiceovers, videos, ads, and other projects without new recordings.

The [Andromeda neural network](/content/blog/andromeda-audio-transformer-neural-network/index.html) was released as the most advanced model in the LALAL.AI stack, trained on roughly four times more data than Perseus and using transformer architecture to deliver up to 40% faster processing and about 10% higher SDR (Signal-to-Distortion Ratio). Andromeda became the default for Stem Splitter ( **Vocal & Instrumental** and **Voice & Noise** stems), Voice Cleaner, Lead & Back Vocal Splitter, and Echo & Reverb Remover, reducing vocal bleed, reverb artifacts, and high-frequency loss and delivering more consistent results.

LALAL.AI remains dedicated to pushing the boundaries of AI-powered audio processing while continuously improving our products. We work hard to create fresh, high-quality solutions, and we always have a lot of ideas and developments in store. Keep your eyes peeled for new possibilities and improvements!

**Omnisale GmbH**

Holzhäusernstrasse 16, CH-6343, Holzhäusern, Switzerland.