The arms race among the AI music technology companies doesn't seem to stop, and that what makes them improve their tech and offerings. And even if the evolution of gen AI tools like Udio or Suno might be scary, AI-powered stem separation brings actual value to the table, firstly because it doesn't necessarily need to use copyright-protected data to train models to isolate vocals and other stems from audio—it doesn't replicate the vocals, after all.
LALAL.AI has just introduced Perseus, their latest AI-driven tool for vocal extraction and removal. For anyone familiar with audio and video production, vocal separation has long been a key feature in remixing, karaoke, video dubbing, and even professional mastering. Now, LALAL.AI is raising the bar with a new neural network, leveraging transformer technology, akin to OpenAI’s approach in ChatGPT, but tailored for sound.
What Makes Perseus AI Stand Out?
At the core of Perseus AI is a transformer model—an advanced machine learning framework first developed by Google and now widely used across various fields. What sets Perseus apart from its predecessors like LALAL.AI's Orion is the use of this transformer-based architecture for audio processing. The outcome is a more precise vocal separation experience.
LALAL.AI reports in their press release that Perseus achieves roughly a 15% improvement in extraction quality compared to Orion and delivers enhanced clarity and separation in both acapella and instrumental creation, which is about a decibel higher across multiple benchmarks, which represents the next step in their neural network journey.
Speaking of the naming, by the way. Continuing their tradition of naming neural networks after constellations, LALAL.AI chose the name “Perseus” for its connection to the myth of transformation.
"The fact that the new neural network works on a transformer model reminded us of the Perseus constellation, more specifically, of the myth of Perseus, which is closely associated with the themes of transformation," the team shares. "It includes Medusa's monstrous transformation from a beautiful woman, the transformative powers of Medusa's head, and Perseus's own heroic transformation in overcoming the challenge. Transformation is a central motif throughout the Perseus myth, and we want our Perseus AI to bring you the joy of transforming your songs into studio-quality-level stems."
How Does Perseus AI Work?
Perseus is designed specifically for vocal and instrumental stem separation, and users won’t need to take any additional steps to activate it—Perseus is enabled by default. If you're familiar with the LALAL.AI platform, the process remains straightforward. Select "Vocal and Instrumental," upload your track, and let Perseus handle the rest.
For users interested in comparing Perseus to a previous model, LALAL.AI allows side-by-side previewing of both models for free, so you can switch between them to hear the difference.
In addition to being the default on the Stem Splitter, Perseus is also integrated into LALAL.AI’s other key services, including the Voice Cleaner, Lead & Back Vocal Splitter, and Echo & Reverb Remover. No matter what tool you’re using, Perseus promises to deliver precise vocal isolation with minimal effort.
To mark the launch of Perseus, LALAL.AI has added extra value to their paid packages. Pro pack users get an additional 50 minutes of processing time, while Plus pack users get 30 extra minutes. This promotion runs from September 25 to October 7.
Curious to see how good it will isolate vocals? Head to the LALAL.AI website to give Perseus AI a try.
