LALAL.AI has launched Andromeda, a major upgrade in audio source separation. Built from six generations of research, it focuses on speed, clarity, and consistency. The team trained the model on far larger datasets and refined its DSP methods. As a result, creators get cleaner stems with far less manual cleanup.
Advanced Audio Analysis
Andromeda uses transformer-based analysis to study sound by time, frequency, and tone. Consequently, it detects timbre, resonance, and instrument presence across the stereo field. Moreover, it reduces common problems like vocal bleed, unwanted reverb, and high-frequency loss. The model runs up to 40% faster than its predecessor and shows up to a 10% SDR improvement on benchmarks. Also, it supports stereo processing up to 22 kHz and accepts many file formats, so workflows stay flexible.
The network now powers LALAL.AI’s core tools: Stem Splitter, Voice Cleaner, Lead & Back Vocal Splitter, and Echo & Reverb Remover. Andromeda is the default engine in these services, so you can extract vocals and instruments with fewer steps. In practice, quiet backing vocals and subtle textures come through sharper. Additionally, the model balances detail extraction with cross-bleed prevention. Thus, you no longer need to choose between minimizing leakage and maximizing detail.
Training and Development
Training took more than a year, including over nine months of pre-training and a multi-stage data cleaning process. The team applied unique filtering techniques and modern DSP advances. Consequently, Andromeda handles a wider variety of audio sources and reduces artifacts. It also makes loudness less of a factor, so soft and loud tracks yield more consistent results.
Looking ahead, LALAL.AI will expand Andromeda beyond the web. The company plans mobile app integration and a VST plugin for in-DAW use. Therefore, producers can isolate stems without leaving their workflow. Meanwhile, LALAL.AI will share updates on Facebook, Instagram, TikTok, Reddit, X, and YouTube. If you work with audio, try Andromeda in the web tools and watch for the upcoming releases.
Andromeda raises the bar for AI-driven audio separation by delivering faster processing, clearer stems, and a simpler user experience. It brings studio-grade isolation to more creators and reduces time spent on manual fixes. Try it today to hear the difference.
About the Author
