We believe great audio
tools should be simple.
AudSep is an AI-powered audio stem separator built for producers, DJs, educators, podcasters, and anyone who has ever needed to pull a vocal or instrument out of a mix. We built it because everything else was too slow, too expensive, or too hard to use.
OUR MISSION
Make every creator's audio sound like it came from a studio session.
For decades, stem separation was a privilege of major labels with expensive licenses and dedicated engineers. Research labs eventually made the models open-source, but the tooling stayed frustratingly inaccessible.
AudSep changes that. We wrap the world's best open separation models in a clean interface that works on any device, in any browser, in under a minute. No software to install. No PhD required.
Stems extracted in real-time
WHAT WE STAND FOR
Our values
Speed over perfection
We ship fast, learn from real users, and iterate relentlessly. A working tool in your hands beats a perfect tool on a whiteboard.
Privacy by default
Your audio is yours. We process files in-memory and delete them within an hour. We never train models on your uploads.
Built for creators
Every decision starts with the creator. Not the enterprise buyer, not the investor deck — the producer, educator, or podcaster who just needs it to work.
Open by nature
We build on open-source models, publish our API, and write openly about how everything works. Transparency is a feature.
THE STORY
How we got here
Frustrated by clunky, expensive stem separators, we asked: what if this was just fast, clean, and free to try?
Wrapped Meta's HTDemucs in a Next.js frontend with real-time streaming progress. First separation took 4 minutes. We knew there was a long way to go.
Shipped v1.0 with Supabase for job tracking, Clerk for auth, and a Railway-hosted Python worker handling the heavy lifting.
Integrated Razorpay with monthly and yearly billing. Added the REST API for developers and studios.
Hundreds of tracks separated weekly. Roadmap includes batch processing, webhook-driven workflows, and multi-speaker diarisation.
Small team, strong opinions
AudSep is built by a small team of engineers who are also musicians. We run the product on honest feedback, real usage logs, and a deep belief that the best audio tools should feel invisible — they just work.
We don't have VC money or a 50-person roadmap committee. That means we move faster, prioritise the features that actually matter, and stay honest in our changelog.