Separate audio with a prompt
— no plugins, no engineering.
Isolate from any track in minutes.
Built for creators who want clean audio fast — with privacy-first processing.
Sievin brings Meta's Segment Anything for Audio (SAM-Audio) model to creators. Describe any sound to isolate or remove — no stems, no presets, just a text prompt.
Built for creators like you
No matter your workflow, Sievin helps you get clean audio fast.
For video creators & editors
Clean up audio for YouTube videos, reels, and ads: isolate voice, remove background noise/music, or extract instrumentals.
For podcasters
Separate speech from background audio, reduce cross-talk, and create cleaner clips for social.
For educators & presenters
Pull voice from recordings, remove distractions, and reuse audio for courses and slides.
For musicians & practice
Isolate vocals or instruments to learn parts, practice covers, or build references.
How it works
Three simple steps to separate any sound from your audio
Upload your audio
Drag and drop any audio file up to 10 minutes. We support MP3, WAV, FLAC, and M4A.
Describe the sound
Tell us what to isolate or remove. "Vocals", "background noise", "drums" — anything.
Download the result
Get your processed audio in minutes. Clean, precise, and ready to use.
Privacy you can trust
Secure processing
Files processed on isolated GPU instances — nothing shared.
24-hour auto-delete
All files are permanently deleted after 24 hours.
No AI training
Your audio is never used to train our models.
Frequently asked questions
Everything you need to know about Segment Anything for Audio and how Sievin works.
Segment Anything for Audio (SAM-Audio) is an AI model developed by Meta that can isolate any sound from a mixed audio recording using a text prompt. Unlike traditional stem splitters that only separate fixed tracks like vocals or drums, Segment Anything for Audio lets you describe any sound — a dog barking, a siren, a specific instrument — and extract or remove it. Sievin is built on this model, making Segment Anything for Audio accessible to creators without any audio engineering skills.
Audio separation is extracting a specific sound from a mixed recording — like isolating vocals, removing background music, separating speech from noise, or extracting drums — so you can reuse it in video editing, podcasts, or music production.
No — but writing a clear prompt helps. SAM-Audio is prompt-sensitive by design, meaning the quality of your results depends on how you describe what to extract. We include tips inside the app to help you write tighter prompts (like "male speech" instead of just "voice"). No audio engineering knowledge required.
Yes — but designed for non-engineers. Traditional stem splitters separate predefined tracks (vocals, drums, bass). Sievin uses Meta's Segment Anything for Audio (SAM-Audio) for prompt-based extraction — describe any sound to isolate or remove. The tradeoff: prompts need to be specific for best results. Learn more about SAM-Audio at ai.meta.com/blog/sam-audio.
Yes. You can remove vocals to create an instrumental, or isolate vocals to extract an acapella. For best results, be specific — "female singing voice" works better than just "vocals". The app includes prompt tips to help you get cleaner separations. Note: SAM-Audio works best with distinct sounds; overlapping voices or similar instruments may require more precise prompts.
We support MP3, WAV, FLAC, and M4A files up to 10 minutes in length. This covers most audio clips from video editing, podcasts, and music production workflows.
Yes — privacy-first audio processing by design. Your files are processed on isolated GPU instances, automatically deleted after 24 hours, and never used to train AI models. Your audio stays yours.