Free Online Stem Splitter
Split songs into vocals, bass, drums and more — classic DSP.
Drag & drop your audio file here
or
Max file size: 100 MB · MP3, WAV, FLAC, OGG
Original
beats per minute
musical key
Processing…
Result
How it works
The Stem Splitter breaks a song into four stems - vocals, bass, drums and other - using classical digital signal processing, not AI. The recipe is transparent: bass is the lowpassed center of the mix, vocals are the center channel with the bass removed, drums are the transient (attack-heavy) part of what remains, found by spectral gating, and "other" is everything left over. By construction the four stems add back up to your original exactly.
Set expectations honestly: DSP separation bleeds. Center extraction cannot remove stereo vocal reverb, and the spectral gate cannot tell a snare crack from a vocal consonant with certainty. What you get is genuinely useful - a vocal stem clean enough for karaoke, a drum stem solid enough to sample - produced in seconds with zero uploads and zero waiting on a server queue.
Each stem gets its own Play/Stop preview and Download button, using the page's output format (MP3 or WAV). Everything - the split, the previews, the encoding - runs client-side in your browser tab, so your music never leaves your device and nothing is uploaded anywhere.
Frequently asked questions
Is this AI stem separation?
No - and that honesty matters. This tool uses classical DSP: vocals are extracted by taking the center channel (what is identical in left and right) minus the low bass, bass is a 250 Hz lowpass of the mid signal, and drums are separated by spectral gating on transients. Expect bleed - a little vocal in the drums, a little snare in the vocals. AI models (Demucs, Spleeter) separate far more cleanly but need gigabytes of model weights and often a server; this runs instantly, offline, in your browser.
Why can I still hear the vocals in the "other" stem?
Because separation is imperfect by design here. Stereo reverb on a vocal lives in the side channels, so center extraction cannot remove it; cymbals share frequency space with vocal sibilance, so the spectral gate lets some through. Think of the stems as "mostly vocals", "mostly drums" - useful for practice, karaoke and sampling, not for forensic isolation.
What are stems useful for?
Practice (mute the vocal stem and sing along), karaoke backing tracks, sampling drum breaks, remixing, and studying how a mix is put together. Each stem can be previewed and downloaded separately as MP3 or WAV.
Is there a file size limit? Is it private?
Files up to 100MB are supported. Separation runs entirely in your browser - your music never leaves your device.