Isolate Vocals from a Song
Pull a clean vocal out of a full mix, separating the singer or speaker from the instrumental backing. Powered by Mel-Band RoFormer, an open-source separation model with commercially licensed weights.
How it works
Mel-Band RoFormer splits a song into vocal and instrumental parts by learning what each source looks like across the frequency bands of the mix. We return the isolated vocal stem with the music removed, useful for remixes, sampling, transcription and karaoke production.
What it's good for
- Acapella extraction for remixes
- Sampling and mashups
- Transcribing lyrics or speech
- Cleaning music off a voiceover
Details
- Engine
- Mel-Band RoFormer
- Formats
- MP3, WAV, M4A, FLAC, OGG, AAC
- Price
- Free to try
Frequently asked questions
On 50 excerpts from the MUSDB18 test set we measured a vocal SDR of 11.3 dB (higher is cleaner), against 8.4 dB for the model we used before. Faint instrumental bleed can remain on dense mixes.
Vocal isolation keeps the vocal and drops everything else; the remove-music tool is the same engine aimed at producing the karaoke (no-vocal) stem instead.
Stereo, full-quality files separate best. Mono or heavily compressed files still work but with a little more bleed.
Most three-to-five-minute songs process in well under a minute; very long files take proportionally longer because the whole track is run through the separation model.
You get a standard audio file at the original sample rate, and stereo input stays stereo so any vocal panning or reverb width is preserved.
The separation itself is yours to use, but the underlying song is still copyrighted, so clearing samples or releases with the rights holder is on you.
This tool returns the vocal stem; run the same file through the remove-vocals tool when you want the backing track as well.