What an acapella extractor does
An acapella is a vocal with no music behind it. Producers used to wait for official acapellas on vinyl or promo packs. Now an acapella extractor can isolate vocals from almost any finished song.
AudioDrop splits the track into four stems, vocals, drums, bass and other, and solos the vocals. The HTDemucs model does the heavy lifting, running on your own device with WebAssembly.
Uses for an isolated vocal
An extracted acapella opens up a lot of creative and practical work:
- Remixes: drop the vocal over a new beat at a new tempo.
- Mashups: combine the acapella from one song with the instrumental of another.
- Vocal study: hear every breath, harmony and ad lib clearly.
- Lyric transcription: write down words that get buried in the mix.
- Singing lessons: compare your phrasing against the original.
- Sampling: chop a vocal hook into a new idea.
Getting a cleaner acapella
Start with a lossless or high bitrate file. Compression artefacts in a low quality MP3 end up in the vocal stem.
Always download after the AI pass completes. The instant preview uses a quicker method and leaves more music behind.
Songs with a dry, upfront vocal extract best. Heavy reverb and delay are part of the vocal sound, and some of that tail can spill into the other stem.
Planning a remix? Run the acapella through the BPM finder and key finder so you know what tempo and key to build around.
Matching an acapella to a new track
When you build a mashup or remix, the numbers that matter are tempo and key. Small moves sound natural. Big moves start to sound processed.
| Change | Tool | How it usually sounds |
|---|---|---|
| Tempo ±5% | BPM changer | Hard to notice |
| Tempo ±15% | BPM changer | Noticeable but usable |
| Key ±1 to 2 semitones | Key changer | Natural on most voices |
| Key ±3 to 5 semitones | Key changer | Voice character starts to shift |
Limits of vocal isolation
Expect a little bleed. Hi-hats, snare hits and bright synths can leave faint traces in an isolated vocal. Backing vocals usually come along with the lead, since both are voices. Distorted, heavily processed or lo-fi vocals are harder for the model to separate, and you may hear more artefacts.
Acapella extractor vs searching for an official acapella
Official acapellas come straight from the multitrack, so they are perfectly clean. But most songs never get one released. An acapella extractor fills that gap, and for many uses the difference is small once the vocal sits in a new mix.
Where you will notice it is on a completely exposed vocal, for example an intro with the voice alone. Here any bleed stands out. In a busy remix, the new beat hides most small artefacts.
A good workflow is to isolate vocals here, drop them into your DAW, and listen in the context of your new track before deciding if the quality is good enough. Often it is. If a line has a noticeable artefact, a short fade or a little reverb on top usually hides it.
- Fully exposed vocal: artefacts easier to hear.
- Vocal over a new beat: small bleed is usually masked.
- Chopped vocal samples: short slices hide most issues.
- Lyric and melody study: quality rarely matters.
Private, free and in your browser
The acapella extractor never uploads your song. The AI model, about 166 MB, downloads to your browser once and is cached, then everything runs locally. There is no account, no watermark and no per-song fee.
It works in Chrome, Edge, Firefox and Safari on desktop, and in mobile browsers too, though long songs are slower on a phone. If you release a remix, clear the rights to the vocal.