How to use Acapella extractor
- 1
Add the song
Choose the best-quality copy you have: a WAV, FLAC or high-bitrate MP3 gives a cleaner a cappella than a clip ripped from a phone recording. Music videos work too.
- 2
Separate
Keep the 2-stem mode, the model pair trained on voice against accompaniment, and press Separate. The first time, a 39 MB model is fetched and stored in the browser.
- 3
Check the vocal on its own
The result opens with the vocal soloed. Scrub through choruses and quiet verses, the parts where leftover drums or guitar are easiest to spot.
- 4
Download the a cappella
Save the vocals as WAV to edit in a DAW, or as MP3 to share with your band; the instrumental is right below if you need both.
Limits and Pro
10 minutes and 500 MB per song; longer files keep their first 10 minutes. Phones and other devices with 4 GB of memory or less process 5 minutes at a time.
Songs up to 20 minutes on a computer. See Pro
About Acapella extractor
An a cappella (acapella) is the vocal part of a song with the band taken away. Producers build remixes and mash-ups on it, cover singers study the original phrasing and breathing, teachers use it to show harmonies, and anyone writing down lyrics hears every word more easily without the drums underneath.
This page opens the vocal remover in vocal mode: the stem you hear first and the main download are the vocals. It uses Deezer's Spleeter model, which decides for every moment and frequency how much of the sound is voice, keeps that share and drops the rest.
What ends up in the vocal track is everything the model hears as singing: the lead, harmonies, backing vocals, ad-libs and spoken parts, all in one stem. Reverb and delay that were added to the voice usually stay with it, which sounds natural on its own but can smear a remix at a different tempo. Very breathy or whispered lines can come out quieter than in the original, and sustained synth pads or a lead guitar in the same range sometimes bleed through. Mono recordings are harder than stereo ones, because the voice and the band share one channel.
The vocal keeps the original key and tempo, so it lines up sample for sample with the source and with the instrumental stem. To change the key or the speed, load the WAV into an audio editor afterwards; this tool only separates.
Privacy
The vocals are isolated on your own device: the song is read and processed by a worker in this browser tab and is never uploaded. The Spleeter model comes once from huggingface.co and the runtime from jsDelivr; neither request contains your music.