How the AI vocal remover works
Old vocal removers cancelled out whatever sat in the centre of the stereo image. That took the bass and kick drum with it and left a hollow, phasey mess. AudioDrop is an AI vocal remover. It separates the song into vocals, drums, bass and other, then simply mutes the vocal stem.
You hear a quick preview split in about a second. Behind it, the HTDemucs model from Meta AI research refines the separation. When it finishes, the instrumental gets noticeably cleaner, with less voice left behind.
Because the vocal is its own stem, you are not limited to on or off. Pull the fader halfway for a quieter guide, or solo the other stems to check what is left behind.
Why use a vocal remover
People remove vocals from a song for all kinds of reasons:
- Make an instrumental for karaoke or a live performance.
- Create a backing track to practise singing over.
- Get a beat or bed to rap or write a new topline on.
- Make background music for a presentation or school project.
- Hear the arrangement and chords without the voice in the way.
- Record a cover with the original music underneath.
AI vocal removal vs the old centre-cancel trick
Here is why a model-based vocal remover sounds better.
| Centre cancel (old) | AI vocal remover (AudioDrop) | |
|---|---|---|
| Bass and kick | Often removed with the voice | Kept in their own stems |
| Stereo width | Collapses to a thin sound | Stays as mixed |
| Reverb on vocals | Left behind in the sides | Mostly removed with the voice |
| Control | On or off | Fader for each stem |
Tips to remove vocals cleanly
Use the highest quality file you have. A lossless file or a 320 kbps MP3 gives the model more to work with.
Let the AI pass finish before you download. The preview is fine for a quick listen, but the refined split is where the vocal remover really earns its keep.
For singing practice, try leaving the vocal fader slightly up instead of fully muted. A faint guide vocal helps you stay on the melody.
Want the opposite? Use the acapella extractor to keep the voice and lose the music.
What a vocal remover cannot do perfectly
Some traces of the voice can remain, especially loud reverb tails, heavy delays and stacked backing vocals. Songs where the voice is heavily distorted, or old lo-fi recordings, separate less cleanly. You may also hear a little bleed of cymbals or guitar where the voice used to be. For most modern songs the result is good enough for karaoke, practice and demos.
Remove vocals from a song on iPhone, Android or Mac
On a Mac or PC, open this page in Chrome, Edge, Firefox or Safari and drag the song onto it. A laptop is the fastest option for full songs, since the AI pass uses several CPU cores at once.
On iPhone, open the page in Safari, tap the drop area and pick a song from the Files app. On Android, use Chrome and pick a file from your downloads or music folder. The vocal remover works the same way on a phone. It just takes longer, and very long tracks may run into the phone's memory limits.
Whichever device you use, the first run downloads the AI model once. After that it is cached, so the next instrumental starts straight away.
Songs bought as downloads, your own recordings and lossless rips all work well. Streaming services usually protect their audio, so you will need a file you can actually save to your device.
Free vocal remover online, with no upload
Many free vocal remover sites upload your song to their servers and make you wait. This one does the work on your device. Your file stays on your computer or phone, and there is no account, no queue and no watermark.
The AI model is about 166 MB and downloads once. After that, your browser caches it and every new song starts immediately. It works on Windows, Mac and Linux in Chrome, Edge, Firefox and Safari, and on iPhone and Android, though phones are slower with long songs. If you publish a cover or remix, respect the rights to the original.