Separate voice and music
Take only the vocals out of a song, remove the vocals to keep the accompaniment (a karaoke track), or keep only the speech or only the background music of a video or a podcast. One separation gives both. The AI runs on your computer, so nothing is uploaded.
- Free
- No upload
- No sign-up
- Works offline
How to use
Press "Get the AI ready (about 470 MB, the first time only)". When it is ready, press "Choose a file" and pick one song, or one sound or video file with voices.
Choose "Keep only the voice (singing, speech)" or "Remove the voice (accompaniment, music only)", then press "Separate". When it is done, listen with "Original", "Voice" and "Music".
Press "Save the voice" or "Save the music" to save it (vocals.mp3, instrumental.mp3 and so on). You can save the other one too without waiting, with "Save the music too" or "Save the voice too".
Questions
Is my file sent anywhere?
No. Reading the file and separating the voice from the music both happen on your computer. The AI is loaded from this site when you press "Get the AI ready". Nothing is sent out of the box after you choose a file.
Can I take only the vocals out of a song? Can I make a karaoke track?
Yes. "Keep only the voice" saves a file of the singing only, and "Remove the voice" saves a file of the accompaniment only (a karaoke track). The AI makes both in one go, so you can save the other one too without waiting.
Can it separate speech from background music in videos and podcasts?
Yes. You can keep only the speech, or remove the narration to keep only the background music. For a video, only its sound is used. Sound effects and background noise can stay a little in either one.
How clean is the separation?
It uses an AI that is good at separating singing from accompaniment (Mel-Band RoFormer, the "Duality" model). An AI is not perfect, though: with some songs, a little of the instruments can stay in the voice, or a little of the backing vocals or the echo of the voice in the accompaniment.
How long does it take?
It depends on the computer (its graphics chip above all). A 4-minute song takes less than a minute on a fast computer, and can take a few minutes on a laptop. A large video also takes time to read its sound. The tool shows how far it is and about how long is left, and you can stop it.
Is the "about 470 MB" downloaded every time?
Only the first time. The AI is kept in your browser's site data, so from the second time on it is ready at once, even without the internet. Nothing is installed. Clearing the data of this site in the browser's settings removes it.
Can I use it on a phone?
No. This tool is for computers only: its AI is very large, and phones and tablets do not have enough memory for it, or take far too long. Please open it in a recent browser on a computer that can use WebGPU (the graphics chip), such as Chrome or Edge.
What about the rights of the song (copyright)?
Use it for your own enjoyment, or with songs and sound that you may use. When you publish or share separated sound, mind the rights of the original song or sound.
Limits
- One file at a time. It reads the same formats as "Convert sound files", and the sound of videos (MP4, MOV and others).
- For computers only, in browsers that can use WebGPU (recent Chrome and Edge on computers, for example).
- The new files are at 44.1 kHz (OGG at 48 kHz), in stereo (in mono when the original is mono), as MP3, M4A, WAV, FLAC or OGG (M4A and OGG only in browsers that can make them).
- For a sound longer than 30 seconds, listening plays 15 seconds where the sound is loudest.
- The AI (about 470 MB) is loaded when you press "Get the AI ready". Once loaded, it stays in this browser.
- A file can be up to 4 GB and 1 hour (in browsers that can only read sound at once, and for Vorbis, about 48 minutes, in stereo at 44.1 kHz). Each new file can be up to 4 GB (1 MB is 1,000,000 bytes).
- We cannot protect against things in your own environment, such as browser extensions.