How to Remove Vocals From a Song Free (Online)
How to Remove Vocals From a Song Free (Online)
Last updated: November 2026
To remove vocals from a song for free, open the Shimga Vocal Remover, load the song (up to 12 minutes), and download the instrumental track. The same pass also gives you the acapella: the vocals with the music dropped, which is also how you remove background music from a clip where someone is talking. The file is processed in your browser, so it is not uploaded to a server. How clean the result is depends on the recording. A dry, modern studio mix separates far better than a live bootleg or an old mono track.
Photo: Anderson Guerra on Pexels
Key facts
- What it does: splits a mixed song into two parts, vocals and everything else (the instrumental).
- Outputs: one pass gives you both an instrumental (vocals removed) and an acapella (vocals only). Download either as WAV or MP3.
- Two methods: AI separation (the open Spleeter model, a one-time download of about 55 MB that your browser then caches) or an instant Quick karaoke mode for stereo songs.
- Other uses: background voice remover (keep the music, lose the talking) and music remover (keep the voice, lose the music bed).
- Where it runs: in your browser. The audio stays on your device.
- Cost: free to use, with no sign-up needed to try it.
- Honest limit: separation is an estimate. Expect some faint vocal "ghosts" in the instrumental and some cymbal or synth bleed in the acapella, especially on busy or heavily reverbed mixes. The AI model also rolls off the very top end, near 11 kHz, so the instrumental can sound a little less airy.
How vocal removal actually works
There are two very different methods, and knowing which one a tool uses explains most of the quality differences you hear.
1. Phase cancellation (the old trick). In many stereo mixes the lead vocal sits dead centre, equally loud in the left and right channels. If you flip one channel upside down and add it to the other, anything identical in both sides cancels out. The vocal disappears, but so does the kick, the bass and the snare, because they are usually centred too. What is left sounds thin and hollow. This is what the classic "karaoke" button on old players did.
2. Source separation with a trained model (the modern way). A neural network trained on thousands of songs where the vocal and backing tracks were available separately learns what a voice "looks like" in a spectrogram. It then predicts which parts of your song belong to the voice and which belong to the instruments, and rebuilds two new audio files. This keeps the bass and drums intact and does not depend on the vocal sitting in the centre. Shimga's vocal remover uses this approach by default, and keeps the old centre-cancel trick as a Quick karaoke option for when you are offline or in a hurry.
Because the model is predicting, not reading a hidden track, the result is never a perfect studio stem. It is usually good enough for karaoke, practice, remix sketches and content where the music sits under speech.
Remove vocals from a song in 6 steps
- Open the tool. Go to shimga.app/vocal-remover. A laptop or desktop with a recent Chrome, Edge or Firefox is fastest. Recent phones work too, but keep the song under about five minutes and the screen on.
- Load your file. Drag in an MP3, WAV, M4A, OGG or FLAC file, up to 12 minutes long. If your song is inside a video, pull the audio out first with the video to audio extractor.
- Pick a method. The AI method gives the cleanest split. Quick karaoke is instant and needs no download, but only works on stereo songs with a centred lead vocal.
- Press Remove vocals. Your device does the work and a progress bar shows the estimate. The first AI run also downloads the model once (about 55 MB). Keep the tab open; Cancel stops it at any time.
- Compare the parts. Switch between the Original, Instrumental and Acapella tabs while it plays; the position is kept. Listen at the loudest part of the chorus and in the quiet verse. Those are where leftovers show up first.
- Download. Save the stem you want as a WAV (lossless) or an MP3 (smaller). If you need a short section only, trim it in the audio trimmer afterwards.
Photo: pedro furtado on Pexels
Which output do you need?
| Goal | Choose | Tip |
|---|---|---|
| Karaoke or singing practice | Instrumental | Backing vocals and harmonies are usually removed along with the lead |
| Cover song backing track | Instrumental | Check the final chorus; stacked vocals are the hardest part to remove |
| Remix, mashup or sample | Acapella | Cut the clean phrases you need rather than using the whole track |
| Remove music under someone talking (vlog, interview) | Acapella (voice) | This is the "background music remover" use case |
| Remove a voice over background music | Instrumental | This is the "background voice remover" use case |
| Transcribing lyrics | Acapella | Words are easier to hear without the drums |
Background voice and background music removal
The same separation works for clips that are not songs. Two common cases:
- You filmed something and music was playing in the room. Keep the voice output. The music will drop a lot, though loud music with lyrics is harder, because the model has two voices to tell apart.
- You have a track with a voice-over or announcer on top. Keep the instrumental output to recover a cleaner music bed.
If the problem is a steady hum, fan, traffic or hiss rather than music, a noise remover is the better tool. See how to remove background noise from audio.
What makes results better or worse
You cannot change the song, but you can pick the best version of it:
- Start from the best source you have. A WAV or a 320 kbps MP3 gives the model more detail than a 96 kbps rip from a video site. Compression artifacts get mistaken for "voice" or "instrument" and turn into a warbly sound.
- Studio mixes beat live recordings. On a live recording the vocal is captured by every microphone on stage and bounces around the room. There is no clean voice to find.
- Heavy reverb and delay stay behind. A model can remove the dry voice but the echo tail of that voice is often treated as part of the backing.
- Harmonies go with the lead. The model separates voice from non-voice, not lead from backing, so harmonies, ad-libs and a second singer usually end up in the acapella together. Choirs and vocal chops may be split inconsistently.
- Instruments that sing. Saxophone, violin and some synth leads overlap with the voice's frequency range and can leak into the acapella.
If a part has small leftovers, it often helps to lower the volume of the remaining artifacts in an editor, or to use a short fade. The volume booster guide covers how to raise the final level without clipping.
Photo: Alina Vilchenko on Pexels
Alternatives worth knowing
A browser tool is the quickest route for one song. For heavier jobs, these are good free options:
| Tool | Type | Good for | Trade-off |
|---|---|---|---|
| Shimga Vocal Remover | Browser | Fast, private, no install | Uses your device's power; long files take time on older machines |
| Ultimate Vocal Remover (UVR) | Free desktop app (open source) | Batch work, choosing between several models | Install and model downloads; a decent GPU helps |
| Demucs | Free command-line tool (open source) | Four stems: vocals, drums, bass, other | Needs Python and some terminal comfort |
| Audacity | Free desktop editor | Phase-cancellation "Vocal Reduction" effect, then editing | The built-in effect is the old method, so it thins out bass and drums |
A note on copyright
Removing vocals does not change who owns the song. An instrumental made from a commercial track is still that track. It is fine for private practice, but publishing a cover, a remix or a karaoke video can need permission from the rights holder or a licence through the platform. If you are making videos of your own music, our music video guide for independent musicians shows how to turn an instrumental into a visualizer video.
FAQ
Is the vocal remover really free? Yes, it is free to use on Shimga, and you do not need an account. The processing runs on your own device, so there is no upload and no queue.
Will the instrumental be perfect? Usually not perfect, but often very usable. Results depend on the recording. Clean studio pop and hip-hop tend to separate well; live recordings, old mono tracks and songs with lots of reverb leave more vocal traces.
Can I get just the vocals (an acapella)? Yes. Choose the acapella or vocals output. You can then trim the phrases you need in the audio cutter and check the tempo with the BPM finder before you build a remix.
Is my song uploaded anywhere? No. The file is read and processed by your browser on your device. The only download is the separation model itself, which your browser caches; your audio is never sent to Shimga's servers.
Can it remove background music from a video? Yes, if you extract the audio first. Use video to audio, run the vocal remover, keep the voice output, then put that audio back in your video editor.
Why does it run slowly on my phone? Source separation is a lot of maths. Phones have less memory and slow down when they get warm, so keep songs under about five minutes, and switch to Quick karaoke or trim the song first if the browser runs out of memory. A laptop or desktop computer will finish the same song much faster.
What can I do with the instrumental next? Make a karaoke or lyric video. Shimga's studio can turn the track into a beat-synced visualizer, and an LRC file adds timed lyrics. See how to make a lyric video.
Explore all of Shimga's free audio and video tools on the tools page.