How to remove vocals from a song

3 min read · Updated November 4, 2026

Removing the vocals from a song used to be a studio trick. Today an AI model can do it on a normal computer or phone in a few minutes. This guide explains how it works, what results to expect and how to get the cleanest instrumental possible.

Why the old trick rarely worked

For years the common advice was "phase cancellation": invert one stereo channel and subtract it from the other. Lead vocals are usually mixed exactly in the centre, so they cancel out. The problem is that bass, kick drum and often the snare are also in the centre, so they disappear too, and any stereo reverb on the voice stays behind as a ghost. The result sounds thin and hollow.

What AI separation does differently

A source separation model has learned from thousands of songs what a voice, a drum kit and a bass guitar look like in the spectrum. Instead of subtracting channels, it estimates each instrument as its own track. That is why you can get a full-sounding backing track with the voice gone, or the voice alone with everything else removed.

Step by step with Audio Splitter Pro

1. Open the free online stem splitter and select your song. MP3, WAV, M4A, FLAC and OGG files all work.
2. Choose No vocals to create only the instrumental parts, or keep All tracks if you also want the vocal track.
3. Press the split button and keep the tab open. A few minutes of waiting is normal, because the AI runs on your own device.
4. In the mixer, turn the Karaoke button on to hear the song without the voice, or mute the vocal track yourself.
5. For a single instrumental file, press Karaoke and use Export with Full mix ticked. For separate stems, tick the tracks you want. Files are saved as WAV.

Getting the best quality

• Start from the best file you have. A lossless or high bitrate source gives the model more detail than a low quality MP3.
• Choose High quality in the options if you are not in a hurry. It processes the song with more overlap between pieces, which smooths the joins.
• Expect small leftovers. Backing vocals, heavy reverb and effects can leave faint traces. That is normal for every separation tool.
• Dense mixes are harder. A thin acoustic recording separates almost perfectly, while a wall of distorted guitars is more challenging.

Common problems and fixes

• A faint voice is still audible. Try the High quality setting, or lower the other tracks slightly so the trace is hidden.
• The instrumental sounds thin. Check that the Bass and Other tracks are on and not muted, then export the Full mix again.
• The page stops or reloads. Close other tabs, try a shorter song, or use a computer. Phones have less memory than computers.

What it is good for

Karaoke nights, practising an instrument along with a backing track, making a cover, creating a clean instrumental for a video you have the rights to, or simply listening to how a song is built. Always make sure you have permission to use the music you process.

Frequently asked questions

Is the result perfect?

No separation is perfect. Modern models are very good, but you may hear small artefacts, especially on busy songs.

Does it work on any language?

Yes. The model separates sounds, it does not need to understand the words.

Is my song uploaded anywhere?

No. Everything is processed inside your browser.