---
title: "How to Remove Voice or Music From a Video"
description: "Mute a video, keep the voice and drop the music, or keep the music and drop the voice. What each job needs and how to check the result."
summary: "Removing sound from a video is three different jobs. Muting drops the whole soundtrack. Keeping the voice and removing the music, or the reverse, needs an AI model that splits a mixed track into layers. Results depend on the material, so listen to the full clip before you publish it."
url: "https://www.gpt-video.com/blog/how-to-remove-voice-or-music-from-video"
published: "2026-10-07"
updated: "2026-10-07"
author: "Alessio Battagliero"
topic: "AI video editing, explained"
keywords: "vocal remover, background music remover, audio separation, video editing"
---

# How to Remove Voice or Music From a Video

Removing sound from a video is three different jobs. Muting drops the whole soundtrack. Keeping the voice and removing the music, or the reverse, needs an AI model that splits a mixed track into layers. Results depend on the material, so listen to the full clip before you publish it.

## Three different jobs hide behind one search

People who search for a way to remove voice or music from a video usually want one of three things. They want to mute the video completely. They want to keep the voice and lose the music. Or they want to keep the music and lose the voice.

These are three different jobs, and they need three different tools. Muting is trivial. The other two require a model that can pull a mixed soundtrack apart, because a voice and a song are mixed into a single audio track and cannot be cut out with a simple filter.

Picking the right job first saves time. If you only need silence, you do not need artificial intelligence at all. If you need to separate sounds, you need a source separation model, and you should expect imperfect results on hard material.

## Muting a video completely

A video file is a container that holds separate tracks. One track carries the picture and another carries the sound. To mute a video, the sound track is left out of a new file and nothing else changes.

The free [audio remover](/tools/audio-remover) does this in your browser. When the picture is already in a format the new file can use, such as H.264 in an MP4, the frames are moved across without being compressed again. That means the picture quality does not change. Other codecs, like HEVC or AV1, are compressed once more, which has the usual small cost of a single encoding pass.

Muting also drops subtitle tracks and chapters, because only one picture track is written. Keep your original file if you need those. The picture keeps its frame size, its frame rate and its length, and the tool offers no menu to change them. It only takes the sound away.

## Keeping the voice and removing the music

This is the most common request for creators. You filmed a clip, but a song plays behind it that you cannot use on a platform, or the music is too loud and fights with the speech.

The [background music remover](/features/background-music-remover) uses source separation. A model listens to the mixed soundtrack and splits it into layers, then keeps the layer that holds speech. You get the voice on its own, and you can add your own music afterward.

Results depend on the material. Clear speech over steady music separates well. Singing, crowd noise or speech that sits in the same range as an instrument is harder, and you may hear a thin or watery sound on the voice. Listen to the whole result with headphones before you publish it.

## Keeping the music and removing the voice

The reverse job is the one karaoke fans know. You keep the instrumental and drop the singer or the speaker.

The [AI vocal remover](/features/vocal-remover) splits a song into a vocal layer and an instrumental layer. You can give it a video and it works on the soundtrack directly, so there is no need to extract the audio first. The instrumental is useful as a backing track under a new voiceover.

Old tricks like phase inversion only work on some stereo recordings, and they often leave ghost traces of the voice. A trained model does a more complete job in one pass, though no method is perfect on every recording.

## How the separation works

A mixed soundtrack is one wave. Voice, drums, guitar and room noise are added together into a single signal, and there is no label on any part of it. That is why you cannot simply select the voice and delete it.

A source separation model is trained on many examples where the separate parts were known. It learns what speech looks like in the frequency picture of a sound, and what steady music looks like. At work, it estimates which part of each moment belongs to which source and rebuilds each layer on its own.

This is an estimate, not a perfect cut. Where voice and music share the same range of pitch, the model has to guess, and the guess shows up as faint traces or a slightly hollow tone. Good recordings guess better. A clean, loud voice over quiet music gives the model the most to work with.

## How to check the result before you post

Do not trust the waveform. Play the whole clip, and pay attention to three moments in particular.

Listen to the first few seconds, where a separation model has the least context and can stumble. Listen to loud passages, where leftover music tends to bleed through. And listen to quiet pauses, where you may hear small artifacts that were hidden under the original music.

If something sounds off, try a cleaner source. A higher quality file always separates better than one that has been compressed many times. If you are laying new music under a voice you have just isolated, keep it low, since the voice may already sound a little dry.

## Rights and platform rules

Removing music from your own recording is a normal editing step. Taking it out of someone else's video gives you no right to reuse the footage. The picture and the speech stay protected even when the soundtrack changes.

A common approach on short-form platforms is to drop the song and pick a track from the platform's own library. Those tracks are licensed for use there. If you add a voiceover afterward, our [AI voiceover](/features/ai-voiceover) feature can generate one. For the broader picture of how AI changes the editing process, read [what AI video editing is](/blog/what-is-ai-video-editing).