Remove Background Music from Audio

How to Remove Background Music

Removing background music from a recording takes three steps. The AI does all the heavy lifting — you just upload, choose the right mode, and download.

1. Upload your audio file. Drag and drop your recording into the converter above, or click to browse. The tool accepts MP3, WAV, FLAC, OGG, M4A, AAC, WMA, and AIFF. Use the highest-quality source file you have — a lossless WAV or FLAC will produce cleaner separation than a compressed MP3.
2. Select "Vocals Only" mode. This is the critical step. The Demucs AI separates your audio into four stems: vocals, drums, bass, and other instruments. Vocals Only mode extracts just the vocal stem — which contains all human speech and singing — and discards the three instrumental stems. The background music ends up in those discarded stems, leaving you with clean dialogue.
3. Download the vocals track. Once processing completes, download the result. The output file contains your speech or vocals with the background music removed. You can use it directly or import it into your audio or video editor to replace the original mixed track.

When You Need to Remove Background Music

This tool solves a specific problem: you have a recording where the speech is good, but unwanted music is playing in the background. Here are the most common scenarios.

• Podcast cleanup. A guest recorded their side of the conversation with music playing in their room, or a co-host had a Spotify playlist running that bled into their microphone. The speech is perfectly usable, but the background music makes the episode sound unprofessional and creates potential copyright issues. Running the audio through Vocals Only mode strips the music while preserving the conversation.
• Interview recordings. Interviews conducted in cafes, restaurants, or events often pick up background music from the venue's sound system. The interviewee's answers are clear enough to understand, but the ambient music is distracting and makes the recording hard to use in a documentary, news piece, or article. AI separation isolates the voices from the venue soundtrack.
• Video narration with soundtrack. You recorded a voiceover or narration over a video that already had background music baked into the audio track. Now you need the narration without the music — perhaps to re-edit the video with different music, or to use the narration in a different context. Demucs separates the spoken narration from the underlying soundtrack.
• Voiceover extraction from video. A training video, explainer, or presentation has a narrator speaking over background music. You want to reuse the narration in a new project, translate it, or transcribe it accurately. Extracting clean speech without the music makes transcription far more accurate and gives you a usable isolated voiceover track.
• Cleaning up recordings with background TV or radio. Someone recorded a voice memo, phone call, or home video while a TV show, radio station, or music stream was playing in the background. The background audio is distracting and may contain copyrighted content. The AI can remove the musical components, significantly cleaning up the recording.

Speech vs Music Separation

Understanding how the AI separates audio helps you set realistic expectations for the output quality.

Demucs is a deep neural network trained on thousands of hours of music. It learned to decompose mixed audio into four stems: vocals (any human voice — singing or speaking), drums (percussion), bass (bass guitar, synth bass, low-frequency instruments), and other (everything else — guitars, keyboards, strings, synths, sound effects). When you select Vocals Only, the model reconstructs just the vocal stem and discards the rest.

This means the AI removes all non-vocal sounds, not just "music" in the traditional sense. Here is what gets separated:

• Removed: background music, instrumental loops, soundtrack, jingles, guitar, piano, synthesizers, drum beats, bass lines, ambient music beds.
• Kept: speech, singing, humming, laughter, vocal breaths, lip sounds — anything produced by the human voice.
• Partially removed: ambient noise, room reverb, wind, traffic, air conditioning hum. These non-musical, non-vocal sounds do not fit neatly into any of the four stem categories. The AI handles them inconsistently — some ambient noise ends up in the vocals stem, some in the other stem. You will get a cleaner recording, but do not expect total ambient noise elimination.

The practical takeaway: if your recording has speech mixed with music, the separation will be very effective. If the unwanted sound is non-musical ambient noise (traffic, wind, HVAC), the results will be partial. For pure noise reduction without music separation, a dedicated noise-reduction tool is more appropriate.

Tips for Clean Speech Extraction

The AI does most of the work, but the quality of your input directly affects the quality of the output. Follow these guidelines for the cleanest possible speech extraction.

• Use the highest quality source file. WAV and FLAC files preserve all audio detail, giving the neural network the most information to work with. If you only have an MP3, use the highest bitrate version available. A 320 kbps MP3 will separate better than a 128 kbps version of the same recording because it retains more spectral information that the AI uses to distinguish speech from music.
• Ensure the speech is louder than the music. AI separation works best when the target signal (speech) is the dominant component. Recordings where speech and music are at similar volume levels produce good results. Recordings where music is significantly louder than the speech are harder — the AI may lose some speech detail along with the music. If possible, adjust the mix before processing so the speech sits on top of the music.
• Minimize other noise sources. Background music is what you want to remove, but other noise layers (room echo, wind, hiss) add complexity. The AI handles one separation task very well — splitting vocals from instruments. Adding noise on top of music on top of speech makes all three harder to untangle. Record in a quiet environment when possible, even if music is unavoidable.
• Trim to the relevant section. If only part of your recording has the background music problem, trim the file to that section before uploading. Shorter files process faster and you avoid re-processing sections that are already clean. You can rejoin the segments afterward in any audio editor.
• Check both the vocals and instrumental outputs. Sometimes a small amount of speech leaks into the instrumental stem, or a small amount of music leaks into the vocals stem. Listening to both outputs helps you identify any separation artifacts. If the vocals stem has music bleed, try processing the file again — the AI can produce slightly different results on a second pass.

Alternative: Extract Audio from Video First

If your source material is a video file (MP4, MOV, AVI, MKV), you need an extra step before the vocal remover can help. The tool processes audio files, not video. Here is the workflow:

1. Extract the audio track from your video. Use a tool like FFmpeg (ffmpeg -i video.mp4 -vn -acodec pcm_s16le audio.wav) or any online video-to-audio converter. Extract as WAV for the best quality. If the video has multiple audio tracks (e.g., narration on track 1, music on track 2), you may already have a clean separation and do not need AI at all — check your video editor's audio track settings first.
2. Upload the extracted audio to the vocal remover. Select Vocals Only mode and process. The AI will separate the speech from the background music in the extracted audio track.
3. Replace the audio in your video editor. Import the cleaned vocal track back into your video editing software (Premiere Pro, DaVinci Resolve, Final Cut Pro, CapCut, or any editor). Mute or delete the original audio track and sync the clean vocals track in its place. Most editors let you snap the new audio to the timeline start position for perfect alignment.

This three-step workflow is standard for video producers who need to clean up interview footage, remove copyrighted music from user-generated content, or isolate narration for re-editing. The extra step of extracting audio first is necessary because video files contain visual data that the AI does not need and cannot process.