← Volver

Cómo funciona Eliminador de voz Vocal isolation Preguntas frecuentes

Background Music
Eliminador

Remove the background music from a recording and keep the voice clean underneath it

MP3, WAV o FLAC · hasta 20 MB y 5 minutos · 25 créditos por pista

How to remove background music

  1. Upload the recording. MP3, WAV or FLAC up to 20 MB and five minutes. If the audio is inside a video, export the audio track from your editor first and put the cleaned result back on the timeline afterwards.
  2. The AI separates music from voice. The model splits the recording in two: everything it reads as music on one track, everything it reads as voice on the other. Nothing is filtered or EQ’d away, so the voice is not left thin and hollow the way a notch filter leaves it.
  3. The music opens muted. This page loads with the music fader at zero, so you hear the voice on its own straight away. If the removal cut something you wanted, lift the music fader part-way instead of all the way down — a quiet bed often sounds better than a hard removal.
  4. Save the clean voice. WAV to take back into a video editor or a podcast edit, MP3 for anything lighter.

When you need the music gone

  • Videos with cleared audio problems — a clip whose backing music has a copyright claim on it, where the speech is the part you need.
  • Interviews and vlogs — footage shot in a bar, a shop or a gym with music playing over the conversation.
  • Voiceovers — a recorded read that came back already mixed against a music bed you did not want.
  • Lectures and talks — a recording where walk-on music sits under the opening minutes.
  • Songs — the same job, done to a record: the instrumental removed and the vocal kept.
  • Sample cleanup — a spoken phrase or a film line you want without the score behind it.

Preguntas frecuentes

Does it work on speech, or only on singing?
It is a music separation model, so it was trained on sung voice — and it handles speech well because a speaking voice sits in the same range and has the same harmonic shape. Where it struggles is the reverse case: very quiet speech under very loud music, and music with no melodic content at all, like a solo drum loop, which the model has little to separate the voice from.
Is this the same as noise removal?
No, and using the wrong one makes things worse. Noise removal targets steady, tonally flat sound — hiss, hum, air conditioning, traffic. Background music is none of those: it is pitched, structured and constantly changing, so a noise reducer either ignores it or chews the voice apart trying. Music needs separation, not reduction.
Will any of the music be left?
Sometimes a trace, and it is predictable where: the moments the music is loudest relative to the voice, and any instrument that overlaps the voice closely — a lead synth or a string line in the same register. It is not a filter with a threshold you can push harder; if a section comes back with too much left in it, that section was genuinely hard to separate.
Can I keep the music instead and remove the voice?
Yes — that is the same separation with the other fader down, and it has its own page: the karaoke maker exports the music on its own. Or use the vocal remover and set both faders yourself.
Do I need an account?
You need to be signed in, because the separation runs on a server rather than in your browser and it costs 25 credits a track. Nothing you upload is published; the results go to your own private library.

Una sola separación, seis puertas de entrada

These are all the same AI separation with the faders in a different place. Whichever one you start from, the split runs once and you get both tracks.