Split into Parts18 October 2026

Separating vocals, instruments and noise with AI

Split into Parts, Pull Out a Sound and Clean Up: one song in six parts, a dog pulled out of a street, and a voice-over cleaned.

By Anand Murugan, who makes Chitra Studio

An audio clip can have many layers. Often you want only part of it: the vocals without the band, the band without the vocals, a dog bark without the street, a voice without the room.

Separation models pull a recording apart. Here are three ways to do it, on real sounds from my projects, with what each cost.

Split a song into its instruments

This one-minute song came from ElevenLabs Music v2.5. Split into Parts turned it into six sounds: vocals, drums, bass, guitar, piano and everything else. The whole split cost 1 credit.

Each part lands in your library as its own sound, so you can lay the instrumental under a clip, or use the vocals alone.

Pull out a sound by describing it

Some recordings are not music. Here, Pull Out a Sound was told what to find, "dog barking", and gave back two sounds: the dog, and everything else. 3.5 credits for fifteen seconds.

Clean up a voice-over

Clean Up is for speech: it takes out background noise and room echo, and evens the level. This 38-second voice-over cost 2 credits.

Tips

• Use Split into Parts for music, Clean Up for speech. Clean Up on a song keeps only the voice; Chitra Studio suggests Split into Parts instead when it can tell a sound is music.
• Name the sound you want plainly. "dog barking" was enough to find it.
• Keep everything. Each part is a new sound beside the original, which stays as it was.