How to separate vocals from a song on iPhone or Android
To separate vocals from a song on iPhone or Android, import it into Denoise, choose AI · Music, pick Vocals to keep the voice or Music to remove it, and tap Extract vocals or Remove vocals. The separation runs on your phone with an open-source AI model, so you get an a cappella, a karaoke-style instrumental or a voice without its background music, without uploading anything.
Last updated · By Routes Software, the makers of Denoise
Separate vocals in three steps
Import the song or video
Tap + and choose Import from Files or Import from Gallery, or share it to Denoise from another app.
Choose AI · Music
Pick AI · Music under Engine. The first time, it downloads the separation model. Then choose Vocals to keep the voice, or Music to keep everything but the voice.
Separate and save
Tap Extract vocals or Remove vocals, compare with Original, then tap Share Cleaned. For a video, the separated sound goes back into the video.
Hear the difference
What you can make with it
A karaoke or backing track
Choose Music and tap Remove vocals: the singer goes and the instrumental stays, ready to sing or play along with. Backing vocals count as vocals to the model, so they usually go with the lead.
An a cappella for a remix or a cover
Choose Vocals and tap Extract vocals to keep only the voice. Bring it into your music app to remix it, or listen to it alone to learn the melody and the words.
A voice without its background music
A vlog, a talk or a lecture with music under the voice works the same way: choose Vocals, and the speech stays while the music goes. The model was built for singing, so listen to the result: other sounds, such as wind or traffic, can end up on either side. For noise rather than music, see removing background noise from a video.
A music video or a concert clip
Import the video itself. Denoise separates its soundtrack and puts the result back into the same video, so you can share the clip with only the voice or only the band.
Good to know
- AI · Music uses SCNet-large, an open-source separation model, downloaded once (about 75 MB). Android phones with less memory use Meta's HTDemucs instead (about 125 MB).
- Separation is very good but not perfect: a little of the music can remain under the voice, and reverb on the singer can stay in the instrumental.
- It takes longer than noise removal, because the separation model is large and runs on the phone's processor.
- Respect the rights of the song's owners when you share what you make.
Frequently asked questions
How do I separate vocals from a song on iPhone?
Install Denoise from the App Store, import the song from Files or your gallery, choose AI · Music and Vocals, and tap Extract vocals. It runs on the iPhone itself, and after the first model download it works offline.
How do I remove the vocals from a song on Android?
Install Denoise from Google Play, import the song, choose AI · Music and Music, and tap Remove vocals. What's left is the instrumental, ready to use as a karaoke track.
Can I get both the vocals and the instrumental?
Yes, in two runs: one with Vocals and one with Music. Each run gives you one part.
Can it remove background music from a video and keep the speech?
Yes. Choose Vocals: speech counts as voice, so it stays and the music goes, and Denoise puts the result back into the video. Check the result, since the model was built for singing.
Does it need an internet connection?
Only once, to download the separation model. After that it works offline.
Is it free?
Separating and listening are free, with no account. Saving or sharing the result is paid: 3 exports cost $1.99, and Denoise Pro $2.99 a month or $14.99 once ($11.99 on Google Play). These are US prices from October 2026; the store shows yours before you pay.