Song Splitter
← Back to blog
April 4, 2026 · 4 min read

How to Remove Vocals from Any Song in 2026

Create a clean instrumental or isolate just the vocals — step by step, no expensive DAW required.

Whether you want a karaoke backing track, a clean instrumental to remix, or an acapella to sample — removing vocals from a finished song is easier than ever in 2026 thanks to AI.

⚠️ Copyright reminder: Before removing vocals from a song for use in a project, make sure you own the track or have a licence to create derivative works. Many streaming platforms prohibit redistribution of modified copyrighted tracks.

Method 1: AI Stem Separation (Best quality)

This is the modern, recommended approach. Tools like Song Splitter use deep learning (HTDemucs) to separate the vocal track from all other instruments with high accuracy.

  1. Go to Song Splitter — stemsplits.com and sign in with Google.
  2. Upload your audio file — Supported formats: MP3, WAV, FLAC, OGG, AAC, M4A. Free tier: max 25 MB / 15 min. With GPU credits: max 50 MB / 30 min.
  3. Wait for processing — The AI runs on GPU and typically finishes in under 60 seconds for a 4-minute track.
  4. Download your stems — You'll get four separate stems: Vocals, Drums, Bass, Other (everything else). For an instrumental, download the Drums, Bass, and Other stems — then mix them back together using the custom mix download feature.

Method 2: Phase cancellation (Old approach, low quality)

The old trick — still found in some tutorials — is to take a stereo track and invert the phase of one channel. Since vocals are often recorded in mono (and appear equally in both channels), they cancel out, leaving the stereo instruments.

Why to avoid it: This only works on tracks where vocals are perfectly centred in mono. Any stereo processing, reverb, or panning on the vocals ruins the cancellation. The result is usually a hollow, unpleasant sound. AI is vastly superior.

Tips for better results

  • Use the highest quality audio file you have — a 320 kbps MP3 or lossless WAV will give better stems than a 128 kbps file.
  • Very dense mixes (lots of instruments on top of each other) are harder for the AI to separate perfectly — some bleed between stems is normal.
  • The "Other" stem contains guitars, keys, synths — everything that isn't vocals, drums, or bass.
  • Use the A–B loop feature in Song Splitter to audition specific sections before downloading.

Try it free → unlimited CPU separations at no cost.