How to Make a Karaoke or Instrumental Track: How Vocal Removal Works (and Its Limits)

Want to practice karaoke or record a cover and wish you had just the instrumental? Our Vocal Remover & Karaoke Maker splits a single song into instrumental, vocals, drums and bass for free. Nothing is uploaded; it all runs in your browser. It works differently from paid AI separation services, though, so some songs work well and some don't. Here is an honest look at how it works and where it falls short.

How it reduces the vocals

Most songs put the lead vocal in the center of the stereo image and spread instruments like guitar and keys to the left and right. So sound that is identical in both channels is likely center sound. The tool slices the audio into very short time frames and frequency bins (STFT), measures how similar left and right are and how balanced they are in each bin, and estimates the "center ratio". That share goes to the vocal track and the rest to the instrumental.

Unlike the old "invert one channel and add" trick, it decides bin by bin. It also protects low sounds that sit in the center but belong in the instrumental, like bass and kick drum (below about 120 Hz). Using the fact that drums are short bursts and harmonic instruments are sustained (HPSS), it splits the instrumental again into drums and the rest, and gathers the low band into a bass track.

How to use it

1. Add a song (MP3, WAV, M4A and more) to the vocal remover. It handles up to 8 minutes at a time.
2. Start with the strength slider at the default 90%. Raise it if vocals remain; lower it if the instrumental sounds thin.
3. Listen to instrumental, vocals, drums and bass separately, then download what you need as WAV or MP3, or all at once as a ZIP.
4. If the key is too high to sing, put the instrumental into Change Pitch and Tempo and shift it by semitones.

Songs that work well, and songs that don't

• Works well: songs with a clear vocal dead center and instruments spread wide. Acoustic arrangements and older pop recordings (pre-2000s) usually work well.
• So-so: vocals with lots of reverb or stereo effects, backing vocals panned left and right, modern dance tracks where synths and vocals sit in the same place. Faint vocal reverb can stay in the instrumental.
• Doesn't work: mono audio has no left/right information, so vocals and instruments can't be separated. In that case only drums and bass are split, and the tool tells you so.

Paid AI separation services use models trained on huge numbers of songs to recognize "this is a human voice", so they are often cleaner on reverb-heavy tracks. We couldn't find a reliable open model licensed for commercial use, so we built this with our own signal processing instead. In return: no upload, no sign-up and no usage limits.

Check the copyright

Practicing on your own with an instrumental you made is one thing, but posting or distributing it on YouTube or social media touches the original song's copyright. For public use, use an official instrumental or karaoke track, or one the rights holder has licensed.

Tips for better results

Start from a high-quality file (320 kbps MP3, FLAC or WAV). Heavily compressed files smear the stereo information and leave more vocals behind. If vocal reverb remains, raise the strength gradually rather than running the instrumental through again; processing twice tends to eat into the instruments. For practice, a little leftover vocal can even serve as a pitch guide. To learn the original key and chords, try the BPM & Key Finder and Chord Finder. See also the limits of vocal removal.

Sources and date checked

Checked: October 6, 2026

• FreeSoundTools vocal remover (own signal processing: STFT center extraction, HPSS)

This article is general information. Devices and services can change over time.

Frequently asked questions