Vocal Remover
Remove vocals from a stereo song in your browser with centre-channel cancellation, keep the bass, or roughly isolate the voice. Download as WAV or MP3.
This is a simple phase trick, not AI source separation: it removes or keeps whatever is panned dead centre in the stereo mix. It works best on studio recordings with a centred lead vocal, and some reverb, backing vocals or centred instruments will remain or be affected.
- Length
- –
- Channels
- –
- Sample rate
- –
- L/R correlation
- –
How to remove vocals from a song
- Drop a stereo song on the drop zone (MP3, WAV, M4A, OGG, FLAC…). It is decoded in your browser and never uploaded.
- Check the file details: Channels must say Stereo. You can play the Original to compare later.
- Choose Remove vocals (karaoke / instrumental) or Isolate vocals (approx.).
- For karaoke, leave Keep the bass and kick drum ticked and adjust the Bass crossover (Hz) if needed (150 Hz is a good start).
- Pick the Format (WAV or MP3) and press Process.
- Listen to the Result player, then press Download.
How centre-channel cancellation works
In most studio mixes, the lead vocal is panned to the centre: the same signal is sent to the left and right channels. Guitars, keyboards, backing vocals and stereo reverbs are usually spread to the sides. If you subtract the right channel from the left, (L − R) / 2, everything that is identical in both channels cancels out, and what is left is the “side” signal: the instruments that differ between left and right.
That is the whole trick behind classic “karaoke” buttons on hi-fi systems, and it is exactly what this tool does. It is a phase trick, not AI source separation: the tool doesn’t know what a voice is, it only removes what sits in the middle.
The catch is that the bass guitar and the kick drum are also mixed in the centre, so plain cancellation leaves a thin result. With Keep the bass and kick drum, the tool splits the audio at the crossover frequency with 4th-order Linkwitz–Riley filters (two cascaded Butterworth biquads, rendered offline by the browser’s Web Audio engine). Below the crossover it keeps the mid signal (L + R) / 2, where the bass lives; above it, it keeps the side signal. The two bands add back without a bump or a dip. A higher crossover brings back more low end, but also more of the lowest part of a male voice.
Isolating the voice (approx.)
Isolate vocals (approx.) does the opposite with a short-time Fourier transform: for each small slice of time and each frequency, it keeps the mid signal and subtracts the size of the side signal from it. Sounds that are equally loud in both channels survive; wide sounds are turned down. Frequencies below 80 Hz and above 12 kHz are removed as they rarely contain voice. Expect a rough, “watery” result with some centred instruments left in: it is handy for practice or to hear a lyric, not for a clean a cappella.
Limits
- A mono file (or a stereo file whose two channels are identical) cannot work: subtracting two equal channels leaves silence. The tool tells you when this happens.
- Live recordings, old stereo mixes with the voice on one side, and heavy stereo reverb on the voice give weaker results.
- The result is mono (both channels are the same) and its level is normalised to −1 dBFS peak.
Frequently asked questions
Is my song uploaded to a server?
No. Decoding, processing and encoding all happen in your browser. The file never leaves your device.
Why can I still hear the singer faintly?
Reverb and echo on the voice are often spread in stereo, so they don’t cancel. Backing vocals and doubled vocals panned left and right also stay. This is the limit of the phase trick.
Why is the result mono?
The side signal (L − R) is a single channel. The tool writes it to both channels of the output so it plays evenly on any device.
What does "L/R correlation" mean?
It measures how similar the two channels are, from 1 (identical, mono) to 0 (unrelated). A value close to 1 means most of the mix is centred and little will remain after vocal removal.
Is this as good as AI vocal separation?
No. AI tools recognise voices with trained models and give much cleaner stems. This tool is instant, private and free of models, but it can only remove or keep what is in the centre of the stereo image.
Related tools
- Audio Converter Convert audio files to MP3, WAV, OGG, M4A or FLAC in your browser, or extract the sound of a video. Batch conversion, no upload, no sign-up.
- Audio Cutter Cut and trim MP3, WAV, M4A, OGG or FLAC files in your browser. Select on the waveform, add fades and export to MP3 or WAV. No upload.
- Audio Speed and Pitch Changer Slow down or speed up an MP3 without changing the pitch, or transpose a song by semitones. Preview, then export to MP3 or WAV, all in your browser.
- Volume Booster and Normalizer Increase the volume of an audio file in dB, normalize its peak or its loudness, with a limiter that prevents clipping. Export to MP3 or WAV in your browser.