FileKeeps
Media

How to Make a Karaoke Track (Remove Vocals from a Song) with AI, in Your Browser

Make a backing track for karaoke practice or covers. An AI source-separation model runs inside your browser — no upload — and you get the isolated vocals too.

FileKeeps2026-08-14 · 5 分钟阅读

You want to practice a favorite song but there is no instrumental, or you want to record a cover without the original vocal in the way — what you need is a backing track (an MR, as karaoke fans call it). This used to mean installing studio software or uploading your music to some unfamiliar site. Now an AI model can do the separation right inside your browser, and the file never leaves your device.

What you need: up-to-date Chrome or Edge on a computer

AI separation is computationally heavy, so it needs GPU acceleration (WebGPU). Open it in a current Chrome or Edge on a desktop or laptop — older browsers and most phones won't work. The first run downloads the separation model (66MB) once and reuses it afterwards.

Three steps

  1. 1Open the Vocal Remover and add a song file (MP3, M4A, WAV and so on, up to 12 minutes).
  2. 2Click "Split vocals & backing". The song is processed in six-second segments with a progress bar.
  3. 3Listen to the two results — backing track and vocals — and download each as WAV or MP3.

TipIf the result disappoints, try a better-quality source file — separation quality depends heavily on the source.

You get the vocal track too, not just the backing

The output is two tracks. The backing track (MR) is your base for karaoke practice and cover recordings; the isolated vocal is great for studying melody and phrasing, or just listening a cappella. The backing track's loudness is automatically matched to the original, so you don't need to boost it yourself.

How this differs from the "phase inversion" trick on free sites

Many free vocal-remover sites subtract the right channel from the left (L-R phase inversion). It's instant, but it has a built-in flaw: bass, kick and lead instruments mixed in the center vanish along with the vocal, and vocal reverb stays behind. AI separation identifies the vocal in the spectrum and lifts it out, so the instruments survive — which is why the backing track doesn't come out hollow.

Honest limits

  • Stacked harmonies, live recordings and low-quality sources can leave vocal traces or reverb behind.
  • Depending on your hardware, a 3-4 minute song takes tens of seconds to a few minutes.
  • Use the stems for personal practice. Separation doesn't change the song's copyright — publishing or distributing them needs the rights holder's permission.

Turn one song file into a backing track and an isolated vocal.

Open Vocal Remover