Vocal remover and stem splitter: vocals, drums, bass and melody, offline

Drop a song and get its four parts as separate files: vocals, drums, bass and melody. Mute the vocals for a karaoke track, solo them for an a cappella, change the pitch of any part and export what you hear. The AI runs on your own computer - no upload, no monthly limit.

Video: separating a song into stems in MiniMax Converter (2:58).

How to use it

  1. Open Audio tools > More… > Stems, or drop one or more songs (a whole folder works too) on the Stem Splitter.
  2. Press Separate. Songs are done one at a time, with a progress bar and a time estimate. The first time, the separation engine and its model are downloaded once.
  3. Each song's four stems are saved in a folder named after the song, and the song appears under Separated songs.
  4. Double-click a finished song to open the Mixer: play the stems together, mute or solo any of them, and press Export mixdown to save what you hear as one file.

Karaoke, a cappella, practice tracks

Mute the Vocals and export: a karaoke instrumental. Solo them: an a cappella. Mute the Bass or the Drums to play along with the song yourself, or keep only the Melody. Export mixdown saves exactly what you hear as WAV, FLAC, MP3, M4A or OGG.

A real mixer

Every stem has Mute, Solo and a volume slider, and the waveform shows each stem in its own colour. Click or drag on the waveform to play from there; it redraws for the current mix and marks where the sum might clip. Normalize loudness keeps the song as loud as the original while you turn parts down, with a limiter so it never clips.

Change the pitch of any part

Each stem has a pitch knob: up to two octaves up or down in fine steps, heard at once while the song plays. Keep voice character keeps a shifted singer sounding natural, and every stem shows its musical key, which follows the knob. With one stem on Solo, drag across the waveform to select a part and change the pitch of just that part.

Private and without limits

Separation uses the well-known open-source Demucs model, on your own computer. Your songs are never uploaded, there is no monthly limit and no watermark. On a computer with a supported NVIDIA graphics card or an Apple Silicon Mac it uses the graphics chip and is much faster.

Many songs at once

Drop a whole album or a folder: the songs wait in a cue list and are separated one after another while you work with the finished ones. Right-click a finished song to copy all stems or a single stem to the clipboard, open its folder, or delete it. Stems you separated before come back when you drop their folder on Separated songs.

Settings

The format and quality of the stems (WAV, FLAC, MP3, M4A or OGG) and where they are saved are in Settings > Defaults > AI. The mixer's own settings (pitch engine, voice character, waveform, limiter and the format of Export mixdown) are in Settings > Defaults > Audio; the gear on the Stem Splitter screen opens them.

Questions and answers

Are my songs uploaded anywhere?

No. The separation runs on your own computer. Only the first time, the separation engine and its model are downloaded.

How good is the separation?

It uses Demucs (htdemucs), one of the best open-source separation models. Vocals and drums come out very cleanly on most songs; quiet or heavily layered parts can keep a little bleed from the other instruments.

How long does a song take?

With a supported NVIDIA graphics card or on an Apple Silicon Mac, usually well under a minute per song. On the processor alone it takes several minutes. The progress bar shows how long is left.

Which computers can run it?

Windows and Linux (64-bit) and Macs with Apple Silicon (M1 or newer). Intel Macs are not supported for stem separation.

Which files can I split?

MP3, WAV, FLAC, M4A, OGG, Opus, WMA, AIFF and the sound of MP4, MKV and WebM videos.

Get MiniMax Converter

Cross-platform desktop app. Linux free for non-commercial use; Windows & macOS: 7 days free, then a one-time €20 license. No subscription, no telemetry, no account.