TresFastAudioMaster

400 tracks in. The weekend stays yours.

Separating one track is easy. Separating a library is a weekend you are not getting back. Point this at the folder and go to bed.

Get the free engine

How it works

Folders, drives, network shares

Point it anywhere. It works through every track without you sitting there.

Tempo and key in the tags

Camelot too. Nothing to re-analyse after the fact — Rekordbox and Serato pick them up the moment you import.

Resumes where it stopped

An interrupted run picks up rather than starting over. Nothing is done twice.

Fails in four seconds, not four hours

Disk, models and drives are checked before a long job starts.

The same performance three times, lined up in time. The vocal stem is silent between phrases because the vocal is.

$24 once

Windows today
Nothing uploaded
No per-track fees
Two machines

Get the free engine Tell me when it is on sale

It is not on sale yet. One message the day it ships, and nothing else. The engine on its own is free to download in the meantime.

OP.STEM.BATCHMODEL.MBR.MITD:\Music\Dancehall

✓Chaka Demus & Pliers - Murder She Wrote[vocals][instrumental]98 BPM8A

✓Mr Vegas - Heads High[vocals][instrumental]96 BPM11B

✓Beenie Man - Who Am I[vocals][instrumental]94 BPM5A

✓Spragga Benz - Pull[vocals][instrumental]97 BPM8A

▶Sean Paul - Gimme The Lightseparating

Elephant Man - Pon De River · Wayne Wonder - No Letting Go · Buju Banton - Champion · Baby Cham - Ghetto Story — queued

147 of 400 done · running 4h 12m · resumes if interrupted

What you receive

Not a pile of files called output_1.wav. 320 kbps, named the way your collection is already named, with tempo and key written into the tags so Rekordbox and Serato pick them up the moment you import.

Artist - Title [vocals].mp3
Artist - Title [instrumental].mp3

The model doing the separation is free and open, and we say which one it is. What a licence buys is the left column becoming the right one while you are not at the desk.

Measured, not claimed

Almost every separator claims studio quality. Almost none says what it was measured against. Ours was tested on a dancehall riddim, where the official instrumental exists and can be used as the answer sheet.

5×

less of what should not be there than leaving the track alone

0.04

of the backing track left in the vocal, where 1.0 would be all of it

44

voiced tracks measured against one official instrumental

One riddim, one official instrumental, 44 tracks. That is the method and the baseline, and it does not prove anything about a genre it was not run on.