Separate Vocals From a DJ Set or Live Show

A recorded DJ set, a live show or a podcast episode is an hour or more. Separation tools cap each run, and AudioForges takes up to thirty minutes in Clean, ten on the free result. The way to do a whole set is to cut it at the transitions and run the parts, and with Batch that is one drop, not twenty. This guide is that workflow, start to finish.

What you need

• The set as one file, or already cut into tracks. MP3, WAV, FLAC, M4A, AAC, OGG and AIFF all work, and video files such as MP4 are read for their audio.
• Each part under thirty minutes for Clean, ten for the free result, and under 500 MB. A ten-minute WAV at 44.1 kHz is about 100 MB, so WAV parts are fine.
• Any audio editor that can cut without re-encoding. Audacity, Ocenaudio, the export markers in rekordbox if you only need a few parts.

Cut the set at the transitions

1. Open the set in your editor and drop a marker where each track takes over. On a mixed set the cleanest cut is the first downbeat of the incoming track, after the outgoing one has faded.
2. Export each marker range as its own file. Keep the original format and sample rate; the parts are going straight into a separation model, so there is nothing to gain from converting.
3. Name the parts in order, 01, 02, 03, so the results sort the same way in your Library and in the DJ pack.
4. Check that no part is over the limit for the engine you are running. A long blend can be split in the middle of the mix; the model does not care where the cut is, only how long the file is.

Run the parts

1. Open the Stem Splitter (or the Vocal Remover for vocals and instrumental only) and drop all the parts at once. Up to twenty files go in as one batch.
2. Batches run in Clean, one song per part. If you only want a rough instrumental of one part to check something, drop it on its own and run it free.
3. Press Run. The parts upload one after another and run on the server. You can leave the page; signed in, every result waits in your Library for 30 days.
4. Open any result in Forge Mixer, scrub through it, and download the stems as WAV or MP3. Download all gives you every part in one go.

What a separated set gives you

Each part comes back with the vocal of that track in the vocal stem and the drums in the drum stem, with the transition intact at the edges. Because the parts were cut at the transitions, you also have every track of the set as its own file, which is what you need for a DJ pack and what a single hour-long run could never give you.

What people do with a separated set

Instrumental of a live show

Remove the vocal from every part, keep everything else, and use the parts as the bed for a video edit or a re-record.

Acapellas for edits

Take the vocal stem of each part and pull out the phrases you want, already labelled by track.

DJ pack

Every stem in one ZIP per part, named with the detected key and BPM, with a rekordbox.xml so the folder imports straight into rekordbox.

Questions people ask

What is the length limit?

Thirty minutes per run in Clean, ten on the free result, and 500 MB per file. A long set goes through as parts. The picker tells you before the upload starts if a file is over the limit.

Why not run the whole set at once?

A separation model treats the input as one song and the GPU time grows with the length. Cutting at the transitions keeps each run fast, keeps every result inside the limit, and gives you the tracks back as separate files, which is what a DJ pack needs anyway.

Does a long part cost more?

A part over ten minutes counts as one song per ten minutes, and the run button shows the total before you pay. The free result costs nothing.

Can I split the parts into four stems?

Yes. Pick 4 stems instead of Vocals and each part comes back as vocals, drums, bass and other, in sync.

Do I have to keep the tab open?

No. The run continues on the server. Sign in first and every result lands in your Library for 30 days.