Stem Splitter

Sampling from a full mix means fighting the vocals. Stem Splitter pulls a song apart into drums, bass, vocals and everything else, on your own computer, so you can sample the part you want. A 6-stem option adds guitar and piano.

Stem Splitter is desktop only and part of Pro, and the 14-day trial includes it. It runs on macOS, Windows and Linux. I built it for sampling rather than mastering, so results vary from song to song.

Download a Model

The first time you open Stem Splitter, it asks you to download a separation model. You only do this once for each model.

1. Open Stem Splitter.
2. Pick a model from Separation Model:
   • 4-stem (fast), 166 MB: drums, bass, vocals and other. The quickest, and the usual choice where it’s available (Macs with Apple silicon and Windows).
   • 4-stem, 84 MB: the same four parts, on every desktop platform.
   • 6-stem, 55 MB: adds guitar and piano. Slower, and results vary more on full mixes.
3. Read the model’s licence, then tick I have read and accept this model’s licence.
4. Select Download Model.

Models download from Hugging Face (huggingface.co), and the app checks each file before it installs it. After that, separation runs on your computer and works offline. Your audio is never uploaded.

To download another model later, open Download Models at the top of Stem Splitter. To remove one, use the bin button next to it in the Model list. With more than one model installed, pick the one to use in that list.

Split a Song

1. Choose a song: select Browse for an Audio File, drag a file onto Stem Splitter or select Choose from Library. WAV, MP3, FLAC, AIFF, OGG and M4A files all work. If the Sample Editor has one sample loaded, Use Current Editor Sample brings that in.
2. To split just part of the song, drag the markers on the waveform. Select at least one second. Preview Selection plays it. With the markers at the ends, the whole song is split.
3. Choose the Bit Depth: 16-bit (the default, and a third smaller) or 24-bit. EP devices accept both.
4. Select Extract Stems (Extract Stems from Selection if you’ve made one).

A full song usually takes one to a few minutes, depending on your computer. You can keep using the app while it runs: Stem Splitter carries on in the background and tells you when the stems are ready.

Use the Stems

When the split finishes, you see Stems Ready with a row for each stem.

• Play All Stems plays them together. Use the mute and solo buttons to hear one part or a mix.
• Play one stem on its own with its play button.
• Preview the stems, then select Save to Library to keep them. They go into your sample library as a pack called “(song name) Stems”. Until you save, the stems only last for this session.
• Create Instrumental mixes everything except the vocals. Once you’ve saved, it’s added to the pack. Before that, you save it as a WAV file.
• Open in the Sample Editor sends a stem to the editor to trim, chop and upload.
• The … menu on a stem has Send to Batch Processor and Export as WAV.

After you save, you can tick several stems and use Send Selected to Batch Processor, or export the whole pack with Export Pack as ZIP.

To try again, select Split This Song Again (for example with a different selection), or Choose a Different Song.

From the Sample Editor

The Extract Stems button in the Sample Editor’s toolbar opens Stem Splitter. If the editor has one sample loaded, select Use Current Editor Sample to split it.

Import a Model File

If you already have a compatible model file, you can import it instead of downloading one.

1. Select Import an Existing Model File (on the setup screen or under Download Models), or Import Another Model below the Model list.
2. Confirm you have the rights to use the file, select Choose a Model File and pick it.

The app imports .bin model files, the same kind it downloads for the 4-stem and 6-stem models, such as ggml-model-htdemucs-4s-f16.bin and ggml-model-htdemucs-6s-f16.bin. The fast model can only be downloaded. Importing a 4-stem file replaces your 4-stem model, and a 6-stem file replaces your 6-stem model. The app checks the file before installing it. If it can’t use a file, it tells you why: the file isn’t a model it recognises, it’s incomplete or it’s a type the app can’t run yet.

Troubleshooting

• Stem Splitter isn’t in the sidebar: it’s desktop only, so it doesn’t appear on phones or tablets. On a computer, reinstall EP-PatchStudio from the download page, and email me if it’s still missing.

• A download fails or stops: the Model Download notification says why:

  • “Not enough disk space. This download needs (size) MB free.”
  • “Download failed: … Try again.” or “Download interrupted: … Try again.”
  • “Downloaded model failed checksum verification. Try again.”
  • “Download cancelled. Partial model removed.”

  Check your connection and disk space, then try again. A failed or cancelled download is removed, so nothing half-finished is left behind.

• “Separation Failed”: select Show Error Details to see what happened, then Retry or choose a different file. If it keeps happening, send me the diagnostic logs.

• Long songs: Stem Splitter can split up to 15 minutes at a time. Drag the markers to select 15 minutes or less, or trim the audio first. On long tracks Play All Stems is turned off to save memory. Opening a stem in the editor, batch processing and export still work.

Good to know

• Stems come out full length. Trim or chop them in the Sample Editor before you send them to your device.
• Vocals can bleed into the other stems. Drums and bass usually come out cleanest.
• Macs with Intel processors and Linux use the 4-stem and 6-stem models. The fast model needs a Mac with Apple silicon or Windows.