Music Tech3 min read

Music Source Separation Models: Demucs, Spleeter, and What Came After

The models behind stem separation — how each works, how they compare, and which is behind the tool you're using.

Most separation tools are wrappers around a handful of models. Knowing which is which tells you what quality to expect.

Spleeter (2019)

Deezer's release that made separation mainstream. A spectrogram-based U-Net that predicts a mask for each source.

Strengths: fast, light, easy to run, still fine for rough work. Weaknesses: noticeable artifacts, especially watery vocals. Works on magnitude spectrograms and reuses the original phase, which is a significant part of the artifact.

Still widely embedded in older tools.

Demucs (2019–present)

Meta's model family, and the significant advance: it works directly on the waveform rather than the spectrogram.

That matters because waveform models handle phase natively rather than reconstructing it. The result sounds meaningfully more natural — less of the phasey quality that characterises spectrogram separation.

Hybrid Demucs added a spectrogram branch alongside the waveform branch, taking advantage of both. Hybrid Transformer Demucs (HTDemucs) added transformer layers for better long-range context.

Strengths: the best widely-available open model, natural-sounding output, actively developed. Weaknesses: computationally heavier, slow on CPU without a GPU.

Band-split approaches

Newer models split the spectrum into bands and process each with dedicated subnetworks, recognising that vocals, bass, and cymbals need different treatment. Strong results, particularly on vocals.

Open-Unmix

An open reference implementation designed for research reproducibility. Not the best-sounding, but valuable as a well-documented baseline.

What commercial tools use

Most separation products build on Demucs variants or proprietary models trained on similar principles, often with:

  • Larger private training sets — the single biggest quality factor
  • More stems — separating guitar and piano rather than lumping them into "other"
  • Post-processing to reduce artifacts
  • Server GPUs for speed

The differences between good commercial tools are usually smaller than their marketing suggests, because they share architectural lineage.

Running it yourself

Demucs is open source, free, and runs locally — which also means your audio never uploads. It needs Python and works far faster with a GPU, but CPU works.

For anyone processing a lot of material, or working with confidential audio, this is the option worth knowing about.

The practical takeaway

Separation quality across good tools is closer than it looks. The differentiator is usually what happens next: whether you get four files in a folder, or stems in a project you can immediately edit — which is where the actual work of fixing separation artifacts happens.

Related reading: how stem separation works, best stem separation tools, and AI music tools and stem bleed.

Frequently asked questions

What is the best stem separation model?

Hybrid Transformer Demucs is the strongest widely-available open model, working on both waveform and spectrogram representations. Most commercial separation tools use it, a variant of it, or a proprietary model built on similar principles.

What is the difference between Spleeter and Demucs?

Spleeter works on spectrograms and is fast but produces more artifacts. Demucs works on the raw waveform, which preserves phase information and sounds noticeably more natural, at higher computational cost.

Can I run stem separation models myself?

Yes. Demucs is open source and runs locally with Python. It needs a reasonably modern machine and is much faster with a GPU, but it works on CPU and costs nothing.

Start making music in Veena

Free, browser-based, no downloads required.

Try Veena Free