Stem Separation

Stem separation splits a finished song into its parts — vocals, drums, bass, and other — so you can remix and recombine them. It's one of the most powerful things you can do in Silverdaw: pull an acapella out of a track, isolate a drum break, or drop the bass line onto its own track.

Like everything else in Silverdaw, it's non-destructive: your original file is never changed.

First time only: a one-off download

The first time you separate stems, Silverdaw downloads the separation model it needs — a one-time download of about 1 GB, so you'll need an internet connection just that once. It's stored on your computer and reused for every separation afterwards, so from then on separation works offline. The models are hosted on Hugging Face.

Separating a track

Right-click a clip and choose Transform ▸ Separate Stems. The Separate Stems dialog opens with no stems selected: tick the parts you want — Vocals, Drums, Bass, and Other — choose a Quality setting, then click Start. Start stays disabled until you've ticked at least one part.

The first time you separate stems, if the model Silverdaw needs isn't on your computer yet, it asks to download it once. Separation begins only after that download finishes; every separation after that starts straight away.

Silverdaw then analyses the audio and produces the separate parts, each on its own track for you to remix. Every stem keeps the original clip's tempo, key, and artwork, so it stays in step with the rest of your project.

Removing reverb & echo from vocals

When you tick Vocals, an extra option appears beneath it: Remove Reverb & Echo. Turn it on to clean room reverb and slap-back echo off the vocals stem as it's separated, leaving a drier, more upfront vocal. Pick how hard it works with the Light, Medium, and Strong strength buttons.

This is a fresh choice each time you separate — it isn't saved as a preference — and, like everything else in Silverdaw, it's non-destructive. The option only applies to the vocals stem, so it's available only when Vocals is ticked.

TIP

This per-run reverb and echo removal is separate from the per-stem Cleanup strengths in Edit ▸ Preferences ▸ Stems, which tidy the finished stems after separation.

The separation models

Silverdaw has two kinds of separation model, and you manage both in Edit ▸ Preferences ▸ Stems:

• High-quality RoFormer models — these give the best results. They total about 1 GB, are downloaded once with the Download models (~1 GB) button, and are then used automatically. The Separation models section shows their status (Not downloaded, Partly installed, or Installed).
• A backup model — a lower-quality fallback. Silverdaw fetches it automatically when a separation needs it — for example, when you select Other without selecting all four parts.

Using models you already have

If you've already downloaded the models — for example on another machine, or because someone shared them with you — you can point Silverdaw at them instead of downloading again:

1. Open Edit ▸ Preferences ▸ Stems.
2. Under Separation models, find "Already have a model? Point Silverdaw at the folder that contains it instead of downloading it again."
3. Click Locate… next to Vocal model and choose the folder that contains the vocal model.
4. Click Locate… next to Drums and bass model and choose that model's folder.

Each model's status changes to Installed once Silverdaw has checked the folder. If a folder doesn't contain the expected model, Silverdaw tells you so you can try another.

The backup model has its own Locate… button in the Backup model section, which works the same way.

Speed, quality, and cleanup

You can trade speed for polish:

• When you run Separate Stems, the dialog offers a Quality setting — Fast, Balanced, or Best — so you can pick how much time Silverdaw spends for cleaner results. Silverdaw remembers your choice for the next separation.
• GPU acceleration is optional. If you have a compatible graphics card, separation can run faster; if not, it still works on the processor. You can turn it on in Edit ▸ Preferences ▸ Stems.
• An optional cleanup can tidy up the separated parts, with Light, Medium, or Strong strength. This is set in Edit ▸ Preferences ▸ Stems.

Getting good performance

Stem separation is one of the most demanding things Silverdaw does: it runs a machine-learning model across your audio, which takes real processing power and time. On a modest laptop — or on any machine without a compatible graphics card — a single separation can take several minutes. A few habits keep it comfortable:

• Start with a faster setting. Choose Fast (or Balanced) for Quality the first time, listen to the result, and only step up to Best if the separation isn't clean enough. Each step up takes noticeably longer, so there's no need to pay for quality you don't need.
• Separate short sections, not whole songs. You rarely need every stem of an entire track. Trim or split the clip down to just the part you want — a single chorus, a drum break, an eight-bar loop — and separate that. Less audio means far less waiting, and you can always separate another section later.
• Use GPU acceleration if you can. If your computer has a compatible graphics card, turning on GPU acceleration in Edit ▸ Preferences ▸ Stems can speed separation up considerably. Without one it still works, just more slowly on the processor.
• Be patient. Especially at higher quality or on a less powerful machine, a separation can take a while. Let it run to the end rather than assuming it has stalled — it's working hard in the background.

After separating

Once you have your stems on separate tracks, they behave like any other clips. You can mute the vocals to make an instrumental, keep only the drums as a loop, add effects, or slice them into loops — all the usual editing applies.