Tips for separating vocals
- Separate with becruily Mel vocal model and its instrumental model variant, then get vocals from the vocal model, and instrumental from instrumental model, import both stems for the DAW of your choice (can be Audacity) so you’ll get a file sounding like original file, then export it - so perform a mixdown of both stems, then separate it with vocal model (mrmason347 /Havoc) might work with some other, newer models with similar characteristics.
- “In my testing, I've found that SCNet very high fullness (on MVSEP) put through Mel-Roformer denoise (average) and UVR denoise (minimum) has the best acapella result” dynamic
- Depending on a model, some Roformer model for vocal separation might be muddy.
Then you might consider using ensembles, although it might be noisy.
- Apollo vocal enhancer model by Lew v2
The enhancer might work the best on BS+Mel ensembles (max spec, though avg might work better in some cases).
v2 model (model/config | Colab | Inference), v1 (model | config).
The v2 is also available in the latest UVR Roformer beta in Tools>Apollo (along with the Lew’s universal model for e.g. music and OG Apollo model).
The ckpt for v1 goes to models\Apollo_Models and the yaml to model_configs subfolder.
- Sometimes using EQ stressing vocals properly might be beneficial for separation too
- You might potentially also try to experiment with demudder added with the beta patch #14 linked above. Normally demudder works only for instrumentals, but when you switch in the config editor to vocal stem being instrumental and in reverse, then demudder will work vocals. If your model have “other” stem instead of “instrumental” or “vocal”, you’ll need to rename it. Demudder requires stem labelled as instrumental to work with.
- Pitching down or negative values set in UCR for pitch shift feature might smooth out results of some models like Bigbeta5e.
- More tricks at the top here
- More tricks for general separation here