Instruments dropouts

[ghost]

I am rather impressed with this software after using it only one time! After using 4 recommended models to remove vocals, I can detect no artifacts. However, in the ranges where the vocals have been removed, the instrumental parts experience rather substantial dropouts in some areas. Are there any specific models that will address this issue?

[Max-Grom]

1. To begin with, it is worth understanding that it is impossible to achieve a result completely without artifacts. Only if you manually remake the track. Even the most advanced AI can (and will) make mistakes.
2. Personally, I really like the result from the MDX-NET-Inst-HQ5 model. The htdemucs v4 model also gives a good result. Currently, there are Roformer models in the beta version. They currently show very HQ separation.
   You can also install your own models. For example, SCNet XL IHF or MB Roformer Inst v7, and try combining different models in Ensemble.

You can also try separating the vocals from the track along with the reverb and then subtracting it from the original track using audio editing software.

[ghost]

thanks for the reply.