A fairly straightforward approach for music source separation is to train
independent models, wherein each model is dedicated for estimating only a
specific source. Training a single model to estimate multiple sources generally
does not perform as well as the independent dedicated mode