[herogary]

Source separation is commonly done by applying masks to the spectrogram. Deep learning is used to train the mask masks for different instruments' parameters. As you mentioned, this is the approach we will follow in the subsequent steps.