Perceptual Evaluation of a Music Source Separation CNN Trained With Binaural and Ambisonic Audio