Fig. 15
From: Piano multipitch estimation using sparse coding embedded deep learning
AMT results of the first 30 s of “MAPS_MUS-deb_clai_ENSTDkCl” produced by plain MPENet, where legends can be referred to Fig. 14. A typical case of false negatives is where notes have long duration (still, MPENet can detect the attack of each note, but decays are discontinuous since the note probabilities are polarized caused by non-linear classifiers, c.f Fig. 7)