Skip to main content

Table 8 Frame-level AMT results using 60 full-length music pieces in “ENSTDkCl/MUS” and “ENSTDkAm/MUS”

From: Piano multipitch estimation using sparse coding embedded deep learning

 

Precision (%)

Recall (%)

F-measure (%)

Hawthorne [40]

88.53

70.89

78.30

Sigtia [38]*

71.99

73.32

72.22

Kelz [39]*

81.18

65.07

71.60

Melodyne [40]

71.85

50.39

58.57

MPENet

90.14

48.96

62.95

  1. Final metrics are the average over all pieces
  2. Results with asterisks are reimplemented by [40]