Table 13 Performance comparison of RP + MFCC and RP + GFCC unimodal systems over stand-alone nonlinear (RP) and linear (MFCC and GFCC) systems. Note that stand-alone systems over MFCC repeated here for comparison.

From: Recurrence plot embeddings as short segment nonlinear features for multimodal speaker identification using air, bone and throat microphones

Feature

Air

Bone

Throat

Val_Acc%

Test_Acc%

Val_Acc%

Test_Acc%

Val_Acc%

Test_Acc%

RP

96.22 ± 0.12

94.43 ± 0.28

96.43 ± 0.08

95.58 ± 0.04

96.04 ± 0.08

95.73 ± 0.15

MFCC

91.20 ± 0.04

86.97 ± 0.01

90.12 ± 0

87.48 ± 0.02

98.91 ± 0.03

96.25 ± 0.05

GFCC

81.67 ± 1.5

84.85 ± 1.2

65.693 ± 1.2

80.29 ± 1.8

82.29 ± 0.42

85.08 ± 0.48

MFCC + RP

96.22 ± 0.16

94.43 ± 0.20

96.43 ± 0.08

95.58 ± 0.14

96.04 ± 0.10

95.73 ± 0.06

GFCC + RP

86.71 ± 0.31

82.37 ± 0.53

91.08 ± 0.14

89.04 ± 0.58

85.94 ± 0.72

82.25 ± 0.48

MFCC + GFCC

92.969 ± 0.08

94.07 ± 0.12

96.10 ± 0.04

96.18 ± 0.04

97.26 ± 0.1

96.50 ± 0.2

  1. Highest accuracy values are in [bold].