Skip to main content
EURASIP Journal on Audio, Speech, and Music Processing

Table 2 Example comparison of top-ranked features for different training objective functions

From: Articulation constrained learning with application to speech emotion recognition

 

ACO

ACL

AR

Rank

Top Features

1

MFB 18, MIN

MFB 18, MIN

MFB 19, MIN

2

MFB 15, MIN

MFB 19, MAX

MFB 19, MEAN

3

MFB 16, RNG

MFB 19, MEAN

MFB 19, MAX

4

MFB 26, MIN

MFB 15, MEAN

MFB 11, MEAN

5

MFB 22, MIN

MFB 21, MEAN

MFB 15, MEAN

UAR (%)

80.1

83.2*

–

CC

− 0.42

0.78

0.90

  1. Results are for valence classification for male speaker and vowel IY. Learning is constrained using jaw position along the Y-axis, with λ1=0.5 and λ2=0.1. UAR unweighted average recall, CC correlation coefficient, MFB Mel filter bank. *Statistically significant improvement (p < 0.05) over compared methods
Back to article page