Adaptive Data Boosting Technique for Robust Personalized Speech Emotion in Emotionally-Imbalanced Small-Sample Environments

. 2018 Nov 2;18(11):3744. doi: 10.3390/s18113744

Algorithm 2 Absent Emotion Data Reinforcement

Input:

I D S (1 \dots M)

—Initial Multiple User Speeches Dataset

CT—Number of Classes from Target User

T F e a t u r e V e c t o r (1 \dots N)

—Target User Speeches Feature Vector

T E m o L a b e l S e t (1 \dots C)

—Acquired Target User Speeches Label Set

NU—Number of Users

TID—Target User ID

Output: SU (1...K)—Selected Similar User Speeches Dataset

for i = 1 to M do

I F e a t u r e V e c t o r_{i}

= extractFeatures(

I D S_{i})

;

{I E m o L a b e l}_{i}

= getLabel(

I D S_{i})

;

I U s e r I D_{i}

= getUserID (

I D S_{i})

;

end

for i = 1 to N do

T C e n t r o i d V a l u e s_{i}

= calculateDistributionFactors (

T F e a t u r e V e c t o r_{i})

;

end

for i = 1 to NU do

for j = 1 to CT do

for k = 1 to M do

T E m o L a b e l S e t_{j}

I E m o L a b e l_{k}

and i =

I U s e r I D_{k}

THEN

I C e n t r o i d V a l u e s_{i}

= calculateDistributionFactors (

I F e a t u r e V e c t o r_{j});

end

for i = 1 to NU do

for j = 1 to CT do

S u m D i s t a n c e s_{i j}

S u m D i s t a n c e s_{i j}

+ EuclidianDistance(

T C e n t r o i d V a l u e s_{j}, I C e n t r o i d V a l u e s_{i j});

end

US = Sorting (

S u m D i s t a n c e s, I D S)

;

Return US;