6. Classification
Support vector machine(SVM) is used to classify emotion class
SVM has utilized Linear and Radial Basis Function (rbf) kernel for generating
feature representative models based on training vectors
8. Result and Analysis
AER using MSPS, MFCC, SDC and their combinations on SAVEE database
AER from single features
For inter-class value of confusion matrix should be high
For intra-class value of confusion matrix should be less
9. Accuracy for single feature selection:-
Neutral emotion has shown better recognition while fear and surprise
have depicted low AER
12. Problem with combine features:-
Redundancy issue
Bias & variance problem
Required more computation time
13. Feature selection
Use to overcome redundancy and computation complexity
Lasso regression shrinks larger dimension to small one by setting some of
coefficient to zero
Ridge regression is also feature shrinking algorithm like Lasso but it doesn’t
make coefficients zero
Feature selection ratio(FSR) is less for ridge regression
16. Performance comparison on eNTERFACE
and with SAVEE database
The performance for different database is different
For savee database accuracy 72.5%
For eNTERFACE database accuracy is 56.41%
17. Conclusion
AER in stressed speech
Initially individually Accuracy is measured for MSPS,MFCC and SDC
For better accuracy this features are combined
This lead to increase in dimensions which required more computation time
To overcome this issue Lasso and ridge regularization is used
Accuracy for Savee database 72.5%
Accuracy for eNTERFACE database 46.41%