Back to news overview

OpenVokaturi 1.0 launched

14 April 2016

Given speech that comes in as a sampled sound, the Vokaturi software computes five acoustic features: pitAve, pitSlo, intAve, intSlo, spcSlo. From these features, the software computes probabilities for the five main emotions: neutrality, happiness, sadness, anger, fear.

The method is linear discriminant analysis, with all off-diagonal cells of all five covariance matrices set to zero. Trained and tested on EmoDB, with ten-fold cross-validation, classification is correct in 62.7 percent of the cases (with homogeneous priors, i.e. each of the five emotions is equally likely a priori).