Automatic emotional speech recognition in Serbian language

The paper addresses automatic emotional speech recognition (ESR) with an emphasis on finding the most efficient set of features of emotional speech. It should be noted that research was conducted on the corpus of emotional speech recorded in Serbian, which is the first of its kind in Serbian language. Our research was focused on the comparison of discrimination capability of prosodic features, spectral features, as well as their combination in order to assess the significance of specified features for ESR and form an efficient feature vector for ESR. The motivation for this study stemmed from the growing need to put human-computer dialogue to a higher level of understanding of the broader communication context including nonverbal messages and human emotions.