Emotion Recognition Using Speaker Cues

The focus of the current research is on emotion identification of a speaker using his/her cues. In this work, emotion identification depends on a two-stage framework. The first stage recognizes the speaker whose emotion is undetermined, while the second stage recognizes the unidentified emotion which was spoken by the speaker whose identity was recognized in the preceding stage. Our proposed architecture has been assessed on an Arabic Emirati-accented speech corpus expressed by fifteen Emirati speakers for every gender. As a classifier, Hidden Markov Model is exploited in this study. Our results show that the proposed two-stage framework is superior to the one-stage framework and the “state-of-the-art classifiers such as Gaussian Mixture Model, Support Vector Machine, and Vector Quantization”.