Comparison of Features Based on MFCCs and Eigen Values of Autocorrelation Matrix for Cross-Lingual Vocal Emotion Recognition in Five Languages of Assam

This work investigates whether vocal emotion expression of discrete emotion (i) can be recognized cross-lingually, (ii) of surprise, which is actually a cognitive component that could be present with any emotion, can also be recognized as a distinct emotion. This study will enable us to get more information regarding nature and function of emotion. Furthermore, this work will help in developing a generalized vocal emotion recognition system, which will increase the efficiency of human-machine interaction systems. In this work, an emotional speech database consisting of short sentences of five full-blown basic emotions, full-blown surprise emotion and ´no-emotion´ (i.e. neutral) is created with 140 acted utterances per speaker of five native languages of Assam. This database is validated by a listening test. A new feature set is proposed based on eigen values of autocorrelation matrix (EVAM) of each frame of the speech signal. The Gaussian mixture model (GMM) is used as the classifier. The performance of the proposed feature set is compared with mel frequency cepstral coefficients (MFCCs) at sampling frequencies of 44.1 kHz and 8.1 kHz and with additive white noise of 5 db and 0 db signal-to noise ratios (SNRs) under matched noise condition.