Emotion Recognition and Spoof Detection from Whispered Speech

Speech Emotion Recognition is used to provide a more natural interaction for Human Computer Interface. Whispered speech is used when confidential matters are discussed. Usually the magnitude information of the spectrum dominates in the features of speech emotion recognition. This paper investigates the effectiveness of time, frequency and phase features for emotion recognition from whispered speech. The various features obtained from the speech sample are energy, entropy, statistical parameters, autocorrelation, zero crossing rate, modified group delay and all-pole group delay function. The resulting feature vector is used to train a Support Vector Machine. Here, a database containing whispered speech of different speakers with different emotions is created to demonstrate the potency of the proposed method.