Emotion-inspired age and gender recognition systems

In this work, emotion-inspired age and gender recognition systems are developed. In the beginning, speakers´ utterances with emotions of angry, happy, calm and sad are analyzed to identify their ages and genders where the recognition engine adopts a Support Vector Machine (SVM). According to the experimental results, the accuracies of the age and gender recognitions under a low arousal emotion tend to be worse and better than those under a high arousal emotion, respectively. In practical applications, a specific emotion may not appear in a speaker´s utterance. Hence, according to the emotional arousal intensity, speech frames of a speaker´s utterance are classified into two groups which are above and below the mean of arousal intensities of speech frames. After that, the age and gender recognitions are conducted at speech frames with higher and lower arousal intensities. Our experiments reveal that the proposed emotion-inspired age and gender recognition systems can be better that those without considering arousal intensities by 8.5% and 9.5% improvements, respectively. Therefore, the recognition systems proposed herein can effectively increase the age and gender recognition accuracy for various multimedia applications.