Speaker Recognition Systems in the Emotional Environment

It is well known that speaker recognition systems perform extremely well in the neutral environment. However, such systems perform poorly in the emotional environment. Our work in this research focuses on text- dependent speaker identification systems in the emotional environment. Our emotional environment consists of five emotions. These emotions are angry, sad, happy, disgust, and fear. Each of the hidden Markov models (HMMs) and the cepstral mean subtraction technique based on HMMs has been used separately in both the training and testing sessions of such systems. Our results show that speaker identification systems in the emotional environment based on the cepstral mean subtraction technique yield better identification performance than that based on HMMs.