Evaluation of Hidden Semi-Markov Models Training Methods for Greek Emotional Text-to-Speech Synthesis
Alexandros Lazaridis, Iosif Mporas · International Journal of Information Technology and Computer Science · 2013
This paper describes and evaluates four different HSMM (h idden semi-Markov model) training methods for HMM-based synthesis of emotional speech.The first method, called emot ion-dependent modelling, uses individual models trained for each emotion separately.In the second method, emotion adaptation modelling, at first a model is tra ined using neutral speech, and thereafter adaptation is performed to each emotion of the database.The third method, emotionindependent approach, is based on an average emotion model which is in itially trained using data fro m all the emotions of the speech database.Consequently, an adaptive model is build for each emot ion.In the fourth method, emot ion adaptive train ing, the average emotion model is trained with simultaneously normalizat ion of the output and state duration distributions.To evaluate these training methods, a Modern Greek speech database which consists of four categories of speech, anger, fear, joy and sadness, was used.Finally, an emotion recognition rate subjective test was performed in order to measure and compare the ability of each of the four approaches in synthesizing emotional speech.The evaluation results showed that the emotion adaptive training achieved the h ighest emotion recognition rates among four evaluated methods, throughout all four emotions of the database.