Emotional speech conversion based on spectrum-prosody dual transformation

A dual transformation system to transform neutral speech to emotional speech is proposed in this paper. Since spectral and prosodic features are key factors that influence the emotional effects of speech, Gaussian Mixture Model (GMM) method and the prosody rules algorithm are applied to transform the spectral and prosodic features respectively. In this paper, we transform neutral speech to angry, happy and sad speech. The training corpus is taken from Danish speech database, and the corpus used to transform is taken from the Berlin Database of Emotional Speech. It is shown in the listening test that the speech synthesized by the proposed method is perceived to portray the targeted speech emotion well. It also shows that emotions can be independent of human language from the same language family.