Emotional speech synthesis using GMM-based voice conversion technique

摘要

Because it is well known that prosodic features play an important role to express emotions, many researches have been conducted to analyze prosodic features and to construct emotional speech synthesis rules. On the other hand, the preceding papers also mention that voice quality control is necessary to improve for synthesizing emotional speech. In this report we propose an emotional speech synthesis method using GMM-based voice conversion technique. First we designed and recorded emotional (angry, sad, happy) and neutral speech databases. Then we synthesize emotional speech from neutral speech using the databases and evaluate the converted speech by objective and subjective experiments. The results show that the proposed method improve emotional quality of speech that have properly generated emotional prosody. In this paper, we report our emotional speech database, the voice conversion method and the experimental results.