Studies on Speech Synthesis under Limited Data Conditions

This monograph book is based on the PhD dissertation of the author who got PhD degree from JAIST in 2013. This book presents methods of speech modification and transformation proposed to improve the naturalness of synthesized speech under limited data conditions based on the concept ¿appropriate smoothness¿ in synthesized speech. These methods showed their efficiencies on improving the usability under limited data conditions for both concatenative and HMM-based speech synthesis. These results contribute to the research fields of speech processing by introducing the new concept of ¿appropriate smoothness¿ in speech. There results also contribute to the research and development fields of speech synthesis in order to make speech synthesis more convenient and more efficient for human-machine interaction systems. These results also contribute to the research fields of speech modification and transformation, in which the efficiency of temporal decomposition to obtain an ¿appropriate smoothness¿ in synthesized speech were verified and confirmed.