Development of a new method for synthesizing natural voices quickly - A neural source-filter model that introduces neural networks into classical methods -

A research team consisting of Project Researcher Xin Wang, Project Assistant Professor Shinji Takaki, and Associate Professor Junichi Yamagishi of the Content Science Research Division at the National Institute of Informatics (NII, Director: Masaru Kiryugawa, Chiyoda-ku, Tokyo), an Inter-University Research Institute Corporation, Information and Systems Research Organization, has developed a neural source filter model (NSF method), a method for synthesizing high-quality voice at high speed. This new method is an extension of a voice generation model published in 1960 through deep learning, and is not only capable of generating high-quality voice waveforms close to the human voice, but also allows for stable training of neural networks.