ISCA Archive Interspeech 2023
ISCA Archive Interspeech 2023

Combining language corpora in a Japanese electromagnetic articulography database for acoustic-to-articulatory inversion

Tianfang Yan, Kikuo Maekawa, Yukiko Nota, Masayuki Hirata

This paper presents an electromagnetic articulography database of Japanese sentences. The database includes aligned acoustics and articulatory data from seven males and three females, with a total of five recorded hours. The database is now in preparation for public release to further research in areas of acoustic-to-articulatory inversion, brain-machine interface communication systems, artificial speech synthesis, and dialect recognition. Moreover, based on this database we established an acoustic-to-articulatory inversion system using a deep, bidirectional, long short-term memory recurrent neural network structure. The results showed that, for the Japanese language, adding English corpora to the training was not beneficial for this speaker-independent model.