Development of Japanese infant speech database from longitudinal recordings

Abstract

Developmental research on speech production requires both a cross-sectional and a longitudinal speech database. Previous longitudinal speech databases are limited in terms of recording period or number of utterances. An infant speech database was developed from 5 years of recordings containing a large number of daily life utterances of five Japanese infants and their parents. The resulting database contains 269,467 utterances with various types of information including a transcription, an F0 value, and a phoneme label. This database can be used in future research on the development of speech production.