A speech improver for hearing-impaired children using the DP matching and the SOM

This paper presents a new system, the speech improver, for hearing-impaired children. They may wish to speak as well as hearing-well people does, although it is possible to communicate with finger language or letters. This system has realized such hope by the DP (dynamic programming) matching and a neural network. The SOM (self-organizing map). The former replaces a child's voice with a reference voice, and the latter deals with many kinds of pronunciations as simple compressed data. We use the pitch of the child's voice for synthesis. Namely, this system has been used to correct wrong pronunciations of the child voice into proper pronunciations of the reference voice, with the remaining the characteristics of the child's voice.