Abstract:
This paper describes a new system, a speech improver, for hearing-impaired children. This system is composed of DP matching and SOM. The child's voice is changed to a normal-hearing person's voice by DP matching. However, the child's voice feature is not changed by retaining its pitch information. In addition, the SOM compresses much input data for reference into a small pattern. The sound source (amplitude and pitch) and articulation of the child's voice are managed separately. The amplitude comes from the LPF, and the pitch is obtained accurately by the authors' method. The LPC power spectrum envelope which represents the articulation is obtained by applying the LPC and the DFT to the input child's voice. Besides, the many spectra of the reference voice have been clustered by the SOM. These spectrum sequences of the input and the reference are the input of the DP matching. According to this output, the child's voice spectrum is replaced by a reference spectrum. A synthesized voice is obtained by concatenating the replaced spectrum transformed into a time domain considering the pitch and the amplitude. The synthesized voice has been improved with the remaining characteristics of the child's voice. Furthermore, this system corresponds to many childrens' input voices, because it uses compressed patterns.