Integrated speech training system for hearing impaired

Abstract

Hearing impaired people have serious difficulties learning to produce intelligible speech. To facilitate speech training for hearing impaired people, many researchers have developed various speech training aids, such as F/sub 0/ (fundamental frequency) indicator, N (nasality) indicator, S (fricative /s/) indicator, and spectra displayer. However, most of these training aids treat only a single speech parameter. An integrated speech training system was implemented. The system makes it possible to display vocal tract shape, as well as other speech parameters, in real time. A self-training program for hearing impaired persons has been developed to help them correct errors. To estimate speech parameters, it is assumed that the speech production process is an autoregressive model. Using a linear predictive analysis, vocal tract lateral shapes and log spectra are estimated. Log energy of the speech signal is used as intensity. Fundamental frequency and nasality are also detected using a vibration sensor. To check the reliability of the developed system, parameters of normal people were tested and training tests for hearing impaired children were performed. The results showed that the authors' proposed system is very useful for speech training of the hearing impaired. >