A system for recognition and translation of the speech of handicapped individuals

There exists a large population whose speech is considerably impaired and effectively unintelligible. This group includes children as well as adults. The lack of successful human communication which they experience is a severe limitation and can even adversely affect their cognitive and social development. The goal of this work is to create a compact and stand alone system for online audio-translation of discrete impaired speech. Such a handy voice-input voice-output communication aid (VIVOCA) does not presently exist on the market. To this end, an appropriate method was developed for handicapped speech recognition. The system consists of two units: one module for recording and preprocessing the handicapped person's speech, based on a PC computer; and a stand alone, portable device for on-line recognition and translation based on a modified neural network. The procedure of translation takes about 0.5 sec., and the testing gives about 90% accuracy.