A basic study on talking aid and voice clarification for the hearing disorder persons

The voice of the handicapped who have lost their hearing is unclear. We propose a voice clarification system which consists of the voice recognition and voice clarification by monitoring the voice signal via a microphone and the shape pattern of mouth via a CCD camera, and by generating the clarified voice through the speaker. In order to recognize their voices, we adopt two methods: 1) a linear prediction coding cepstrum (LPC cepstrum), and 2) a neural network. The LPC cepstrum obtains the characteristic vector which is calculated by the frequency ingredient, and the neural network performs the pattern recognition for the typical wave shape corresponding to the each vowel. Furthermore, the image recognition system recognizes the shape pattern of mouth, based on the color information, estimation rules and membership functions. We confirm the efficiency of this total system to recognize the vowels.