Development of speech recognition system in native language for hearing impaired

Abstract

This paper presents the performance of the speech recognition system with reference to children with normal hearing and children with hearing impairment. Though the nasal and oral cavities of the hearing impaired are perfect, they cannot produce sounds since they cannot hear anything. The reason is that the ability to understand language and speech production is coordinated by the brain. So a person with a problem in the ear or damage in brain activities due to an accident, stroke or birth defect may have problems in producing speech. They are classified as profoundly deaf and hard of hearing, based on the degree of hearing ability. Early detection of deafness would enable the hearing impaired to produce sounds by speech therapy. If deafness is detected at a later stage, it is difficult to make the speech of the hearing impaired understandable. So, it is necessary to develop the system for recognizing their speeches, especially in the native language. In this paper, a system is developed for Tamil language by using, Melfrequency cepstral coefficient feature extraction at the front end and Hidden Markov Model tool kit at the back end. System is evaluated and the comparison is done between the speeches of normal speakers and the hearing impaired. Recognition accuracy is 92.4% for hearing impaired speeches and 98.4% for normal speeches. Though it is difficult for the unfamiliar listeners to understand the hearing impaired speeches, this system can be utilized for recognizing the speeches of Hearing impaired by others.