Transcribing deaf and hard of hearing speech using Hidden markov model

This paper presents the performance of the deaf speech recognition using Hidden Markov model. Even persons those having perfect nasal and oral cavity cannot produce sounds if they are deaf, since they cannot hear anything. If deafness is found earlier, then using speech therapist they can be made to reproduce sounds at the maximum. Depending on the degree of hearing they are deaf, profoundly deaf and hard of hearing. At the same time if deafness is found by later stage, they cannot produce sounds properly and no one can understand their speech. For analysis we have considered the deaf students whose speeches are intelligible to some extent. The characteristics of deaf speech such as pitch, formants are different from that of normal speech and due to high variability in deaf speech signal characteristics, performance of the recognition system may also degrade compared to normal speech. The uttered deaf speech is recognized using HMM with MFCC features and recognition accuracy is compared for each isolated digit. If it is extended for continuous speech then communication between deaf and normal is satisfied. Deaf speech is taken from 10 children in the age group of 5-10 years from Maharishi vidya mandir centre for deaf.