Performance Evaluation of the Hearing Impaired Speech Recognition in Noisy Environment
C. Jeyalakshmi, Vijaykumar Krishnamurthi, Adhvikaa Ambikapathi Revathi · International Review on Computers and Software (IRECOS) · 2013
Most of the noise suppression algorithms concentrated on normal hearing listeners. This study investigates the use of Recursive least square filter in improving automatic speech recognition of the hearing impaired speech in noisy environment. The children who are deaf or hearing impaired suffers in many ways regarding education and in public places to communicate with the normal speakers since their voices are difficult to understand due to guttural and monotone nature even though they undergone speech therapy. Because some of the sounds cannot be taught, only we can feel. To develop the recognition system for their speeches in practical situation, we have analyzed the recognition accuracy of the hearing impaired speech in various noise environments where their speech becomes still worse due to background noises. Here we have considered babble, white and factory noise in different SNR levels such as 10db, 5db, 0db,-5db and the recognition results were obtained as 27.8% for babble noise, 10.6% for white noise and 13.6% for factory noise when SNR is 0db. Spectral subtraction is initially applied and due to its poor performance it is shown that, application of Adaptive recursive least square filter improves significantly the recognition performance with filter order 128, even for the low SNR level 0db, such as 70.0% for babble noise, 67.4% for white noise and 61% for factory noise. The result suggests that regardless of impaired speech, in the practical noisy environment also we can achieve comparable recognition accuracy for hearing impaired speeches.