Automation of Hearing Impairment Test Based On Automatic Speech Recognition

Abstract

Hearing impairment is found in 5% of the total world’s population, out of which 10% are children and remaining 90% adults. If we focus on the age group of 65 years and above, one third of the group is facing hearing impairment issues. Different surveys carried out in USA revealed that Speech Recognition Threshold (SRT) Test is being conducted by 99.5% and 83% of audiologists for hearing assessment of the patients. However, not only the non-availability of expert audiologist but also the low literacy rate is a hurdle to conduct a successful Speech Recognition Threshold Test in Pakistan. A per the surveys Khyber Pakhtoon Kha (KPK) region of Pakistan has a literacy rate of 50%. Such less literacy rate along with Pushto language barrier between patient and audiologists makes hearing impairment diagnosis a troublesome process. This paper proposes an AHIT system based on SRT in Pushto language, which will be capable of detecting hearing impairment using Pushto as a test language, helping in automating the process of hearing impairment testing. The technique involves the extraction of Mel-Frequency Cepstral Coefficients (MFCCs) from a large data set of 15 different Pushto language spondee words. A Convolutional Neural Network (CNN) based machine learning model is then trained and tested using these MFCCs, whereas an interpreter code is used to test the system.