AutoVAT: An Automated Visual Acuity Test Using Spoken Digit Recognition with Mel Frequency Cepstral Coefficients and Convolutional Neural Network
Derryl Taufik, Novita Hanafiah
Abstract
Since we live in the digital era, refractive error cases such as myopia have been increased steadily from year to year. As a result, the number of people looking for a visual acuity test has also increased. However, conventional visual acuity test requires knowledgeable examiner, thus some people may find it difficult to undergo a visual acuity test. It would be much easier if visual acuity test can be done at home anytime. Our research would like to develop an automated visual acuity test (AutoVAT) that can run on general computer using microphone as an input device and monitor. Visual acuity is measured using Snellen chart with digit optotype and evaluated based on the user’s answer in a form of speech (spoken digit). This study uses Mel Frequency Cepstral Coefficients (MFCC) as the feature point to characterize the spoken digit and Convolutional Neural Networks (CNN) for classification. The results of this study indicate that AutoVAT successfully evaluated visual acuity score with less than one row difference on average based on Snellen chart.