An automatic speech recognition system for helping visually impaired children to learn Braille

Abstract

We present an automatic speech recognition (ASR) system which, along with a haptic interface, is aimed at helping preschool children to learn Braille. The ASR algorithm extracts a set of Mel-Frequency Cepstral Coefficients (MFCC) from the speech signal, followed by a Dynamic Time Warping (DTW) approach, thus allowing to recognize vowels pronounced by the user. The ASR algorithm was tested on 9 subjects and its sensitivity was measured in terms of the percentage of true positives. The highest accuracy values were obtained for the a, e, o and u vowels (with hit ratios of 88.8% in all cases), whereas the i vowel exhibited the lowest sensitivity (77.7%). Validation of user interaction with the haptic system is currently underway, and additional testing is needed to determine the potential benefits that this system offers in the context of preschool education in Colombia.