A smartphone-based digital hearing aid to mitigate hearing loss at specific frequencies

Abstract

Hearing Loss is one of the three most common chronic conditions among the elderly. In many cases, an individuals hearing is only impaired at certain (not all) frequencies. Analog hearing aids boost all sound frequencies equally including frequencies in which the individuals hearing is good, causing discomfort to the user. Digital hearing aids can amplify only the specific frequencies at which a persons hearing is impaired. In this paper, we describe the design, implementation and evaluation of a smartphone digital hearing aid app. Our digital hearing aid implementation has two parts: speech processing in the frequency domain and sound classification. We used Weighted Over-Lap Add (WOLA) filter bank to decompose microphone sounds into different frequency bands that are then amplified in the frequency domain. Mel-frequency cepstral coefficients (MFCC) of input sounds are computed and used as features for sound classification by the Gaussian Mixture Model (GMM) machine learning model. Our digital hearing aid app amplifies select frequency bands and correctly classifies speech in quiet and noisy environments. The results of a small user evaluation of our prototype are also promising.