Automatic Detection and Filtering of Listening Conditions In Hearing Aids Using Convolutional Neural Networks

Currently, individuals with hearing impairments struggle to effectively filter and enhance sounds across diverse listening environments, particularly in noisy conditions. This makes speech comprehension and communication in daily life challenging. This study introduces a novel method for detecting varied listening environments in hearing aid technology by integrating a Convolutional Neural Network (CNN) model with Spectral Subtraction algorithms. Achieving a training accuracy of 99.97% and a testing accuracy of 97.57%, the CNN model is designed to capture complex features from audio spectrograms, which are vital for distinguishing between different acoustic conditions. Spectral Subtraction is employed as a preprocessing step to reduce noise interference, thereby enhancing the input quality for the CNN. The approach was tested across four noise levels (0 dB, 5 dB, 10 dB, and 15 dB), with clean audio serving as a baseline. Evaluation metrics including Perceptual Evaluation of Speech Quality (PESQ), Short-Time Objective Intelligibility (STOI), and Log-Likelihood Ratio (LLR) were applied. Results demonstrated notable improvements in both speech intelligibility and overall audio quality, highlighting the effectiveness of this method in advancing hearing aid technology.