| The development of intelligent devices is becoming a popular trend in the hearing aid industry. Such devices aim at making the user's listening experience more natural and at improving customer satisfaction. One focus of interest in this dissertation is the automatic adjustment of the hearing aid control settings to minimize the need for manual user interventions. The proposed system is based on computational intelligence tools, namely artificial neural networks and neurofuzzy systems, which have the ability to learn the dynamics of highly nonlinear systems without the need for the explicit knowledge of their mathematical models. Such techniques are adopted here to map the acoustic features (input space) to the desired volume setting (output space) of the hearing aid user. Two computational intelligence tools, a multilayer perceptron and an adaptive network-based fuzzy inference system were analyzed on three simulated users with moderate, severe, and profound hearing losses. A hearing aid simulation system provided target volume settings to train and test the learning networks, selected to optimize the speech intelligibility index in each acoustic situation. The performances of both soft computing models obtained from over 2000 recordings demonstrated a high efficiency of the adopted approach in automatically optimizing volume settings for the three simulated users. In worst case scenario 95% of the testing patterns obtained 0.06 SII error or less, over 400 audio files. A future step is to extend to an online adaptation and eventually the proposed system would be integrated into a trainable self-learning hearing aid. |
