Enhancing acoustic environment classification for hearingimpaired individuals using hybrid CNN and RFE

Abstract

Individuals who are deaf or hard of hearing experience considerable difficulties in distinguishing sounds in various acoustic environments, which affects their communication ability and overall quality of life. Existing auditory assistive technologies currently face challenges with real-time classification and adaptation to changing noise conditions, underscoring the need for more reliable and accurate classification models. This research bridges the existing gap by creating a hybrid classification framework that integrates convolutional neural networks (CNN) and random forest ensemble (RFE) to enhance the accuracy of environmental sound classification. The study utilizes Mel-frequency cepstral coefficients (MFCCs) for feature extraction and principal component analysis (PCA) for dimensionality reduction, thus facilitating the efficient processing of real-world audio data. The proposed methodology improves classification accuracy across various environmental conditions. Experimental evaluations demonstrate superior performance, achieving a training accuracy of 94.93% and a testing accuracy of 93.41%, thereby exceeding conventional machine learning methods. By overcoming limitations in existing models, this research contributes to the development of adaptive hearing assistance systems with enhanced noise classification capabilities. The results have significant implications for the development of smart hearing aids, real-time noise classification, and auditory scene analysis. Ultimately, this research enhances assistive hearing technologies, promoting greater accessibility, communication, and inclusion for hearing-impaired individuals, thus contributing positively to society.