DOI: https://doie.org/10.0802/Jbse.2024984437
Sunilkumar M. Hattaraki , Shankarayya G. Kambalimath
Detection, Classification, Environmental Sound, Machine Learning Model (ML), Recurrent Neural Network (RNN), FIR filter and Hearing-Impaired Individuals.
Machine learning has numerous applications in audio-signal classification. It helps to find and sort different kinds of sounds, such as talking, music, and noise, from the world around us. Before applying machine learning to classify audio signals, the audio is first converted into a format that the computer can understand. Sound is shown using methods such as pictures of sound waves, special numbers called Mel-frequency Cepstral coefficients (MFCC), a method of predicting sound patterns called linear predictive coding, and breaking down sound into tiny parts using wavelet decomposition. Once the audio has been formatted appropriately, it can be used as input for a machine learning (ML) model intended for classification. This paper introduces an approach utilizing a Recurrent Neural Network (RNN) model integrated with Finite Impulse Response (FIR) filtering. This combination aims to effectively detect and classify various listening conditions, particularly tailored to individuals with hearing impairment. The pre-trained RNN model accurately categorizes audio, while dynamic FIR filtering enhances audio quality based on the predicted environment, tailored to address the needs of hearing-impaired individuals. This model enables the detection and classification of diverse listening conditions with a training accuracy of 98.50% and a testing accuracy of 94.97%, offering personalized filtering to enhance auditory experiences for hearing-impaired individuals.