Publication Date: 2024/01/11
Abstract: "This research introduces a groundbreaking project, "Enabling Independence through Sound Classification," leveraging Artificial Neural Networks (ANNs), Convolutional Neural Networks (CNNs), Mel- frequency cepstral coefficients (MFCCs), and the Librosa library to offer real-time auditory feedback to individuals who are hearing impaired. The project's core objective is to enhance the independence and safety of this community by translating environmental sounds into meaningful alerts and descriptions. Beyond the technical aspects of sound classification, the study emphasizes the profound social impact of promoting inclusivity, self-reliance, and equity for those with auditory challenges. Through a comprehensive exploration of CNN and RNN architectures, along with comparisons to TensorFlow and PyTorch models on a prototype dataset, the proposed approach, incorporating envelope functions, normalization, segmentation, regularization, and dropout layers, demonstrates superior accuracy and reduced loss percentages. This research signifies a pivotal step towards a more accessible and inclusive society, harmonizing technology and empathy for the benefit of individuals with sensory challenges."
Keywords: ANN, MFCC, DL, ML, RNN, CNN, API, ReLu, CPU, GPU.
DOI: https://doi.org/10.5281/zenodo.10490435
PDF: https://ijirst.demo4.arinfotech.co/assets/upload/files/IJISRT24JAN136.pdf
REFERENCES