Automated Classification of Irregularities in Magnetic Audio Tapes Using Various Machine Learning Techniques

This thesis presents an automated system for classifying irregularities in magnetic audio tape recordings, incorporating a variety of advanced machine learning models to enhance the accuracy and efficiency of the classification process. The study employs multiple architectures, including VGG-16, ResNet, YOLO, and EfficientNet, to analyze the characteristics of irregularities on magnetic tapes. This work evaluates the performance of different models in classifying these irregularities. A diverse dataset, comprising screenshots and processed frames representing various types of irregularities, serves as the input for these models. The comparative analysis of model outcomes provides insights into the effectiveness of each model in the context of audio tape irregularity classification. This research advances the field of digital archiving and preservation by highlighting the potential of using multiple machine learning approaches for the automated classification of tape irregularities, aligning with the Moving Picture, Audio, and Data Coding by Artificial Intelligence (MPAI) Context-based Audio Enhancement (CAE) standard.