Detection of magnetic audio tape degradation with neural networks and Lasso

摘要

Audio magnetic tapes manufactured using polyester urethane are known to become nonplayable over time due to the degradation of the magnetic layer. Attempting to play degraded tapes to digitize them can cause extensive damage to the tape as well as to the play back device. For this reason, most of the magnetic tapes in cultural heritage institutions are in critical state. The purpose of our study is to preserve historical recordings in magnetic tapes by developing a nondestructive technique to determine degradation status. Our approach is to combine attenuated total reflectance Fourier transform infrared spectroscopy (ATR FT-IR) with chemometric techniques, especially neural networks and least absolute shrinkage and selection operator (Lasso). The model built using neural networking was able to successfully classify playable and nonplayable with 97% to 98% accuracy when similar tape brands/models were in the training and the test set. With different brands/models in the test set, neural network model performed poorly. However, Lasso showed 95.5% accuracy for similar brand/models and 80.5% accuracy for different tape brands/models. This suggests that Lasso is the better technique to determine if a tape is degraded or not.