Audio Spectrogram Transformer-based Audio Classification using Voice data of Dementia Patients
Younjeong Lee, Chanho Park, Yeeun Moon, Jongpil Jeong · 2023
Globally, aging is ongoing, and the probability of senile disease and dementia patients continue to increase as human lifespan and the proportion of the elderly increase. Dementia is a disease that affects the depression and quality of life of the elderly, and it is a socially and economically dangerous disease that cannot live independently. Currently, the only treatment for dementia is through training if the disease is delayed or detected early. Diagnosing dementia early is the most important step at this point, and studies are currently being conducted non-face-to-face or through voice. This study aims to predict dementia early through human voice. The study was conducted with the aim of optimizing and lightening the characteristics of dementia patients by applying the Audio Spectrogram Transformer analysis model as a voice classification model. The model used in this paper classifies dementia with an accuracy of 89%. In future research, we will try a study that combines Explainable AI (XAI), an interpretable artificial intelligence of deep learning.