| dc.description.abstract | This thesis addresses a critical component of Alzheimer's disease (AD) diagnosis by utilizing transcripts of condensed speech data from Dementia Bank. The work uses pre-trained big language models and creates efficient binary classification methods emphasizing early identification. Sturdy feature vectors are carefully created by utilizing feature extraction methods on various condensed voice sample datasets, guaranteeing accurate preprocessing. Many assessment criteria, including AUC-ROC, accuracy, precision, F1 score, and recall, are used to evaluate the models' ability to differentiate between 'ProbableAD' and 'Control' situations.
The findings demonstrate encouraging developments in the quick and precise detection of AD using brief speech samples. The study raises the possibility of using small speech samples for applications involving remote monitoring. The study also emphasizes how important particular auditory traits are in identifying cognitive impairments. To sum up, this thesis makes a substantial contribution to AD diagnosis by introducing a strong paradigm for using short speech data. The knowledge gained can potentially improve the lives of those affected by AD and aid in the process of early diagnosis. | |