Dementia classification using attention mechanism on audio data

Dementia is one of the most common diseases among elderly people, which affects patients’ motor skills, memory, and social abilities. Its early detection can be beneficial in controlling the progression of the disease. This paper concerns the use of Alzheimer’s Dementia Recognition through Spontaneous Speech in both 2020 and 2021, as well as external audio data in DementiaBank, to classify Alzheimer’s disease. We suggest a solution using an attention mechanism on the mel-spectrogram audio representation. The perspective of using model pre-training and model fine-tuning to improve classification accuracy is also covered in the research experiments. The created models outperform the 2020 baseline by 22.91% in terms of accuracy and by 11.26% in terms of accuracy score for the 2021 dataset.