2.1. Dataset

The dataset for the 2020 ADReSS challenge consists of speech recordings elicited for the Cookie Theft picture description task from the Boston Diagnostic Aphasia Exam (Goodglass et al., 2001). These data were balanced by the organizers in terms of age, gender, and the distribution of labels between the training and test partitions in order to minimize the risk of bias in the prediction tasks. The dataset from 78 non-AD subjects, and 78 AD subjects, was labeled for binary classification and regression tasks. The labels for the binary classification include Alzheimer's dementia and healthy control, whereas the labels for the regression task are Mini-Mental State Examination (MMSE) scores (Folstein et al., 1975), which provide a means for dementia diagnosis based on linguistic tests. For more details regarding the dataset, including the segmentation and voice activity detection algorithm, we refer the reader to the ADReSS challenge baseline paper (Luz et al., 2020).