AutoSpeech - Automatic Speech Analysis of Verbal Fluency for Older Adults

Description:

Verbal Fluency tests are quick and flexible tests used in the area of Neuropsychology to evaluate executive and language functions of subjects. These tests consist in the production of as many words as possible within a 1 minute time constraint, with respect to a given category.

There are two main types of verbal fluency tests administered: phonemic (where words that begin with a given letter are produced, e.g., the letter P), and semantic (where words that belong to the same semantic group are produced, e.g., animals). Verbal fluency performance is shown to be related to cognitive impairment and can be used to distinguish different types of brain lesions, mental disorders, and neurocognitive disorders (such as dementia).

Verbal fluency tasks have two main factors influencing the overall performance obtained. Subjects must be able to identify a possible subcategory that can be explored for the production of exemplars, while inhibiting incorrect words, and searching for new possible subgroups to use afterward – this is known as cluster switching, which is more related to executive functions. Besides this, a subject has to effectively produce from memory words belonging to a given subcategory – this determines cluster size, a task mainly associated with language and memory functions.

The increase of incidence of neurocognitive disorders and the lack of available specialists to perform screening examinations is a motivation for the automation of these tasks.

An automatic pipeline for the analysis of both phonemic (letter P) and semantic (animal category) verbal fluency tests is here proposed, for European Portuguese. This pipeline uses Google Cloud Speech-to-Text online service to perform Automatic Speech Recognition, and Natural Language Processing techniques to automatically evaluate temporal, phonemic, and semantic clusters produced.

This pipeline is tested and used to determine the impact of physical activity on cognitive function and verbal fluency performance on elderly subjects (N=57). Subjects with a lower level of physical activity tend to have lower cognitive and verbal fluency performance than more active subjects. However, due to other factors that affect this result, such as age and general cognitive status of subjects, no definitive conclusions can be taken regarding the influence of physical activity on elderly cognition.

 

Author: João Costa

Type: MSc thesis

Partner: Faculdade de Engenharia da Universidade do Porto

Year: 2019