Speech Pathology Voice Application for ALS Patients : Remote speech sample collection for ALS testing

Sammanfattning

Amyotrophic Lateral Sclerosis (ALS) is a disease that leads to the degeneration of muscles, including the bulbar muscles located in the head and neck. Degeneration of these muscles can lead to slurred or dysarthric speech. For that reason, speech tests as a method for assessment is a growing practice that is showing great potential. This thesis explores the viability of using artificial intelligence to conduct such speech tests remotely using an automated telephony system. A combination of machine learning models were used to create a modular, customizable system that could call patients and guide them through standardized ALS speech tests, collect their speech samples, and provide real-time test evaluations directly in the call. The system was assessed using a combination of quantitative data collected on healthy individuals, and a qualitative case study where experienced speech therapists and researchers tested the user-friendliness and clinical relevance of the system. To make future research possible, all code written was made publicly available as open-source. Analysis on the collected data indicated that the system was able to successfully collect high-quality speech samples, and that the automated test evaluations mapped closely to manual evaluations in most cases. The system was also regarded as highly intuitive (scores of 9 or 10 out of 10 total) by all users and the collected speech samples were deemed to be of high clinical relevance. However, there were limiting factors identified, including the challenges related to deploying resource-intensive machine learning models, and the need for verification of the system on a test group of real ALS patients. Overall, the project demonstrated that an automated, phone-based system had the potential of being a scalable and cost-effective tool for remotely collecting high-quality speech samples. Future work should focus on validating the system with ALS patients to determine its real-world applicability.