Impaired speech recognition

• Main Author: Lam, Michelle Su-Ann
• Other Authors: Goh Wooi Boon
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Computer and Information Science
• Online Access: https://hdl.handle.net/10356/181137
• Institution: Nanyang Technological University

Dysarthria is a speech disorder which often leads to speech that is difficult to understand or easily misinterpreted, resulting in communication challenges for affected individuals. With recent advancements in automatic speech recognition technologies, it has the potential to assist dysarthric speakers in their communication needs. This paper presents the development of an automatic speech recognition system that can effectively assist such individuals in their daily communication with their counterparts. The ability of two state-of-the-art models to accurately transcribe dysarthric speech will be explored. A large language model is incorporated as an automatic speech recognition correction system to further enhance the accuracy of resulting transcriptions. The effectiveness of the resulting automatic speech recognition system as a communication assistance tool will be demonstrated together with a text-to-speech synthesizer, with both components being integrated into a mobile application that aims to recreate clearly spoken words from the original dysarthric speech. While the resulting automatic speech recognition system faced challenges in generalizing across different dysarthric datasets, including a large language model into the process yielded positive outcomes as demonstrated by the qualitative user testing.