Enhancing speech technology accessibility for individuals with Parkinson’s
Zheng, Xiuwen
Loading…
Permalink
https://hdl.handle.net/2142/127266
Description
Title
Enhancing speech technology accessibility for individuals with Parkinson’s
Author(s)
Zheng, Xiuwen
Issue Date
2024-12-06
Director of Research (if dissertation) or Advisor (if thesis)
Hasegawa-Johnson, Mark Allan
Department of Study
Electrical & Computer Eng
Discipline
Electrical & Computer Engr
Degree Granting Institution
University of Illinois at Urbana-Champaign
Degree Name
M.S.
Degree Level
Thesis
Keyword(s)
Accessibility
Automatic Speech Recognition
Dysarthria
Language
eng
Abstract
Accessibility is a human right. While automatic speech recognition (ASR) has been widely used in our daily life, it struggles when recognizing dysarthric and dysphonic speech, due to acoustic impairment and lack of available training data. This thesis aims to enhance speech accessibility for people with Parkinson's, by fine-tuning pre-trained ASR systems using the 2023-10-05 data package collected by the Speech Accessibility Project (SAP), which includes speech data from 253 individuals with Parkinson's disease. The proposed method significantly outperforms a baseline model fine-tuned with typical speech, yielding improvements in word error rate of 45.15% to 38.76% compared to models fine-tuned with 100 hours and 960 hours of Librispeech data, respectively. Furthermore, this research explores cluster-dependent fine-tuning and multi-task learning methods that have shown effectiveness in the previous research in addressing speech impairments associated with Cerebral Palsy. The most promising results were obtained using a multi-task learning approach, in which the ASR model is trained to predict the speaker's impairment severity as an auxiliary task.
Use this login method if you
don't
have an
@illinois.edu
email address.
(Oops, I do have one)
IDEALS migrated to a new platform on June 23, 2022. If you created
your account prior to this date, you will have to reset your password
using the forgot-password link below.