Zhu, W (2024) Enhancing Speech Recognition of Welsh for Older Adults Using Data Augmentation Techniques. Master thesis, Voice Technology (VT).
|
PDF
MA-S5551390-W-Zhu.pdf Download (446kB) | Preview |
Abstract
Automatic Speech Recognition (ASR) is widely used in various applications, enhancing clarity in educational, daily, and cross-cultural interactions. While promising for older adults, ASR systems often struggle with their speech due to physiological and cognitive changes. This study addresses this challenge by fine-tuning ASR models with older adults’ speech data and employing data augmentation techniques. Focusing on Welsh, a low-resource language, the research demonstrates that fine-tuning the XLSR model reduced word error rate (WER) from 62.19% to 57.64%. Further improvements were achieved using advanced techniques such as speed perturbation with a factor of 0.9, reducing WER to 54.30%. These results underscore the potential for enhancing ASR performanceforolder adults through tailored augmentation methods, contributing to more inclusive speech technology for low-resource languages.
| Item Type: | Thesis (Master) |
|---|---|
| Name supervisor: | Do, T.P. |
| Date Deposited: | 19 Sep 2025 13:38 |
| Last Modified: | 19 Sep 2025 13:38 |
| URI: | https://campus-fryslan.studenttheses.ub.rug.nl/id/eprint/508 |
Actions (login required)
 |
View Item |