Li, Chenyu (2024) Exploring the Potential of Accent Conversion Techniques to Enhance Fairness in Language Assessment. Master thesis, Voice Technology (VT).
|
PDF
thesisChenyuLiS2615401-1.pdf Download (2MB) | Preview |
Abstract
In the context of global mobility, ensuring equitable language proficiency assessments is crucial for fair immigration and integration policies. This thesis investigates the feasibility of using machine learning to neutralize Indian accents in English speech to enhance objectivity in language proficiency evaluations. The primary aim is to determine whether machine learning can effectively neutralize accents in English spoken by speakers whose native language is Hindi, thereby addressing identity anonymity in language test settings. Existing foreign accent conversion (FAC) models are predominantly speaker-dependent, trained on datasets from specific speakers, and only effective for those individuals. The models that claim to work for unseen speakers typically involve a complicated pipeline structure, high data requirements, and are not easy to implement. This research aims to develop a speaker-independent model by training on a diverse dataset of Indian-accented English speakers. By doing so, it seeks to create a generalized accent conversion model with a simple structure that can be applied broadly, setting a precedent for extending this approach to other accents and thereby broadening the inclusivity of linguistic applications. Applications of FAC include computer-aided language learning and entertainment, such as movie dubbing. However, the impact of FAC on language assessment has seldom been discussed. This study addresses a significant gap in language proficiency assessments, where the influence of accents on evaluation outcomes remains a challenge. Despite the recognition of accent-related issues by major testing organizations, explicit measures to mitigate their impact on scoring are lacking. Through innovative approaches to neutralize foreign accents in spoken language evaluations, this research aims to ensure fair and unbiased assessments for individuals from diverse linguistic backgrounds. By identifying and addressing these challenges, the study contributes to the advancement of equitable evaluation practices in multicultural societies.
| Item Type: | Thesis (Master) |
|---|---|
| Name supervisor: | Coler, M.L. |
| Date Deposited: | 15 Aug 2024 11:27 |
| Last Modified: | 15 Aug 2024 11:27 |
| URI: | https://campus-fryslan.studenttheses.ub.rug.nl/id/eprint/550 |
Actions (login required)
 |
View Item |