University of Victoria (Canada)
Applying automatic speech recognition to Indigenous language documentation: A case study with Hul’q’umi’num’
Abstract
dc:description.abstractThe process of documenting Indigenous languages can create a large amount of audio recordings that are difficult to convert into a written form. Speeding up the transcription process using automatic speech recognition could help the Hul’q’umi’num’ Language & Culture Society to create pedagogical materials and make their recordings more accessible. In this project, I trained a language model known as XLS-R on Hul’q’umi’num’ audio recordings to determine how accurately it can transcribe Hul’q’umi’num’, whether particular linguistic and orthographic features are more difficult for XLS-R to transcribe, and what amount of time and computational resources the training takes. The model reached a CER of 11.1% and WER of 50% using 26 minutes of continuous speech. Most phonemes could be transcribed with high accuracy but the model showed difficulties with segmenting words, differentiating glottalized consonants from plain consonants, determining vowel length, and predicting the placement of glottal stops.
Degree
thesis:*- Level thesis:degree_level
- Master of Arts MA
- Department thesis:degree_department
- School of Languages, Linguistics and Cultures
- Year dc:date.issued
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Jiang, Xin He
- Advisors dc:contributor.supervisor
-
- Bird, Sonya
- Urbanczyk, Suzanne Claire
Subjects
dc:subject × 6Rights
dc:rights- Statement dc:rights
-
- Available to the World Wide Web
- Language dc:language.iso
- en, English
Identifiers
dc:identifier.*- Handle dc:identifier.uri
- https://hdl.handle.net/1828/23969