Chris Bartley
PhD researcher at the University of Sheffield, building speech and language technology to help preserve endangered languages, starting with Manx Gaelic.
Research interests
My doctoral research at the University of Sheffield focuses on making modern speech and language technology accessible to endangered and low-resource languages. I work primarily with Manx Gaelic (Gaelg), a Celtic language from the Isle of Man, but the methods I develop are designed to be language-agnostic and transferable.
Low-resource speech technology — Building ASR, TTS, and voice conversion systems that bootstrap from very small corpora (tens of minutes to a few hours of audio) using semi-supervised learning, data augmentation, and cross-lingual transfer from related languages.
Endangered language technology — Bringing these methods to endangered and minority languages, primarily Manx Gaelic, and building practical tools for revitalisation such as speech synthesis, recognition, and translation. The approaches are language-agnostic and transfer across the Celtic family and beyond.
Domain adaptation — Closing the gap between the clean, read speech that models are trained on and the spontaneous speech that dominates real recordings. I use voice conversion (Grad-VC / DiffVC) as offline data augmentation, training on cross-speaker, content-matched word and phone pairs rather than same-voice reconstruction. Listen to sample training pairs.
Unsupervised learning — Speech models that learn from audio without transcriptions, with a focus on graphemic approaches, including a graphemic formulation of wav2vec-U 2.0 that maps speech directly to characters rather than phonemes, removing the dependence on a phonemic lexicon.
Publications
-
Bootstrapping Endangered Language ASR with Short-Form Corpora
Christopher Bartley, Anton Ragni
Interspeech 2026Links to follow after publication -
How I Built ASR for Endangered Languages with a Spoken Dictionary
Christopher Bartley, Anton Ragni
Preprint · 2025
Further publications in preparation.