wav2vec2-HAT
A Wav2Vec2-Base self-supervised model of speech processing for Haitian Creole (0.2K-ALH)
wav2vec2-HAT-0.2K-ALH-base is a Wav2Vec2-Base self-supervised model of speech processing for Haitian Creole (Kreyòl). It follows the mobilising the archive approach: it was trained on digitalised archival fieldwork recordings of Haitian Creole, collected about 40 years ago on cassettes and digitised by the French National Library (BnF). This model is described in the following papers (Havard et al., 2024) and (Havard et al., 2025).
The model is available on Hugging Face: LLL-CREAM/wav2vec2-HAT-0.2K-ALH-base. Access is granted upon request.
The code used to pre-train the SSL models created during the LLL-CREAM project is available on GIN: CREAM/SSL-Haitian.
References
Havard, W. N., Govain, R., Gonçalves Teixeira, D., Lecouteux, B., & Schang, E. (2024). Technologies de la parole et données de terrain : le cas du créole haïtien. In M. Balaguer, N. Bendahman, L.-M. Ho-dac, J. Mauclair, J. G Moreno, & J. Pinquier (Eds.), Actes de la 31ème Conférence sur le Traitement Automatique des Langues Naturelles, volume 1 : articles longs et prises de position (pp. 686–694). ATALA and AFPC. https://aclanthology.org/2024.jeptalnrecital-taln.45
Havard, W. N., Govain, R., Lecouteux, B., & Schang, E. (2025). Speech Technologies with Fieldwork Recordings: the Case of Haitian Creole. Proceedings of the Eight Workshop on the Use of Computational Methods in the Study of Endangered Languages (ComputEL-8), 40–46. https://aclanthology.org/2025.computel-main.5/