The AutoIPA project is a collaboration between Virginia Partridge of the UMass Center for Data Science and Artificial Intelligence and Joe Pater of UMass Linguistics. Its goal is to make automated IPA transcription more useful to linguists (and others!). Our first step was to fine-tune a Wav2Vec 2.0 model on the Buckeye corpus (see below for details). Our next steps will be to extend our work to other varieties of English and other languages. Please reach out to us if you have any questions or comments about our work or have related work to share! (pater@umass.edu, vcpartridge@umass.edu).

News

September 2025

We presented our AutoIPA model at the special session on Deep Phonology at AMP 2025 at UC Berkeley. The slides can be found here. The audio files for the examples in the slides are here. The references cited in the slides are here. We have made the various versions of our model referred to in the paper available through a web interface on Hugging Face. The web interface includes support for Praat text grid input and output. The models can also be downloaded from Hugging Face; software and detailed testing results can be found on GitHub.

If you use our software, please cite our AMP paper:

Partidge, Virginia, Joe Pater, Parth Bhangla, Ali Nirheche and Brandon Prickett. 2025/to appear. AI-assisted analysis of phonological variation in English. Special session on Deep Phonology, AMP 2025, UC Berkeley. To appear in the Proceedings of AMP 2025.

Funding

We gratefully acknowledge the support of the Center for Data Science and Artificial Intelligence, the HFA/CICS Collaborative Seed Fund, and the Public Interest Technology Initiative, all at UMass Amherst.