Most speech technologies still struggle with dialects. Augusta addresses this challenge. It is an open Automatic Speech Recognition (ASR) model developed at Eurac Research to transcribe South Tyrolean German dialects into Standard German, making spoken, informal language accessible for research, media, public administration and archives.
Built on open-source ASR models and collaborative datasets, Augusta combines publicly available resources with community-contributed recordings and an iterative correction workflow. Users actively help improve the system by feeding corrections back into the training loop.
We are actively seeking use cases and collaborators across sectors to test, refine and extend Augusta in real-world scenarios. Collaborations are conceived as mutually beneficial and research-oriented, enabling partners to address their own needs while contributing to the advancement of open, dialect-aware speech technology.
Our lightning talk presents Augusta as a concrete example of how open technologies can advance speech recognition for low-resource languages, supporting digital sovereignty, linguistic diversity and community- driven approaches to machine learning and AI beyond standard-language settings.
