Task 3: Swiss German Speech to Standard German Text


The goal of this task is to build a system able to translate Swiss German speech in various dialects to Standard German text. We provide a labeled dataset with 293 hours of recordings (mostly Bern dialect) and an unlabeled dataset with 1208 hours (mostly Zurich dialect). The team with the best BLEU score on a 13 hours test set with speakers from all German-speaking parts of Switzerland wins the contest. The dialect distribution of the test set is close to the real Swiss German dialect distribution in Switzerland.

We encourage participants to explore and combine suitable supervised, semi-supervised and unsupervised learning approaches.

Registration, more details, and the data for the task can be found here.

Organizers