FHNW Institute for Data Science Datasets

Swiss German Speech to Standard German Text

Swiss Parliaments Corpus

Dataset based on Swiss German parliament debates and their Standard German transcripts.
Parliament: Grosser Rat Kanton Bern
License: MIT

Version 1, 2020-01-28

Initial version, used in GermEval 2020 Task 4, 70 hours of training data

Version 2, 2020-08-25

Improved and extended version, used in SwissText 2021 Task 3, 293 hours of training data

All Swiss German Dialects Test Set

Test set with speech from people all over German-speaking Switzerland, with a dialect distribution close to the real dialect distribution.
Text data is from the German Common Voice project.
License: MIT

Version 1, 2021-03-10

Initial version, used in SwissText 2021 Task 3, 13 hours of data

Gemeinderat Zürich Audio Corpus

Unlabeled audio dataset containing recorded parliament debates.
Parliament: Gemeinderat Zürich
License: MIT

Version 1, 2021-03-10

Initial version, used in SwissText 2021 Task 3, 1208 hours of data

Swiss German STT Metric Evaluation

GER-HSR-1K

Human annotated ratings of a Swiss German speech-to-text system.
License: MIT

Version 1, 2023-05-22

Initial version