Fact · Languages and accents
In the FHNW team's 2022 evaluation, their own Swiss German model had almost half the word error rate of the best commercial system.
Checked against its source on October 17, 2026
In the source's own words
“Our best model has almost half the word error rate and a 15 points higher BLEU score than the best commercial system.”
Swiss German Speech to Text system evaluation, Yanick Schraner et al., arXiv (Computation and Language), 2022, Section 4, Evaluation.
Open the original- What it measures
- The commercial systems are anonymized. The test sets are the STT4SG-350 test set and the Swiss Parliament Corpus.
- What it does not show
- The authors built the model that wins. Read the figure as their own measurement from May 2022.