Automatic Speech Recognition on STT4SG-350 (test)
12.1WERTimmel/FHNW ’24
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Timmel/FHNW ’24Model=large-v2 Full FT, Data=∼1,033h, Notes=Same-conv. eval§2026.05 | 12.1 | — | |
| TimmelModel Available=No, Data Available=No, Honest Eval=No‡2026.05 | 12.11 | — | |
| Timmel large-v2Method=Full FT, Training Data=∼1,033h (502h conv.+PL+CV), Evaluation Protocol=Contaminated evaluation (ASGDTS or same-distribution data in training)2026.05 | 12.11 | — | |
| ZHAWModel Available=License req., Data Available=Proprietary, Honest Eval=No2026.05 | 17.1 | — | |
| MichaudModel Available=Restrictive lic., Data Available=No, Honest Eval=No2026.05 | 17.5 | — | |
| flix-swissgerman-loraModel Available=Yes (HF), Data Available=Yes†, Honest Eval=Yes2026.05 | 25.32 | 13.9 | |
| flix-swissgerman-fullModel Available=Yes (HF), Data Available=Yes†, Honest Eval=Yes2026.05 | 25.6 | 13.8 | |
| Flurin17Model Available=Yes (HF), Data Available=Yes, Honest Eval=Yes2026.05 | 28 | — |