Automatic Speech Recognition on GCND (test)
27.1Brabants WER (%)GLoRIA
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| GLoRIAFine-Tuning=Geo-Conditioned, Rank=1282026.03 | 27.1 | 40.84 | 30.16 | 30.28 | 26.7 | 49.41 | 37.61 | 37.17 | 32.09 | 34.59 | |
| Modified FeedforwardFine-Tuning=Geo-Conditioned Full Fine-Tuning2026.03 | 27.44 | 42.67 | 31.56 | 31.98 | 27.36 | 50.27 | 38.78 | 38.52 | 32.6 | 35.69 | |
| Joint ModelFine-Tuning=Full Fine-Tuning, Metadata=No Metadata2026.03 | 28.67 | 46.01 | 32.83 | 32.65 | 28.61 | 48.38 | 39.28 | 37.6 | 34.01 | 36.45 | |
| Coordinate EmbeddingFine-Tuning=Geo-Conditioned Full Fine-Tuning2026.03 | 28.7 | 44.84 | 33.04 | 33.21 | 29.17 | 52.25 | 41.08 | 41.55 | 35.1 | 37.66 | |
| Dialect-Specific (Best)Fine-Tuning=Full Fine-Tuning, Metadata=No Metadata2026.03 | 30.02 | 45.13 | 32.42 | 35.89 | 29.53 | 47.86 | 38.74 | 41.96 | 36.52 | 37.56 | |
| Pre-trained ModelModel Category=General-Purpose Model2026.03 | 56.52 | 75.38 | 66.22 | 64.51 | 58.95 | 61.74 | 58.32 | 58.69 | 66.37 | 62.97 | |
| Whisper Large-V3Model Category=General-Purpose Model2026.03 | 71.69 | 75.11 | 68.3 | 68.04 | 62.91 | 68.56 | 62.07 | 61.37 | 71.6 | 67.74 | |
| OWSM V4-1BModel Category=General-Purpose Model2026.03 | 78.32 | 83.65 | 80.54 | 75.34 | 76.25 | 76.78 | 76.01 | 75.43 | 83.61 | 78.44 |