System-level Speech Translation Quality Estimation on IWSLT Metrics Shared Task 2026 (test)
95.1Soft Pairwise Accuracy (en-de)HydraQE (all heads avg.)
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| HydraQE (all heads avg.)2026.06 | 95.1 | 98.4 | 96.8 | |
| HydraQE (xCOMET head)2026.06 | 94.8 | 98.7 | 96.7 | |
| HydraQE (DA head)2026.06 | 94.7 | 98.1 | 96.4 | |
| HydraQE (primary)2026.06 | 94.5 | 97.9 | 96.2 | |
| HydraQE (MetricX head)2026.06 | 94.2 | 97.1 | 95.7 | |
| CometKiwisource=organizers' results2026.06 | 93.1 | 97.2 | 95.2 | |
| SpeechQE2026.06 | 89.8 | 96.1 | 92.9 | |
| BLASER 2.0 QE2026.06 | 88.8 | 88.2 | 88.5 | |
| COMET (partial)2026.06 | 85.7 | 78.2 | 81.9 |