Hallucination detection on CALLHOME (test)
11.56Hallucination Rate (%)RAUQ Entropy
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| RAUQ EntropyModel=Voxtral-3B, Category=Baseline2026.04 | 11.56 | 86 | 51 | 60 | 44 | 52 | 0.53 | |
| Mean EntropyModel=Voxtral-3B, Category=Baseline2026.04 | 15.83 | 86 | 55 | 55 | 55 | 59 | 0.57 | |
| PerplexityModel=Voxtral-3B, Category=Baseline2026.04 | 25.18 | 79 | 50 | 40 | 64 | 56 | 0.54 | |
| Logistic RegressionModel=Voxtral-3B, Feature Configuration=Top 75, Number of features=3002026.04 | 27.26 | 80 | 55 | 43 | 74 | 61 | 0.57 | |
| Logistic RegressionModel=Voxtral-3B, Feature Configuration=AUDIORATIO, Number of features=9602026.04 | 29.13 | 79 | 52 | 40 | 74 | 58 | 0.55 | |
| Logistic RegressionModel=Voxtral-3B, Feature Configuration=Combined, Number of features=38402026.04 | 32.8 | 76 | 50 | 37 | 77 | 55 | 0.52 | |
| RandomModel=Voxtral-3B, Category=Baseline2026.04 | 50.03 | 49 | 22 | 15 | 46 | — | — | |
| Attention ScoreModel=Voxtral-3B, Category=Baseline2026.04 | 53.92 | 46 | 23 | 15 | 51 | 15 | -0.01 |