Speaker-attributed Automatic Speech Recognition on AISHELL4 Long-form (test)
15.32DERGemini-2.5-Pro
Evaluation results
| Method | Links | ||
|---|---|---|---|
| Gemini-2.5-ProEvaluation protocol=without manual segmentation2026.04 | 15.32 | 31.59 | |
| Speaker-Reasoner Multi-turn w/ SACEvaluation protocol=without manual segmentation2026.04 | 21.6 | 36.2 |