SOAP Note Generation on Synthetic Clinical Dialogue 1.0 (dev)
4.9FaithfulnessReference (Oracle + Kimi K2 Thinking)
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| Reference (Oracle + Kimi K2 Thinking)Architecture=Oracle-grounded Generation2026.04 | 4.9 | 4 | 4.6 | 3.9 | |
| Whisper Large V3 + Qwen3-ThinkingArchitecture=Cascaded (Whisper Large V3 + Qwen3-Thinking)2026.04 | 3.2 | 4.2 | 4 | 3.6 | |
| Qwen3-ASR+ThinkingArchitecture=Cascaded (ASR + Thinking LLM)2026.04 | 3.1 | 4 | 3.9 | 3.5 | |
| Qwen3-Omni-InstructArchitecture=End-to-end (Omni-Instruct)2026.04 | 1 | 1.3 | 4.5 | 2.4 | |
| Qwen3-Omni-ThinkingArchitecture=End-to-end (Omni-Thinking)2026.04 | 1 | 1.5 | 4.4 | 2.4 |