Streaming Voice-Agent Interaction Efficiency on BigBench Audio
80.73NFEPredGen
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| PredGenThinking Mode=No-think2026.01 | 80.73 | 79.38 | 98.33 | |
| PredGenThinking Mode=Think2026.01 | 80.73 | 79.73 | 98.76 | |
| LTS-VADThinking Mode=Think2026.01 | 6.29 | 5.26 | 83.62 | |
| LTS-VADThinking Mode=No-think2026.01 | 6.28 | 4.68 | 74.52 | |
| LTS-VoiceAgent2026.01 | 2.05 | 0.2 | 9.76 |