Speech Question Answering on Web Questions
32.5Speech QA ScoreTurnGuide
Evaluation Results
| Method | Links | |
|---|---|---|
| TurnGuideloss ratio=2:12025.08 | 32.5 | |
| GLM-4-Voice2025.08 | 32.2 | |
| Moshi2025.08 | 26.6 | |
| Scaling InterleaveModel=9B, Train Params=9B, Data=600B tok2026.06 | 15.9 | |
| Scaling InterleaveModel=9B, Train Params=9B, Data=200B tok2026.06 | 13.3 | |
| Ours (8B)Model=8B, Train Params=~150M, Data=2.5k hrs2026.06 | 12.2 | |
| Scaling InterleaveModel=9B, Train Params=9B, Data=100B tok2026.06 | 9.3 | |
| MoshiModel=7B, Train Params=7B, Data=7M hrs2026.06 | 9.2 | |
| Ours (4B)Model=4B, Train Params=~100M, Data=2.5k hrs2026.06 | 7.9 | |
| SpeechGPT2025.08 | 6.5 | |
| Spectron2025.08 | 6.1 |