Automatic Speech Recognition on WoW Out-of-domain (test)
12.7WERMulti-turn ASR (Raw Context)
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| Multi-turn ASR (Raw Context)Configuration=Multi-turn, Raw, Ntrain=R[1,10], Ndecode=102026.03 | 12.7 | 23.3 | |
| Multi-turn ASR (Raw Context)Configuration=Multi-turn, Raw, Ntrain=R[1,10], Ndecode=52026.03 | 12.9 | 24.2 | |
| Abstract CompressionConfiguration=Multi-turn, Compressed, Ntrain=0→10, Ndecode=102026.03 | 13.2 | 24.5 | |
| PHI-4-MULTIMODAL (Fine-tuned)Configuration=Single-turn, Ntrain=0, Ndecode=02026.03 | 13.4 | 25.6 | |
| PHI-4-MULTIMODAL (Fine-tuned)Configuration=Single-turn, Ntrain=0, Ndecode=102026.03 | 14.5 | 27.2 | |
| PHI-4-MULTIMODAL (No Fine-tuning)Configuration=Single-turn†, Ntrain=0, Ndecode=02026.03 | 20.4 | 30 |