Speech-to-Text Translation on FLEURS
15.7Error Rate (ara -> X)Seamless M4T
Evaluation results
| Method | Links | ||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Seamless M4TModel architecture category=End-to-End2026.05 | 15.7 | 13.5 | 31.8 | 15.3 | 15.2 | 11.9 | 14.2 | 12.6 | 11.5 | 15.8 | 14.1 | 15.6 | |
| Whisper + LLaMAX3Model architecture category=Cascaded2026.05 | 16.1 | 14.7 | 25.2 | 17.4 | 19.8 | 14.4 | 15.7 | 5.4 | 12.3 | 18 | 15 | 15.8 | |
| MCAT-27BModel architecture category=End-to-End, Model size=27B2026.05 | 20 | 17.7 | 31.9 | 19.6 | 24.2 | 18 | 20.9 | 13 | 15.6 | 24 | 18 | 20.3 | |
| ESRT-4BModel architecture category=End-to-End, Model size=4B2026.05 | 20.8 | 18 | 32.5 | 20.6 | 24.7 | 17.7 | 20.2 | 13.1 | 17.2 | 23.5 | 18.5 | 20.6 | |
| Whisper + NLLBModel architecture category=Cascaded2026.05 | 21.1 | 18.5 | 30.5 | 21.2 | 23.7 | 18.9 | 19.5 | 12.9 | 17.1 | 23.4 | 18.7 | 20.5 | |
| ESRT-12BModel architecture category=End-to-End, Model size=12B2026.05 | 25.8 | 21.3 | 35.4 | 24.8 | 27.6 | 21.4 | 23.7 | 17.6 | 21.1 | 27.1 | 21.9 | 24.3 |