X-to-English Speech-to-text Translation on CoVoST v2 (test)
35.7Fr -> En Scorew2v2-mBART
Evaluation Results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| w2v2-mBARTSetup=Bilingual, Learning Strategy=Supervised learning + pre-training2022.10 | 35.7 | 36.2 | 39.4 | 5.7 | 13.5 | 26.1 | |
| w2v2-mBART (Babu et al., 2021)Setup=Multilingual (21 langs), Learning Strategy=Supervised learning + pre-training2022.10 | 32.9 | 34.1 | 26.4 | 3.5 | 6 | 20.6 | |
| S2T TransformerSetup=Multilingual (21 langs), Learning Strategy=Supervised learning2022.10 | 26.9 | 26.3 | 9.6 | 0.4 | 0.6 | 12.8 | |
| S2T TransformerSetup=Bilingual, Learning Strategy=Supervised learning2022.10 | 26.3 | 23 | 14.8 | 0.1 | 2.5 | 13.3 | |
| Cascaded (ASR-TDN-MT)Setup=Bilingual, Learning Strategy=Unsupervised learning2022.10 | 24.4 | 23.4 | 27.8 | 8.5 | 7.6 | 18.3 | |
| w2v2-mBARTSetup=Multilingual ({Fr,Es,Ru,Et,Lv}), Learning Strategy=Unsupervised learning2022.10 | 24.3 | 24 | 22.8 | 3.1 | 1 | 15 | |
| w2v2-mBARTSetup=Bilingual, Learning Strategy=Unsupervised learning2022.10 | 24.2 | 24 | 25.6 | 3.9 | 2.8 | 16.1 |