Speech Audio Generation from Dynamic MRI on Open-source vocal tract dataset
0.818Corr2DKE-CVAE
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| KE-CVAE#Param=121.1 M2025.03 | 0.818 | 1.251 | 4.13 | |
| KE-CVAE w/o Flow#Param=120.6 M, Flow=false2025.03 | 0.798 | 1.237 | 4.02 | |
| KE-CVAE w/o adversarial training#Param=60.0 M, adversarial training=false2025.03 | 0.759 | 1.19 | 3.8 | |
| Vanilla (Transformer) w/ KE#Param=146.5 M, Knowledge Enhancement (KE)=true2025.03 | 0.688 | 1.14 | 3.57 | |
| Vanilla (CNN) w/ KE#Param=124.0 M, Knowledge Enhancement (KE)=true2025.03 | 0.672 | 1.135 | 3.48 | |
| Vanilla (Transformer)#Param=146.5 M, Knowledge Enhancement (KE)=false2025.03 | 0.649 | 1.131 | 3.22 | |
| Vanilla (CNN)#Param=124.0 M, Knowledge Enhancement (KE)=false2025.03 | 0.641 | 1.116 | 3.16 |