Speech Synthesis on LibriTTS (dev)
1.086M-STFTHiFi-GAN (v1)
Evaluation results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| HiFi-GAN (v1)Method=LSGAN, Training time (days)=3.6, Training steps=1 M, # Params=14.0, xRT=1102026.03 | 1.086 | 2.735 | 0.171 | 90.3 | 3.369 | 23.7 | 3.453 | |
| HiFi-GAN (v1)Method=RAF, Training time (days)=7.3, Training steps=1 M, # Params=14.0, xRT=1102026.03 | 0.991 | 3.018 | 0.146 | 92.4 | 3.492 | 20.4 | 3.539 | |
| VocosMethod=HingeGAN, Training time (days)=4.5, Training steps=1 M, # Params=13.5, xRT=7712026.03 | 0.933 | 3.317 | 0.133 | 93.8 | 3.312 | 23.8 | 3.373 | |
| VocosMethod=RAF, Training time (days)=6.1, Training steps=1 M, # Params=13.5, xRT=7712026.03 | 0.913 | 3.428 | 0.12 | 94.2 | 3.495 | 19 | 3.513 | |
| WaveFMMethod=FM, Inference mode=1 step, Training steps=1 M, # Params=19.5, xRT=80, Official Checkpoint=true2026.03 | 0.884 | 3.51 | 0.126 | 94.4 | 2.927 | 34.9 | 3.031 | |
| BigVGAN-baseMethod=LSGAN, Training time (days)=5.9, Training steps=1 M, # Params=14.0, xRT=722026.03 | 0.874 | 3.452 | 0.134 | 92.4 | 3.45 | 19.5 | 3.515 | |
| BigVGAN-baseMethod=RAF, Training time (days)=4.8, Training steps=0.5 M, # Params=14.0, xRT=722026.03 | 0.866 | 3.619 | 0.119 | 94 | 3.617 | 16.1 | 3.635 | |
| WaveFMMethod=FM, Inference mode=6 steps, Training steps=1 M, # Params=19.5, xRT=18, Official Checkpoint=true2026.03 | 0.853 | 3.909 | 0.103 | 95.3 | 3.138 | 29 | 3.212 | |
| BigVGAN-baseMethod=RAF, Training time (days)=9.4, Training steps=1 M, # Params=14.0, xRT=722026.03 | 0.841 | 3.767 | 0.111 | 94.4 | 3.651 | 14.8 | 3.667 | |
| BigVGANMethod=LSGAN, Training time (days)=11.9, Training steps=1 M, # Params=112.4, xRT=442026.03 | 0.822 | 3.824 | 0.11 | 94.4 | 3.509 | 18 | 3.556 | |
| BigVSANMethod=LSSAN, Training steps=1 M, # Params=112.4, xRT=44, Official Checkpoint=true2026.03 | 0.779 | 4.14 | 0.084 | 95.9 | 3.472 | 18.1 | 3.531 |