Neural Vocoding on LJSpeech
4.49MOSDiffWave
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| DiffWaveSteps=2002022.03 | 4.49 | 7.3 | — | — | — | — | — | |
| BDDMSteps=12, ÂN=0.67, βΝ=0.122022.03 | 4.48 | 0.438 | — | — | — | — | — | |
| BDDMSteps=7, ân=0.62, β₁=0.422022.03 | 4.43 | 0.256 | — | — | — | — | — | |
| WaveGradSteps=10002022.03 | 4.36 | 38.2 | — | — | — | — | — | |
| HiFi-GAN2022.03 | 4.33 | 0.0134 | — | — | — | — | — | |
| BDDMSteps=3, ân=0.68, β₁=0.532022.03 | 3.64 | 0.11 | — | — | — | — | — | |
| WaveNet (MoL)2022.03 | 3.52 | 318.6 | — | — | — | — | — | |
| MelGAN2022.03 | 3.48 | 0.004 | — | — | — | — | — | |
| WaveGlow2022.03 | 3.03 | 0.0198 | — | — | — | — | — | |
| DiffwaveSampling steps=6, RTF=0.0932022.04 | — | — | — | 0.966 | 3.62 | 72 | 0.007 | |
| FastDiffSampling steps=4, RTF=0.0172022.04 | — | — | — | 0.976 | 3.71 | 49 | 0.006 | |
| HIFI-GANRTF=0.0022022.04 | — | — | — | 0.956 | 3.28 | 72 | 0.01 | |
| UnivNetRTF=0.0022022.04 | — | — | — | 0.971 | 3.45 | 68 | 0.013 | |
| WaveGlowRTF=0.0292022.04 | — | — | — | 0.961 | 3.16 | 66 | 0.014 | |
| WaveGradSampling steps=50, RTF=0.3902022.04 | — | — | — | 0.911 | 2.7 | 61 | 0.008 | |
| WaveNet (MOL)RTF=85.2302022.04 | — | — | — | — | — | 33 | 0.002 |