List your company

Neural Vocoding on LibriTTS-R clean-100 (test)

4.08NISQAGround Truth
3.12323.37163.623.8684May 25, 2026

Evaluation results

MethodLinks
Ground Truth2026.054.08———4.11———
WaveFitIterations=52026.054.02—0.02265.364.040.90.3115.51
GAN-WaveNeXt 2Iterations=42026.054.01—0.00660.24.040.950.359.94
GAN-WaveNeXt 2Iterations=52026.054.01—0.0090.244.040.950.374.93
HiFi-GAN V12026.053.99—0.0110.84.052.340.1613.9
WaveFitIterations=42026.053.97—0.02134.283.991.010.3215.51
GAN-WaveNeXt 2Iterations=32026.053.92—0.00540.153.910.960.344.96
WaveFitIterations=32026.053.91—0.01513.223.981.010.3215.51
Diff-WaveNeXt 2Sub-modeling strategy=with2026.053.81—0.01640.163.874.160.1257.68
WaveFitIterations=22026.053.8—0.01112.153.891.030.3215.51
GAN-WaveNeXt 2Iterations=22026.053.77—0.00330.13.880.970.3129.97
FastDiffSub-modeling strategy=with2026.053.67—0.02820.83.784.320.2462.52
Diff-WaveNeXt 2Sub-modeling strategy=without2026.053.45—0.03350.163.557.340.1614.42
FastDiffSub-modeling strategy=without2026.053.43—0.06250.83.54.760.1615.63
WaveNeXtIterations=12026.053.16—0.00220.063.20.920.3114.98
Diff-WaveNeXt 2Hardware=Single GPU2026.05—32——————
FastdiffHardware=Single GPU2026.05—96——————
GAN-WaveNeXt 2Hardware=Single GPU2026.05—410——————
HiFi-GANHardware=Single GPU2026.05—270——————
WaveFitHardware=Single GPU2026.05—410——————