Neural Vocoding on LibriTTS-R clean-100 (test) benchmark leaderboard · SOTA2 Research

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

4.08NISQAGround Truth

Evaluation Results

Method	Links
Ground Truth2026.05		4.08	—	—	—	4.11	—	—	—
WaveFit2026.05		4.02	—	0.0226	5.36	4.04	0.9	0.31	15.51
GAN-WaveNeXt 22026.05		4.01	—	0.0066	0.2	4.04	0.95	0.3	59.94
GAN-WaveNeXt 22026.05		4.01	—	0.009	0.24	4.04	0.95	0.3	74.93
HiFi-GAN V12026.05		3.99	—	0.011	0.8	4.05	2.34	0.16	13.9
WaveFit2026.05		3.97	—	0.0213	4.28	3.99	1.01	0.32	15.51
GAN-WaveNeXt 22026.05		3.92	—	0.0054	0.15	3.91	0.96	0.3	44.96
WaveFit2026.05		3.91	—	0.0151	3.22	3.98	1.01	0.32	15.51
Diff-WaveNeXt 22026.05		3.81	—	0.0164	0.16	3.87	4.16	0.12	57.68
WaveFit2026.05		3.8	—	0.0111	2.15	3.89	1.03	0.32	15.51
GAN-WaveNeXt 22026.05		3.77	—	0.0033	0.1	3.88	0.97	0.31	29.97
FastDiff2026.05		3.67	—	0.0282	0.8	3.78	4.32	0.24	62.52
Diff-WaveNeXt 22026.05		3.45	—	0.0335	0.16	3.55	7.34	0.16	14.42
FastDiff2026.05		3.43	—	0.0625	0.8	3.5	4.76	0.16	15.63
WaveNeXt2026.05		3.16	—	0.0022	0.06	3.2	0.92	0.31	14.98
Diff-WaveNeXt 22026.05		—	32	—	—	—	—	—	—
Fastdiff2026.05		—	96	—	—	—	—	—	—
GAN-WaveNeXt 22026.05		—	410	—	—	—	—	—	—
HiFi-GAN2026.05		—	270	—	—	—	—	—	—
WaveFit2026.05		—	410	—	—	—	—	—	—