List your company

Neural Vocoding on LibriTTS

4.058UTMOSGround truth
2.335242.7824953.229753.677005Jun 1, 2023Nov 17, 2023May 4, 2024Oct 20, 2024Apr 7, 2025Sep 23, 2025Mar 12, 2026

Evaluation results

MethodLinks
Ground truth2023.064.058—————
BigVGANtraining iterations=2 million2023.063.7494.653.6930.95570.108—
VocosBackbone architecture=ConvNeXt2023.063.7344.663.70.95820.101—
Vocos (w/ Snake)Activation function=Snake2023.063.6994.663.6290.95790.102—
ComVo2026.033.6901—3.823996.090.09030.8439
HiFi-GANtraining iterations=2 million2023.063.6694.573.0930.94570.129—
Vocos (w/o ConvNeXt)Backbone architecture=ResBlocks with dilated convolutions2023.063.6584.653.5280.95340.109—
Vocos (w/ absolute phase)Phase prediction method=absolute phase prediction2023.063.594.653.5650.95560.108—
iSTFTNettraining iterations=2 million2023.063.5644.562.9420.93720.141—
APNet22026.032.7379—2.774892.430.14481.1582
FreeV2026.032.6971—2.79691.050.15811.1782
APNet2026.032.4015—2.845791.850.15821.3375