Kouon_Vocoder

中文文档(同步更新)

The kouon_vocoder project is a vocoder project driven by the SVS community and the producers involved in singing synthesis. It aims to solve the range problem of the OpenVPI NSF-HiFiGAN vocoder and the sound quality problem of the Fish NSF-HiFiGAN vocoder.

This project is part of the Multi-langs project and runs as a separate sub-project.

This Vocoder project is intended for research and non-commercial use only.

Changelog

• 2024.07.10 nsf_hifigan v1.0: The onnx file has been released.
• 2024.10.31 nsf_hifigan v2.0: The onnx file has been released.
• 2024.11.01 mini nsf_hifigan: The onnx file has been released.
• For the latest news of the version, please join the QQ group.

Computing power specifications

• The first version of NSF uses computing power for: 4*NVIDIA A800 PCIE 80GB
• The first version of MINI-NSF uses computing power for: 8*NVIDIA H800 SXM 80GB
• The second version of NSF uses computing power for: 8*NVIDIA H100 SXM 80GB

NEW: Our's NSF-HiFiGAN is Training SUCCESS, And release ONNX model 和 OpenUTAU's OUDEP before 12:00 AM on 2024/07/10 (20240710)

Training Code

• RefineGAN: [Fish-Diffusion]
• NSF-HiFiGAN: OpenVPI's Singing Vocoders Code

Major contributors to this repository:

• CN17161
• [旋转_turning_point]
• [可米斯圣]
• [星空]
• [爱摸鱼的zzc]
• 白菜工厂1145号员工

Special thanks

• 花宴ichika
• 进击的jinji
• 夕沢yuusawa
• 灌肠
• 星海智算(算力提供)

For more information on the individual vocoders, please see our relese in the repository.