HiFTNet Release

HiFTNet is much faster than BigVGAN

BigVGAN Release

Mix Encoder : hubert soft & whisper large v2 audio encoder's output of 24L

dependency

about mix

final model architecture of hifigan

code: so-vits-svc-5.0-hifigan-code.zip
pretrain: sovits5.0_main_1500.pth

6G memory GPU can be used to trained

final model architecture of bigvgan

pth is pretrained model file, can be used for fine-tune
模型包含生成器和判别器，可用于微调

sovits5.0 48k debug

Debug model for sovits5.0 48k. Train 2 days, just for test.

sovits5.0 16k debug

For code debug, not last state.

sovits5.0 preview release

预览模型，未进行第二阶段训练，存在音色泄漏；附上训练日志；

v3.0 32k

disable fp16