48K Ai Zijin Male Wenqing Wide Range Relatively Clean RVC Model

Recently, I've been busy with AI alchemy because I've been constantly experimenting with the fusion of various tones. Zijin's model theoretically supports Chinese, English, and Japanese.

During the blending process, under similar timbres, we collected datasets from different languages to fuse them, a process that took "two days."

Yes! ～

For two whole days, this included model refinement, using oeksound+ manual tone refinement. However, this model is not perfect. During reasoning, although the vocal range is wide (through the singing part and the sound after reasoning), the singing still feels somewhat unclean, which is not recommended if you don't recommend this issue.

This model is quite good. Next is the sound description after reasoning.

Because the original tone

before inference is also warm blue, the inference effect is the same. If your output is the same as the listening sample, you'll need to keep adjusting the pitch. Everyone's original input is different, so the effect will vary.

The final inference will output a 48K sampling rate. You just need to adjust the pitch or set breath protection; other settings are the default.