Mouth Hu is good, articulation is good, AI Zhizi 48k mature voice

Please pay attention to the Cn_hubert version, strictly follow the https://klrvc.com/zh/tutorial/3714 tutorial to deploy the model.

data-end="220" This model is based on the Cn_HuBERT self-supervised voice model and the mygf_f048k base model Trained together.
Compared to the traditional ContentVec / HuBERT scheme, this model > the enunciation accuracy > , especially in Chinese-speaking scenarios, where more natural pronunciation structures are maintained.

model will show a a slight husky texture in timbre performance.
It should be noted:

• This husky feel is part of the model's own vocal style

• The issue is not caused by background noise or training noise

• are normal phenomena during inference and do not affect usage quality

Secondly, regarding the model's inference speed, thanks to the new self-supervised voice and base model, we tested it on the 3060ti and 4070ti, and found it was almost identical to the original plan.

If the underlying feature file is not replaced as required or the old version of the feature model is still used, it may cause:

• unstable sound

• serious nonsense

• data-end="596" nonsense / mispronunciation

• reasoning quality significantly declines

The testing environment for this model is based on RVC V1006. Users are advised to maintain a consistent version to ensure compatibility and optimal performance.

theoretically supports multiple languages.