48K unique, noise-free AI Tiantian, a Shaoyu RVC model with a husky feel

The tone of this training session is quite unique; compared to previous mature tones, this model has a slightly hoarse sound.

The dataset for this training isn't very large, but it is relatively good for tuning the tone and supports rap songs.

The input requires the original voice to be the inference demonstration effect, or you can adjust the pitch according to actual conditions for optimal results. Specific parameters can be set as defaults. The training algorithm is rmvpe.

This model is not a full-range model and does not support optimization effects for tone such as "screaming" or "laughing." It only handles noise and treble and bass very well.

If you ignore the above "drawbacks," the remaining advantages are almost all of this model's strengths. I optimized the sound quality to a 48K sampling rate, which is a very good sampling rate.

Currently supported languages: Chinese and English. The Japanese language still needs to be tested.

⚠️：（This model can only be used for non-commercial learning and research, and please do not use it in violation of local laws and regulations.）

Supported: Google Drive download