Updated: 2025/5/22

This update fixes background noise and optimizes some singing details, solving a large amount of background noise during covers.

Third Version Listening:

<!--[if lt IE 9]>document.createElement('audio');<![endif]-->

——————————————————————— ——————————————

Earlier, I open-sourced a doro RVC model, but its tone leaned toward a girlish and unrefined tone, failing to evoke that cute feeling.

Doro RVC model

The above are free and open source. If you know how to change the key or add a magic sound effect, you can still create a cute Doro-like tone, though it can be a bit complicated.

This time, I'm bringing you the cute version of Doro's RVC model. The dataset for this model isn't easy to find, and its tone is quite similar to 'Mambo.' Here, we've processed the cute version datasets you've heard and some that have been post-processed.

Currently, the output audio may have a slight noise. Although previous datasets have been heavily denoised, there are still strange issues that can't be processed properly.

Secondary creation for non-commercial insurance is a very good choice. After output, you may still need to add some effects to professional processing software to remove some noise to achieve the optimal coverage of the output source.

Processing the dataset took us almost an entire day to work through, and we hope to continue supporting this motivation.

As you can see in the alchemy screenshot, at 6,000 steps, the model starts to fit. Although the 6,000-step pth file will be packaged, the reasoning effect will also be the same as at 6,000 steps.

The preview after reasoning will use dry preview; other inferences will be done later.

This model requires foundational post-production.