A 48K RVC model Ai Susu that is both sweet and manageable

This dataset blends 2-3 similar tones to create this tone.

This timbre dataset is about 30 minutes long, processed by the author second by second, making it very clean to use.

Since most of the dataset's accents come from the stage accent, if your style is set to sweet girl, you can add a bit of speech during the original input to make it look even cuter.

For mature and mature types, it depends on your original input. If your original tone is soft, you can adjust the pitch to -1-3 to solve this problem.

Of course, the singing section is generally not suitable for the ancient-style genre, which means you need to articulate more clearly than other models when using it.

Frequently Asked Questions:

Can the model laugh and scream?

Optimization has been made in the dataset for this aspect, but since everyone uses different

methods for laughter and shouting, it is not recommended to use large-scale language for model modification.

Does the model need to adjust other parameters?

Everyone's input environment is different. We recommend adjusting the pitch and breathing protection according to the default settings in RVC-ui.

Why does the tone sound different from what you reasoned about?

Many people have encountered this problem: why does the sound of the trial sound differ from the inference sound you listen to? This is a matter of input. Each input tone has its own merits, and the timbre before and after training with the dataset will also differ. Specifically, use your best tone.