48K high-quality Qingshuyin AI Yaoyang RVC model

Using a dataset of about 30 minutes, 500 rounds were trained.

Moreover, when the dataset is processed and output with specialization, noise and other issues are extremely rare. If you're using it for entertainment singing, be sure to lower your original tone.

You can also achieve the best effect by setting the pitch.

The reasoning effects I provide are for reference only; each output has its merits, but most are pretty similar.

Below is the loss value curve for training: