Cute Shaoluoyin AI Xiaorong RVC model

Originally, the model was intended to have a 48K sampling rate, and of course, a model with a 48K sampling rate was developed. Later, it was found that the model had some background noise during output, so the dataset was re-optimized and the model was retrained with a 44K (40K) sampling rate.

Essentially, 48K has much higher sound quality than 40K, but the 48K sampling rate is extremely sensitive to datasets. To ensure output quality, we ultimately decided to release the 40K Xiaorong model.

This model can be laughable to some extent, but also laughable. If it's unstable, you can also use the FCPE algorithm, which performs decently in inference tests. But you also need to support a sound card.

The above is the loss curve of this model. After 5k steps of training, the model did not overfit, and the training results were valid.

This model achieves the same effect as 48K Ai Yalin. A mature RVC model with a wide vocal range, but one has a 40K sampling rate and the other a 48K sampling rate.

If interested, you can try downloading and using it.

The original input and output should also differ, and the appropriate pitch should be adjusted to achieve the best effect.