Sashiba Shoroon, a singing RVC model

This time, we trained a new RVC model named Zaozhi. It uses two Shao Luo datasets with similar timbres, and is trained on the V1006.2 version, which handles bass much better.

This time, Ai+ was again used to manually complete denoising and processing of the dataset, with the finished dataset taking about 40 minutes.

Below is a comparison of the effects before and after the reasoning. The reasoning results are for reference only; the actual sound after reasoning is based on the original timbre inference.

Before reasoning

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

After reasoning

Singing and reasoning

parameters

Alchemy log