《EVA》Asuka (Sohryu/Asuka Langley) RVC model

Making the dataset for Asuka was truly exhausting. The old version of the EVA audio wasn't very good, and the new theatrical version took over two hours to edit all kinds of ambient sounds.

After various denoising and reverb removal processes, I gradually obtained a dataset of about 13 minutes. The dataset didn't include Chinese (originally, I planned to create Chinese data from the entire hodgepodge and throw it into training), but after thinking about it, it was a bit troublesome. The effect is only in Japanese; I tried it while reasoning, but it didn't have a strong leftist vibe. You can also try out Chinese.

As for singing, I reasoned about Japanese songs and the results are quite good, almost reaching 80% tone reproduction. After all, it's an RVC, so the tone similarity has always been average.

Let's see how this model looks. If it looks good, I'll continue making models for Ayanami Rei and Shinji. I hope everyone will support the models.

Below is my alchemy log

still the same, comment to get this model for free.

CV: Yuko Miyamura

Download