Aidai Haibara / Japanese dual-pair RVC children's voice model

Miaoyin has trained many daily models, so recently we started training a batch of anime RVC models.

Among them, Ai Haibara, a representative among Japanese anime fans, is more suited to Japanese dubbing, so her daily voice acting accounts for about 70% of the dataset.

But considering the Taiwanese

version of Wei Jingqi's performance is also quite good, we allocated 30% of the data to Chinese characters. The effect is quite good.

Secondly, considering the model's orientation, many users may experience inconsistencies in tone and color during use. Therefore, I will teach you a tutorial on how to convert male and female parameters. Of course, everyone's original tone is different, so try to incorporate some during use.

No reasoning in Chinese

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

After reasoning

After reasoning here, since I forgot to set the pitch, the original tone at 0 still doesn't look very similar. Previously, when the tone was -2, it was still quite similar.

Japanese original audio

After reasoning

When using it, you can set the pitch according to your own original tone

Below is a segment of deductive singing: the original sound does not remove reverb, so there will be tearing sounds and noise.

Ai Haibara's voice is relatively steady, leaning towards girly + childish tones. When using it, the user's voice can be slightly muffled, and the pitch should not be too high.

Ai Haibara's original audio switches to girlish voice, then turns to Ai Haibara

No reasoning has been

made

After reasoning

parameters

Training log