Mitu is a girl vocal supporting the singing RVC model

When training Mi Tu, I was surprised that 20% of his tone leaned toward Xiao Tuantuan's voice.

Miaoyin collected 20 minutes of dry notes and about 15 minutes of singing to generate the MiTu model. From a reasoning perspective, this model can be rated A+ in all aspects.

Next, I will reason about the performance and the singing parts through several different timbres.

By the way, I'll answer questions I've answered one by one below for friends who have encountered problems using models these days.

Original sound

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

After reasoning

Original sound

After reasoning

From the above reasoning, you should be able to tell that different original voices produce different inference effects, but the tone tone remains similar. So when using this kind of unspecified model, if you feel the effect is not great or differs from Miaoyin Workshop's reasoning, try to use the effect of mimicking the tone. Of course, if you really can't solve the problem, you can upload the original audio to a cloud drive and send me a private message on the site. I will solve it for you during my free time.

Singing portion performance

Parameters

Training log

I will also package the npy file into the compressed file. Although I don't need the npy file now, it still helps a lot for some older versions of the RVC main program.