Doraemon female child voice model

This time, we're bringing a model of Doraemon. For this Doraemon-based model, Miao Yin collected a 30-minute original Japanese dataset + a 20-minute raw dataset, and after optimizing and streamlining, finally obtained a noise-free dataset of about 30 minutes.

Since the training is on the older Doraemon dataset, there will be some minor noise—here's the setting recommendation

• Protect voiceless consonants and breathed consonants: 0.2-0.4.
• Feature retrieval rate: 0.6-0.8.

Why use bilingualism as the training goal?

Dual language does not affect the final output quality. In your images, Doraemon's sound and color are very similar whether dubbed in Taiwan or Japan, so they don't really affect the quality of the model. Moreover, this type of training results in higher model quality.

Here are the results after reasoning

Before reasoning

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

After reasoning

parameters

Training log