sophiefy / VITS

ACG Text-to-Speech

Stella TTS Based on VITS

Contents

• Update
• Introduction
• MoeGoe
• Models
  • Shiki Natsume
  • Chipanese Bilingual Model
  • Café Stella and the Reaper's Butterflies
  • Yosuga No Sora
  • Bishojo Mangekyo
  • Genshin
  • A Certain Scientific Railgun
• Contact
• References

Update

• japanese_triphone_cleaners is available.

Introduction

Just have some fun...

The datasets and models are only for research use, NOT commercial!

MoeGoe

You can run my models on MoeGoe developed by CjangCjengh! Please first read the instructions of MoeGoe and then download my models and corresponding configuration files.

Models

---

Shiki Natsume

• Description

To further utilize the contextual information of the text, I wrote a physiscal triphone cleaners for Japanese. For more details about triphone, refer to this material.

Raw text:

さあ、はりきって行きましょう。

Phonemes:

s a a,h a r i k i Q t e i k i m a ʃ o o.

Triphones (physical):

s+a s-a+a a-a,h+a h-a+r a-r+i r-i+k i-k+i k-i+Q i-Q+t Q-t+e t-e+i e-i+k i-k+i k-i+m i-m+a m-a+ʃ a-ʃ+o ʃ-o+o o-o.

The following 2 speakers are supported.

Name	ID
四季ナツメ	0
司波深雪	1

• Model:

• Configuration file:

• Demo:

---

Chipanese Bilingual Model

• Description

I modified the Japanese cleaner and conbined it with a Chinese cleaner into a bilingual one, called "Chipanese cleaner"! The training data are from Genshin, Café Stella and the Reaper's Butterflies and Kami-sama no You na Kimi e. For more information, please refer to this repo: VITS-Bilingual.

MoeGoe is NOT supported currently!

The following 2 speakers are supported.

Name	ID
四季ナツメ	0
派蒙	1

• Model

• Configuration file

• Demo

  • Colab: VITS (中日双语)
  • BILIBILI: 基于VITS的中日双语语音合成模型

---

Café Stella and the Reaper's Butterflies

• Description

Name	ID
四季ナツメ	0
明月栞那	1
墨染希	2
火打谷愛衣	3
汐山涼音	4

• Model:

  • Google Drive (380 epochs)
  • 百度网盘 (198 epochs) 提取码0721

• Configuration file:

  • Google Drive
  • 百度网盘 提取码0721

• Demo

  • Colab: VITS (星光咖啡馆)
  • BILIBLI: 基于VITS的星光咖啡馆5人语音合成模型

---

Yosuga No Sora

• Description

Name	ID
春日野穹	0
天女目瑛	1
依媛奈緒	2
渚一葉	3
春日野悠 (reserved)	4

• Model:

  • Google Drive (600 epochs)

• Configuration file:

  • Google Drive

• Demo

  • Colab: VITS (缘之空)
  • AcFun: 基于VITS的缘之空4人语音合成模型 (完整视频)
  • BILIBILI: 基于VITS的in solitude, where we are least alone语音合成模型 (阉割视频)

---

Bishojo Mangekyo

• Description

Name	ID
蓮華	0
篝ノ霧枝	1
沢渡雫	2
亜璃子	3
灯露椎	4
覡夕莉	5

• Model:

  • Google Drive (715 epochs)

• Configuration file:

  • Google Drive

• Demo

  • Colab: VITS (美少女万华镜)
  • AcFUN: 基于VITS的美少女万华镜语音合成模型

---

Genshin

• Description

Both single speaker model of Paimon and multi-speaker model of 46 characters will be supported!

• Model:

  • Google Drive: Paimon
  • Google Drive: 46 characters

• Configuration file:

  • Google Drive: Paimon
  • Google Drive: 46 characters

• Demo

  • Colab: Paimon
  • BILIBILI: 派蒙Vtuber出道计划——基于AI深度学习VITS和VSeeFace的派蒙语音合成/套皮
  • Colab: 46 characters
  • BILIBILI: on the road-map

---

A Certain Scientific Railgun (collecting data...)

• Characters

Name	ID	Name	ID
上条当麻	0	削板軍覇	7
一方通行	1	御坂妹	8
垣根帝督	2	最終信号	9
御坂美琴	3	白井黒子	10
麦野沈利	4	佐天涙子	11
食蜂操祈	5	飾利初春	12
正体不明	6	インデックス	13

• Model:

• Configuration file:

• Demo

Contact

QQ: 2235306122

BILIBILI: Francis-Komizu

References

Original code

Reference code

Triphone