---
title: README
emoji: 🌍
colorFrom: indigo
colorTo: gray
sdk: static
pinned: false
---

## External Mirrors

ArkEcho model series (China mirror): https://pan.baidu.com/s/1usxdcbG75RSWSWyTTaerzg?pwd=etns

Legacy address / Issue reporting: https://github.com/Ethernos-Studio/Models

QQ: 3110197220

## Additional Notes

**[Model Naming Convention]**

$$ Series-ModelType-CharacterAbbr-TrainingLanguage-Size $$

**[Series Names]**
| Series | Description |
| ----- | ---- |
| `ArkEcho` | *Arknights* |


**[Model Types]**
| Type | Description |
| ----- | ---- |
| R2 | RVC model |
| GSV | GPT-SoVITS model |

**[Character Abbreviations]**
 - No fixed rule. If the character's English name is too long, Chinese pinyin initials are used instead (e.g., Rosmontis is `MDX`)

**[Training Languages]**
| Language | Notes |
| ----- | ---- |
| `Chinese` | |
| `Japanese` | |
| `English` | |
| `U_CJ` | Special, short for Universal_Chinese-Japanese |
| `U_JE` | Special, short for Universal_Japanese-English |
| `U_CJE` | Special, short for Universal_Chinese-Japanese-English |

> **Note: Since Universal models are trained with different voice actors per language, unexpected issues may occur. Unless necessary, please use the dedicated single-language models.**

**[Sizes]**
| Size | Data Amount |
| ----- | ------- |
| `Lite` | Small (approx. 1/3 of data) |
| `Medium` | Medium (approx. 1/2 of data) |
| `Full` | Complete (1/1 data) |
| `Heavy` | Large (multiple times the data) |

> **Note: Size does not equal training epochs or model quality**

**[Example]**

$$
ArkEcho-R2-Amiya-Chinese-Heavy
$$

means:
$$Arknights model - RVCv2 model - Character: Amiya - Chinese training set - Multiplied training data$$



# Notice

ArkEcho series model documentation (continuously updated): https://ethernos.net/blog.html?id=arkecho_model_shuoming

Downloading any ArkEcho series model means you have **fully read, thoroughly understood, and unconditionally accepted** all terms of the [ArkEcho Project Unified Statement](http://models.ethernos.net/third-pirty).