rAVEUK/genshin-voice

---
language:
- zh
- en
- ja
- ko
task_categories:
- audio-classification
- automatic-speech-recognition
- text-to-speech
pretty_name: Genshin Voice
dataset_info:
features:
- name: audio
dtype: audio
- name: transcription
dtype: string
- name: language
dtype: string
- name: speaker
dtype: string
- name: speaker_type
dtype: string
- name: type
dtype: string
- name: inGameFilename
dtype: string
splits:
- name: train
num_bytes: 246839751624.224
num_examples: 424011
download_size: 209621886182
dataset_size: 246839751624.224
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
---

# Genshin Voice

Genshin Voice is a dataset of voice lines from the popular game [Genshin Impact](https://genshin.hoyoverse.com/).

Hugging Face 🤗 [Genshin-Voice](https://huggingface.co/datasets/simon3000/genshin-voice)

<!-- STATS -->
Last update at `2025-04-22`

`424011` wavs

`40907` without speaker (10%)

`40000` without transcription (9%)

`10313` without inGameFilename (2%)
<!-- STATS_END -->

## Dataset Details

### Dataset Description

The dataset contains voice lines from the game's characters in multiple languages, including Chinese, English, Japanese, and Korean.

The voice lines are spoken by the characters in the game and cover a wide range of topics, including greetings, combat, and story dialogue.

- **Language(s) (NLP):** Chinese, English, Japanese, Korean

## Uses

To install Hugging Face's datasets library, follow the instructions from [this link](https://huggingface.co/docs/datasets/installation#audio).

### Example: Load the dataset and filter for Chinese voices of Ganyu with transcriptions

```python
from datasets import load_dataset
import soundfile as sf
import os

# Load the dataset
dataset = load_dataset('simon3000/genshin-voice', split='train', streaming=True)

# Filter the dataset for Chinese voices of Ganyu with transcriptions
chinese_ganyu = dataset.filter(lambda voice: voice['language'] == 'Chinese' and voice['speaker'] == 'Ganyu' and voice['transcription'] != '')

# Create a folder to store the audio and transcription files
ganyu_folder = 'ganyu'
os.makedirs(ganyu_folder, exist_ok=True)

# Process each voice in the filtered dataset
for i, voice in enumerate(chinese_ganyu):
audio_path = os.path.join(ganyu_folder, f'{i}_audio.wav') # Path to save the audio file
transcription_path = os.path.join(ganyu_folder, f'{i}_transcription.txt') # Path to save the transcription file

# Save the audio file
sf.write(audio_path, voice['audio']['array'], voice['audio']['sampling_rate'])

# Save the transcription file
with open(transcription_path, 'w') as transcription_file:
transcription_file.write(voice['transcription'])

print(f'{i} done') # Print the progress
```

### You unpacked the game and just want to know what the wavs are about

result.json format: (subject to change)

```json
{
"9b5502fb1b83cb97.wav": {
"inGameFilename": "VO_friendship\\VO_raidenShogun\\vo_raidenEi_dialog_pendant.wem",
"filename": "9b5502fb1b83cb97.wav",
"language": "English(US)",
"transcription": "Really? So in all this time, no new Electro Visions have appeared in the outside world? Well, what I can say on this topic is subject to certain constraints, but... it is not by my will that Visions are granted or denied. The key is people's desire, and... well, there's another side to it too.",
"speaker": "Raiden Shogun",
"talkRoleType": "",
"talkRoleID": "",
"guid": "f8e72b65-6c0a-4df1-a2f0-2bb08dbeab75",
"voiceConfigs": [
{
"gameTrigger": "Fetter",
"gameTriggerArgs": 3001,
"avatarName": "Switch_raidenShogun"
}
]
}
}
```

## Dataset Creation

### Source Data

The data was obtained by unpacking the [Genshin Impact](https://genshin.hoyoverse.com/) game.

#### Data Collection and Processing

Please refer to [Genshin-Voice](https://github.com/simon300000/genshin-voice) and [w4123/GenshinVoice](https://github.com/w4123/GenshinVoice) for more information on how the data was processed.

#### Who are the source data producers?

The source data producers are the developers of the game, miHoYo.

### Annotations

The dataset contains official annotations from the game, including language, speaker name, and transcription.

## Bias, Risks, and Limitations

Annotations are incomplete. Some voice lines are missing speaker names and transcriptions.

Speakers and transcriptions may contain markups and placeholders: `#<color=#37FFFF>パイモン：</color>{NICKNAME}、すごく怖い悪夢を見たことってあるか？\\n<color=#37FFFF>{NICKNAME}：...`

### Recommendations

Users should be made aware of the risks, biases and limitations of the dataset.

Speaker names can be partially inferred from the ingame filenames.

## Licensing Information

Copyright © COGNOSPHERE. All Rights Reserved.

## More Information

I can upload wav files on demand.

language:
- 中文
- 英语
- 日语
- 韩语
task_categories:
- 音频分类（audio-classification）
- 自动语音识别（automatic-speech-recognition）
- 文本转语音（text-to-speech）
pretty_name: Genshin Voice
dataset_info:
features:
- 名称：音频（audio），数据类型：音频格式
- 名称：转录文本（transcription），数据类型：字符串
- 名称：语言（language），数据类型：字符串
- 名称：说话者（speaker），数据类型：字符串
- 名称：说话者类型（speaker_type），数据类型：字符串
- 名称：类型（type），数据类型：字符串
- 名称：游戏内文件名（inGameFilename），数据类型：字符串
splits:
- 名称：训练集（train），字节数：246839751624.224，样本数：424011
download_size: 209621886182
dataset_size: 246839751624.224
configs:
- config_name: 默认配置（default），data_files:
- split: 训练集（train），path: data/train-*

# Genshin Voice

Genshin Voice 是一款源自热门游戏《原神》（Genshin Impact）的语音台词数据集。

Hugging Face 🤗 [Genshin-Voice](https://huggingface.co/datasets/simon3000/genshin-voice)

<!-- STATS -->
最后更新于 `2025-04-22`

`424011` 条WAV音频文件
`40907` 条无说话者信息（占比10%）
`40000` 条无转录文本（占比9%）
`10313` 条无游戏内文件名（占比2%）
<!-- STATS_END -->

## 数据集详情

### 数据集描述

本数据集收录了该游戏多名角色的多语言语音台词，涵盖中文、英语、日语及韩语。

这些语音台词均由游戏内角色配音，内容涵盖问候、战斗及剧情对话等广泛主题。

- **自然语言处理所用语言**：中文、英语、日语、韩语

## 应用场景

如需安装Hugging Face的datasets库，请参照[此链接](https://huggingface.co/docs/datasets/installation#audio)中的说明操作。

### 示例：加载数据集并筛选出带有转录文本的甘雨中文语音

python
from datasets import load_dataset
import soundfile as sf
import os

# 加载数据集
dataset = load_dataset('simon3000/genshin-voice', split='train', streaming=True)

# 筛选出带有转录文本的甘雨中文语音
chinese_ganyu = dataset.filter(lambda voice: voice['language'] == 'Chinese' and voice['speaker'] == 'Ganyu' and voice['transcription'] != '')

# 创建文件夹存储音频与转录文本文件
ganyu_folder = 'ganyu'
os.makedirs(ganyu_folder, exist_ok=True)

# 处理筛选后的数据集的每条语音
for i, voice in enumerate(chinese_ganyu):
audio_path = os.path.join(ganyu_folder, f'{i}_audio.wav') # 音频文件保存路径
transcription_path = os.path.join(ganyu_folder, f'{i}_transcription.txt') # 转录文本文件保存路径

# 保存音频文件
sf.write(audio_path, voice['audio']['array'], voice['audio']['sampling_rate'])

# 保存转录文本文件
with open(transcription_path, 'w') as transcription_file:
transcription_file.write(voice['transcription'])

print(f'{i} 已完成') # 打印处理进度

### 若你已解压游戏文件并希望了解WAV文件对应的内容

result.json 格式（可能会变更）：

json
{
"9b5502fb1b83cb97.wav": {
"inGameFilename": "VO_friendship\VO_raidenShogun\vo_raidenEi_dialog_pendant.wem",
"filename": "9b5502fb1b83cb97.wav",
"language": "English(US)",
"transcription": "Really? So in all this time, no new Electro Visions have appeared in the outside world? Well, what I can say on this topic is subject to certain constraints, but... it is not by my will that Visions are granted or denied. The key is people's desire, and... well, there's another side to it too.",
"speaker": "Raiden Shogun",
"talkRoleType": "",
"talkRoleID": "",
"guid": "f8e72b65-6c0a-4df1-a2f0-2bb08dbeab75",
"voiceConfigs": [
{
"gameTrigger": "Fetter",
"gameTriggerArgs": 3001,
"avatarName": "Switch_raidenShogun"
}
]
}
}

## 数据集构建

### 源数据

本数据集的源数据通过解压《原神》（Genshin Impact）游戏文件获取。

#### 数据收集与处理

如需了解数据处理的详细流程，请参照[Genshin-Voice](https://github.com/simon300000/genshin-voice)与[w4123/GenshinVoice](https://github.com/w4123/GenshinVoice)两个仓库的说明。

#### 源数据产出方

本数据集的源数据由游戏开发商米哈游（miHoYo）提供。

### 标注信息

本数据集包含游戏官方提供的标注信息，涵盖语言、说话者姓名及转录文本。

## 偏差、风险与局限性

标注信息存在不完整的情况，部分语音台词缺失说话者姓名与转录文本。

说话者名称与转录文本中可能包含标记与占位符：`#<color=#37FFFF>パイモン：</color>{NICKNAME}、すごく怖い悪夢を見たことってあるか？\n<color=#37FFFF>{NICKNAME}：...`

### 使用建议

使用者应知晓本数据集存在的风险、偏差与局限性。

可通过游戏内文件名部分推断说话者名称。

## 授权信息

版权所有 © COGNOSPHERE。保留所有权利。

## 补充说明

可根据需求上传WAV音频文件。