Datasets:

ultemica
/

genshin-impact-voices

pretty_name: Genshin Impact Voice Dataset
license: other
task_categories:
  - automatic-speech-recognition
  - text-to-speech
language:
  - zh
  - ja
  - en
  - ko
tags:
  - voice
  - speech
  - genshin-impact
  - hoyoverse
  - character-voice
  - audio
size_categories:
  - 100K<n<1M
configs:
  - config_name: zh
    data_files:
      - split: train
        path: data/zh/train-*.parquet
  - config_name: ja
    data_files:
      - split: train
        path: data/ja/train-*.parquet
  - config_name: en
    data_files:
      - split: train
        path: data/en/train-*.parquet
  - config_name: ko
    data_files:
      - split: train
        path: data/ko/train-*.parquet

Genshin Impact — Voice Lines (Multi-Language)

An archive of character voice data extracted from Genshin Impact (原神), repackaged as Parquet shards per audio language.

Dataset Summary

Field	Value
Game	Genshin Impact (原神)
Publisher	HoYoverse / miHoYo Co., Ltd.
Languages	中文 (zh), 日本語 (ja), English (en), 한국어 (ko)
Game version	6.3
Source format	WAV + sidecar transcripts (`.lab` / `.txt`)
Distribution format	Apache Parquet (zstd), ~500 MiB audio per shard
Extraction tool	AI-Hobbyist / Genshin_Datasets (ModelScope mirror)

Directory Layout (HF repo)

data/
├── zh/train-NNNNN-of-NNNNN.parquet
├── ja/train-NNNNN-of-NNNNN.parquet
├── en/train-NNNNN-of-NNNNN.parquet
└── ko/train-NNNNN-of-NNNNN.parquet

Each language is a separate config. HF auto-detects the layout — load by passing the language code as the second argument to load_dataset.

Loading

from datasets import load_dataset

ds = load_dataset("ultemica/genshin-impact-voices", "ja", split="train", streaming=True)
sample = next(iter(ds))
print(sample["transcription"], sample["speaker"])

Schema

Field	Type	Notes
`audio`	`{ bytes, path }`	WAV bytes + original filename
`transcription`	string \| null	Sidecar `.lab` / `.txt` text if available
`speaker`	string	Top-level folder in the source archive (character / NPC)
`emotion`	string \| null	Reserved; currently null
`language`	string	Matches the config name
`source_path`	string	Original WAV path inside the 7z archive

Intended Use

Training / evaluation data for ASR (automatic speech recognition) models
Reference audio for TTS (text-to-speech) systems
Speaker diarization / speaker recognition tasks
Research on multilingual anime-style voice acting

Note: This dataset is intended for research and non-commercial use only.

Data Source & Extraction

The original assets were obtained from the official Genshin Impact client distributed by HoYoverse / miHoYo, then decrypted and packaged by the AI-Hobbyist community on ModelScope. This repository repackages those archives into language-keyed Parquet shards suitable for streaming via the Hugging Face datasets library.

GitHub: https://github.com/AI-Hobbyist/Genshin_Datasets
ModelScope: https://modelscope.cn/datasets/aihobbyist/Genshin_Dataset

License & Copyright

All assets are © HoYoverse / miHoYo. Genshin Impact is a copyrighted work of miHoYo Co., Ltd. (HoYoverse). This repository is a fan-made archive intended solely for research and educational purposes; no commercial use is permitted. We will respond promptly to any takedown request from the rights holders.

Copyright of the original assets belongs entirely to HoYoverse / miHoYo.
Commercial use of any model or derivative work trained on this dataset is strongly discouraged.
If you redistribute this dataset, please include this notice.

Citation

@misc{genshin_voice_2026,
  title  = {Genshin Impact Voice Lines (Game Version 6.3)},
  author = {ultemica},
  year   = {2026},
  howpublished = {\url{https://huggingface.co/datasets/ultemica/genshin-impact-voices}},
  note   = {Extracted assets © HoYoverse / miHoYo}
}