NandemoGHS/Japanese-Eroge-Voice

Japanese-Eroge-Voice

Description

This dataset contains pairs of audio data and corresponding transcriptions extracted from Japanese eroge (adult games) that I have personally purchased. The transcriptions are generated using the [litagin/anime-whisper](https://huggingface.co/litagin/anime-whisper) model.

Preprocessing Steps

The raw audio data has undergone the following preprocessing steps:

1. 1.Loudness Normalization: Audio loudness is normalized using ffmpeg's 2-pass `loudnorm` filter to target parameters of -23.0 LUFS integrated loudness, -1.0 dB true peak, and 11.0 LU loudness range (LRA).

1. 1.Transcription: Each clip is transcribed with [litagin/anime-whisper](https://huggingface.co/litagin/anime-whisper) model.

1. 1.Data Shuffling, Anonymization, and WebDataset Conversion: The processed data is shuffled, and unique identifiers (UIDs) are hashed for anonymization. The data is then packaged into [WebDataset](https://github.com/webdataset/webdataset) format. Due to the shuffling and anonymization, it is difficult to reconstruct the original works in their entirety, aiming to limit the enjoyment of the original copyrighted works under Japanese copyright law.

Dataset Format

This dataset is provided in WebDataset format. Each .tar shard typically contains 1024 clips (the final shard may contain fewer). Every clip consists of three files sharing the same base filename:

0a1be9a22ae956c9.flac  (FLAC audio file)
0a1be9a22ae956c9.json  (JSON metadata file)
0a1be9a22ae956c9.txt   (Text transcription file)
...

Dataset Statistics

• —Total duration: 409.33 hours

Biases and Limitations

• —NSFW Content: This dataset is derived from adult games and contains a significant amount of Not-Safe-For-Work (NSFW) content.
• —Gender Bias: Due to the nature of the source material, the dataset is heavily skewed towards female voices.
• —Potential Transcription Errors: Transcriptions are generated automatically by an AI model and have not been manually verified. They are likely to contain errors and inaccuracies.

License

This dataset is licensed under the [MIT License](https://choosealicense.com/licenses/mit/).

Intended use — This dataset is primarily designed for educational and academic research. All use is at your own risk, and you must ensure compliance with applicable law.

NO WARRANTY – This dataset is provided “as is” without any express or implied warranty.

How to Cite

@misc{Japanese-Eroge-Voice,
  title        = {Japanese-Eroge-Voice},
  author       = {OmniAICreator},
  year         = {2025},
  howpublished = {\url{https://huggingface.co/datasets/NandemoGHS/Japanese-Eroge-Voice}},
}