SteinsGateSg/mayuri-voice-dataset

Mayuri Voice Dataset

Hugging Face packaging for the Shiina Mayuri speech dataset used in the mayuri_voice project.

Contents

• data/audio/wav/ Full WAV collection.
• metadata/mayuri_asr_raw.csv Raw transcription CSV.
• metadata/mayuri_ja_filtered.list Filtered GPT-SoVITS training manifest.
• metadata/mayuri_ja_filtered.stats.json Filtering statistics.
• metadata/mayuri_ja_filtered.rejects.csv Rejected samples from the default filtered training set.
• metadata/emotion_labels.csv Emotion labels generated for all filtered samples.
• metadata/emotion_labels.jsonl JSONL version of the emotion labels.
• metadata/emotion_summary.json Summary counts by emotion and reference candidacy.
• metadata/reference_shortlist.csv Top-ranked reference candidates per emotion.
• metadata/dataset_manifest.json Repo-level manifest describing the packaged dataset.

Current Snapshot

• Raw WAV files: 1526
• Filtered training samples: 1416
• Default speaker tag: shiina_mayuri
• Default language tag: ja

Intended Use

This repository is intended for:

• character voice training
• reference bank curation
• GPT-SoVITS style experiments
• emotion-aware reference selection

Upload Notes

• This repository is prepared for Git LFS upload.
• Run git lfs install before pushing large audio files.
• WAV files are tracked via .gitattributes.

Chinese Note

这个目录是面向 Hugging Face 数据集仓库的整理结果，已经把全量音频、训练 manifest、转写、情感标注和参考候选表都放好了，可以直接初始化成独立仓库并上传。