JSUT (Japanese speech corpus of Saruwatari-lab, U-Tokyo)

A ~10-hour single-speaker Japanese read-speech corpus designed for end-to-end TTS, covering the main pronunciations of daily-use Japanese characters.

• Task: tts
• Languages: ja
• Hours: 10
• Domain: read (studio, single female speaker)
• License: see source terms
• Homepage: https://sites.google.com/site/shinnosuketakamichi/publication/jsut
• Paper: https://arxiv.org/abs/1711.00354

Recommendation

Strong default for Japanese single-speaker TTS baselines and research prototypes — the canonical reference corpus alongside JVS. Pick when you need a clean, well-known ja-JP voice for studio-quality TTS.

Getting the data

Obtain from the dataset homepage.

License is split — text is CC BY-SA 4.0, audio is research-only by default; commercial audio use requires emailing Takamichi / U-Tokyo TLO for permission.

Suggested processing

A recommended VoxKitchen pipeline ships in the repository at voxkitchen/templates/pipelines/tts-data-prep.yaml — run it with vkit docker run.