AIAny
Icon for item

Joyo Kanji Yomi Benchmark

Provides kanji-level evaluation data for Japanese TTS: disambiguated sentence contexts targeting 4,378 kanji-reading pairs (2,136 Jōyō kanji) with 13,095 native-speaker–verified sentences and katakana-marked ground-truth readings for kanji-level error metrics.

Introduction

Oops! Something went wrong

[next-mdx-remote-client] error compiling MDX: Expected a closing tag for `<>` (6:125-6:127) before the end of `paragraph` 4 | - Coverage and granularity: covers all 2,136 Jōyō kanji and 4,378 kanji-reading pairs with three sentence contexts per reading (13,095 sentences), so you can evaluate per-reading behaviour rather than only word- or sentence-level quality — useful for pinpointing specific polyphony errors. 5 | - Native verification and disambiguation: sentences and annotations were reviewed by 35 native Japanese speakers through a multi-stage process, and ambiguous kanji-reading pairs that cannot be uniquely disambiguated by context were excluded — improving label reliability for evaluation. > 6 | - Evaluation-ready format: each sample includes a full-sentence katakana transcription with the target reading delimited by <> for automatic extraction; an accompanying evaluation toolkit handles TTS synthesis, ASR transcription, alignment, and metric computation, streamlining kanji-level experiments. | ^ 7 | - Focused on TTS/ASR pipelines: designed to measure pronunciation selection in synthesized speech (kanji→phoneme mapping under sentential context), not as a general-purpose language modeling corpus. 8 | More information: https://mdxjs.com/docs/troubleshooting-mdx

Information

  • Websitehuggingface.co
  • Organizationssbintuitions
  • Published date2026/06/11

Categories

More Items

Hugging Face

Provides about 10,000 OpenUSD objects for robotics simulation, including articulated and static assets with physics properties and robot-interaction annotations.

Hugging Face

Pairs metric-scale 3D scans of real environments with egocentric task recordings, estimated body and hand motion, and hierarchical annotations for physical AI research. The first release contains 126 scenes; the dataset is licensed CC BY-NC 4.0.

Turns LLMs’ physical knowledge into video training data using simulator constraints, physical verification, and iterative correction. It also repairs physical errors in generated videos and releases 100K videos spanning 2,500 physical events.