OpenJTalk Dictionary for Piper (pyopenjtalk-plus)

openjtalk.zip contains the compiled MeCab/OpenJTalk system dictionary used by the Japanese Piper voices (for example ja_JP-hi_fi_captain-medium).

This is the custom dictionary bundled with pyopenjtalk-plus (mecab-naist-jdic base + jpreprocess fixes and corrections) - the same dictionary the Japanese Piper voices were trained with. Using the stock open_jtalk_dic_utf_8-1.11 dictionary instead produces different readings/accents on some words.

Contents (files at zip root)

sys.dic (103,131,410 bytes)
matrix.bin
char.bin
unk.dic
left-id.def, right-id.def, pos-id.def, rewrite.def
COPYING (dictionary license / attribution)

Unpacked size is about 107 MB. The Piper iOS/macOS app downloads this archive once, on the first Japanese voice install, and stores it in the shared group.pipertts.data/openjtalk folder. The dictionary is memory-mapped by MeCab at runtime, not loaded into heap.

License / attribution

The dictionary is redistributed under its BSD-style license (NAIST + UniDic Consortium); see COPYING inside the archive for the full text and attribution. Source: the pyopenjtalk-plus project (dictionary as shipped in pyopenjtalk-plus 0.4.1), originally derived from mecab-naist-jdic / Open JTalk.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support