This repository provides an openjtalk-style format of tdmelodic, an accent dictionary of Japanese that is robust to many neologisms.
- tdmelodic: an opensource accent dictionary Japanese that estimates the accent of the words listed in NEologd dictionary by a machine learning-based technique.
- openjtalk: an opensource Japanese text-to-speech (TTS) engine that includes TTS front-end, e.g., full-context labels generation. Its Python wrapper pyopenjtalk is used in ESPnet.
You can install this dictionary by adding it to the pre-stored dictionaries.
- Download
tdmelodic_openjtalk.csv
- Download openjtalk from http://open-jtalk.sp.nitech.ac.jp/.
- Unzip the downloaded zip file and move to the directory.
- concatenate
tdmelodic_openjtalk.csv
tomecab-naist-jdic/*.csv
, e.g.,
$ cp mecab-naist-jdic/unidic-csj.csv mecab-naist-jdic/unidic-csj_org.csv
$ cat tdmelodic_openjtalk.csv mecab-naist-jdic/unidic-csj_org.csv > mecab-naist-jdic/unidic-csj.csv
- Compile the source code of openjtalk.
for pyopenjtalk (Python)
Someone, tell me the way.
# clone pyopenjtalk to your pc
$ git clone https://github.com/r9y9/pyopenjtalk.git
$ cd pyopenjtalk
# clone submodules
$ git submodule update --recursive --init
# next append the dic file
$ cp lib/open_jtalk/src/mecab-naist-jdic/naist-jdic.csv lib/open_jtalk/src/mecab-naist-jdic/naist-jdic_org.csv
$ cat this-repo/tdmelodic_openjtalk.csv lib/open_jtalk/src/mecab-naist-jdic/naist-jdic_org.csv > lib/open_jtalk/src/mecab-naist-jdic/naist-jdic.csv
# finally install modified pyopenjtalk
$ pip install -e .
- Shinnosuke Takamichi (Univ. of Tokyo, Japan)
- Junya Koguchi (Meiji Univ., Japan)
- Non-commercial purpose (research purpose) only