Tools to create ARPA models from cmu pocketsphinx dictionaries for proper g2p generation
Python 33.60%Shell 1.45%C++ 63.20%Makefile 1.75%
g2p's Introduction
Steps
Dependencies
C++11 with TR1 for opemfst is required;
A Linux or OS X environment
Install openfst from deps/ folder (configure with --enable-far)
Install phonetisaurus from root folder
Install mitlm from deps/ folder (requires fortran)
Install Sphinxbase (attention to sphinxbase version)
Sphinxbase: Since gecko-dev uses an outdated version of sphinxbase, this specific commit (https://github.com/cmusphinx/sphinxbase/tree/18aec4d11c5fc724a15f899bc1222bfcfe589def) should be used, otherwise generated models will be incompatible with Gecko's sphinxbase and will fail to load. Build with traditional ./configure && make && make install
Generating the models
From inside dicts/ folder, execute gen_arpa.sh from each language you want arpa models to be build
The .dmp and lm.bin (for faster loading) binary files and its text version (model.arpa) are generated the language root folder
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent