Phonetisaurus G2P model development web app

This is a Python/Flask project that serves as a simple GUI to develop and test Phonetisaurus models.

The Phonetisaurus project is available under https://github.com/AdolfVonKleist/Phonetisaurus It is a very popular G2P tool used (among others) in Kaldi. Another popular alternative is Sequitur G2P.

The following features are included in the current version of the web app:

• upload/view/edit word lists
• upload/view/edit G2P lexica
• upload/delete/download G2P models
• generate lexica from word lists
• train G2P models from lexica
• test G2P models

Some features missing/under development:

• only FST models, for now
• only SRILM used for training
• simplistic file management (no checks for duplicats/security)
• no access control (no users/passwords)
• missing collaboration tools (many people working at the same time)

Usage

If you have Docker, simply run:

docker run --rm -it -p 80:80 danijel3/phonetisaurus-web-app

This will run (interactively) an instance of the server on port 80 on localhost. Simply open http://localhost in your browser and start using.

For a slightly more permenant solution, docker-compose is recommended. Copy the docker-compose.yml to a server of your choice and create the data subdirectory:

mkdir -p data/model data/lex data/vocab

Edit the docker-compose.yml file if necessary and run:

docker-compose up -d

This will keep the server up indefinitely.

Build

The image is built automatically by DockerHub here: https://hub.docker.com/repository/docker/danijel3/phonetisaurus-web-app