Tools for the construction of an articulatory speech database.

Starting from a plain text corpus, such as a cleaned wikipedia dump, these tools are designed to help the user create a phonetically balanced list of sentences to use as prompts for data collection. The tools were developed for Italian, using the PAISA plain text corpus as the starting point http://www.corpusitaliano.it/en/index.html

text2phones: a scheme script for Italian transciptions using Festival http://www.cstr.ed.ac.uk/projects/festival/

transcribe.py: designed to clean the raw PAISA text and call text2phones to produce transcriptions

triphones.py: contains utilities for collecting statistics on the transcriptions and picking a subset of sentences for best phonetic coverage. Optionally uses the Google Translate API for language verification, since PAISA contains some sentences in other languages, mostly English.

Also included are transcribebrownnltk.py and text2phonesEN for testing the algorithm on English data sets.

The file all_trans.tar.gz is a sample database of Italian sentences taken from classic novels. Included in the triphones.py script is a demo that makes use of this database. To run the demo, first uncompress the file.