Uses rules to split bits of text into graphems more similar to their pronunciation. Example: 1993 -> mille - neuf - cent - quatre - vingt¬ - treize All the other models and tools operate on deformated text.
The rules, their applying orders, and test cases are versionned in yaml files Hosted on a Github Repo
from asr_toolkit.format import FormatHandler as FH
format_handler = FH('assets_dir')
deformatted = format_handler.deformat('1,23 mm³')
# -> "1 - virgule - vingt - trois millimètres cubes"
reformatted = format_handler.format(deformatted)
# -> "1,23 mm³"