The technical term is Inverse Text Normalization (ITN).

If you search for it, you will find huge collections of rules and data.

Here are the best resources for filling maps without typing everything yourself:

1. ITN rule collections (the “gold standard”)

  • itnpy: A simple, deterministic Python tool designed for this very purpose. It uses CSV files to convert spoken words into written characters (numbers, currencies, dates). You can copy the CSVs almost 1:1 into your map.

  • NVIDIA NeMo ITN: Very powerful. They have huge grammar files for almost all languages. There you will find lists for units of measurement, titles, and date formats.

2. Data sources for punctuation & case

  • Vosk recasepunc: This is the standard tool for Vosk. It uses models, but the source code often contains lists of abbreviations and proper names that can be extracted.

  • Google Text Normalization Dataset: A huge dataset (created for a Kaggle challenge) containing millions of examples of how spoken language is converted into written language.

3. “Dictation helper” libraries

  • num2words: If you need number mapping, you can find lists for “one” to “one million” here.