Question: About romanized languages detection

[loretoparisi]

First thanks a lot for this amazing work! I'm trying to address to task of the language detection for indian languages. Currently my neural network can detected with a good accuracy most of the languages I wanted to have. The problem is that I need to detect the romanized version of these language, but I'm not sure that training that network (fasttext) on a romanized languages corpus could bring to something meaningful. I can see that here this tool is using ngram - trigram and a frequency based model.
My question is if this approach could be used for romanized version of a language like "hin", or "urd", etc. or this cannot be done in any case.
Thanks a lot.