We're using MS custom speech for german language in our application. When the user says the decimal number "1653,57" (in german: "Eintausendsechshundertdreiundfünfzig Komma fünf sieben") the MS speech service recognizes "1600 53,57" which are two numbers. Saying the number digit by digit doesn't help either; this results in "165 3,57" which are two numbers too.
The results are nearly the same when sayin "point" instead of "comma": "1600 53.57".
We have not tested it using a speech model for the english language.
I think this is an error in the baseline model. What can we do to be able to recognize decimal numbers?