German small model and the umlaut

Hello, folks.

I am trying out the small German model "vosk-model-small-de-0.15" by passing in grammar to the recognizer. What I am experiencing is that it works great, except for when a word that is passed in grammar contains an umlaut.

For instance, passing in the following json:
"[\"eins\",\"zwei\",\"drei\",\"vier\",\"fünf\",\"sechs\",\"sieben\",\"acht\",\"neun\",\"zehn\",\"elf\",\"zwölf\"]"

The recognizer will respond perfectly with everything but "fünf" and "zwölf".

I tried with the large model "vosk-model-de-0.21", and it seems to work with words that contain an umlaut. Is there something I am missing in regard to the small model or is this a known limitation? I haven't been able to find information on this, so I am hoping it is just me ;)

Thank you for all of your hard work on this.