It seems that somehow this 2 languages switched places. No matter what language I choose for Conversation agent (HA) Serbian or Serbian (Latin), input for TTS is in Serbian (Latin) when responding. Microsoft STT can handle both alphabets and when Serbian is selected in Conversation agent correctly uses Cyrillic alphabet. When Serbian (Latin) is selected I get Latin alphabet for Serbian, that is OK.
What is not OK is that language for response stays always on Serbian (Latin) and Microsoft TTS can't process that.
Any advice is welcome, thanks