Tacotron ignores marks on the end of the sentences

Hello everyone!

Is it possible to make Tacotron 2 pay attention to the end of the sentence, such as ‘!’, ‘?’
I’ve got my own small dataset, where I carefully pronounce sentences with the right intonation, but cant make the Tacotron 2 do the same.
Should be here added an extra embedding layer on the characters and mixed with the rest of the model, or anything?
Maybe I need to use a Tacotron GST for that purpose?

Thank you a lot!