Saving and loading voices in MaryTTS and AbstractOutputNode

[alexanderkoller]

Describe the bug When saving and loading dialogs to/from a file, information about voices is not handled correctly:

• the MaryTTS Settings does not save the default voice at all;
• the AbstractOutputNode can create invalid voices or duplicate the default-voice entry in the combobox.

To Reproduce It may be enough to select a non-default voice in a MaryTTS node and then save+load the dialog, but I have not tested this systematically. The problem arises because the default voice is selected first in AbstractOutputNode#readAttributes, and then overwritten by the correct value.

I am not entirely sure that there is a generic way to save and load voice information, because the voice objects are regenerated at each runtime and I don't quite trust the consistency of the voice names. This needs some looking into.

[alexanderkoller]

The voice handling in the Google TTS plugin seems to work correctly.