See bug #352578 for more background.

The Orca speech routines (apart from at startup and shutdown time),
ignore the settings.enableSpeech value. Instead they will speak if
there is a valid acss value passed into them.

When the Orca Preferences GUI is shown (via Insert-Space), part of the
initialisation down in the _init() routine is to work out all the valid
speech servers. This is creating acss values.

There are two possible solutions here:

1/ Adjust one or more of the speech routines to just return if
   settings.enableSpeech is false.

2/ Adjust the orca_gui_prefs.py code to only check for valid speech
   servers if the "Enable Speech" checkbox is checked.