On platforms where automatic downloads are enabled (desktop) we
triggered a second download of the voices from our 1st-run-dialog
leading to a race for the target download file.

Now voices are automatically registered later (in Window's
onCompleted()).