Automatic Restoration of Diacritics for Speech Data Sets