Automatic selection of recognition errors by respeaking the intended text | IEEE Conference Publication | IEEE Xplore