Articulatory feature based continuous speech recognition using probabilistic lexical modeling