Context-Dependent Multiple Distribution Phonetic Modeling with MLPs