Integrating articulatory features using Kullback-Leibler divergence based acoustic model for phoneme recognition