Analysis of Predictive Coding Models for Phonemic Representations in Small Datasets