Text this: Improving ASR performance using context-dependent phoneme model