Evaluation of Speech Representations for MOS prediction