Evaluation of objective measures for intelligibility prediction of HMM-based synthetic speech in noise