Text this: Modeling of Speech Parameter Sequence Considering Global Variance for HMM-Based Speech Synthesis