Sequence-wise Optimization for Quasi-Harmonic Speech Waveform Modeling