Average-Voice-Based Speech Synthesis Using HSMM-Based Speaker Adaptation and Adaptive Training