Meta-Learning Approaches for Speaker-Dependent Voice Fatigue Models