Text this: Modeling monophthongal versus diphthongal /aɪ/ in sung vocal performance with interpretable machine learning.