Interpreting Pretrained Speech Models for Automatic Speech Assessment of Voice Disorders