Speech biomarkers of disease have attracted increased clinical interest in recent years, but interpretation of clinical features derived from signal processing or machine learning approaches remains challenging. As an example, the second Mel frequency cepstral coefficient (MFCC2) has been identified in several studies as a useful marker of disease, but continues to be treated as uninterpretable. Here we show that MFCC2 can be interpreted as a weighted ratio of low- to high-frequency energy, a concept which has been previously linked to disease-induced increases in aspiration noise caused by incomplete vocal fold closure, and also show how its sensitivity to disease can be increased by adjusting computation parameters.