Despite recent developments, ASR systems still struggle to handle atypical speech. This difference in performance has been leveraged to use ASR to automatically assess speech quality in the context of speech disorders. In the first part of this paper, we confirm the correlation between ASR sensitivity and speech quality. At the same time, ASR systems remain black-box and difficult to interpret. Therefore, we explore ASR performance for Head and Neck Cancer and Parkinson's disease speech disorders in the second part of this paper. We hypothesise that ASR errors are representative of pronunciation patterns caused by these disorders. We build pronunciation profiles from ASR transcription errors for each category and assess the representativeness of these profiles by synthesising speech variants including the variations identified in the profiles. Analysis of these variants shows that ASR transcription errors are characteristic of pronunciation patterns caused by speech disorders.