The talk briefly outlines a project undertaken to develop a system to aid clinicians with diagnosis and monitoring of diseases affecting speech. The system was designed to analyse subjects' voices automatically, drawing on linguistic and engineering expertise, with the results of the analysis being directed to an appropriate researcher and/or clinician. This research involved collection of 25,000 subjects' voices over the telephone, initially focussing on subjects suffering from Parkinson's Disease. During this process a number of unforeseen issues became apparent, related to scientific and behavioural preconceptions. This talk focuses on these problems and other challenges often taken for granted by engineers and researchers.