Influence of sampling rate on voice analysis for assessment of Parkinson's disease

Parkinson's diagnosis through voice analysis (PDVA) has been attracting increasing attention. In this paper, the influence of sampling rate on PDVA is studied. By analyzing the main difficulties that hamper the development of PDVA, the significance of seeking guidelines on sampling rate is discussed. Then voices from both healthy controls and patients with Parkinson's disease are recorded, for which the sampling rate used is given special consideration. Recordings of other sampling rates are generated via down sampling the recorded voices. Then it is proposed to adopt six metrics from four levels to assess the impacts of sampling rate, which are information entropy, reconstruction error, feature correlation, classification accuracy, computational cost, and the storage cost. Through extensive experiments, basic guideline to seek an appropriate sampling rate is provided. It is concluded that a sampling rate of 96 kHz is preferred when no limits of storage and computational costs are imposed. However, a lower sampling rate may be needed if the storage size and computational complexity are the main concerns.