Automatic Detection of Parkinson's Disease in Reverberant Environments

摘要

Automatic classification of speakers with Parkinson's disease (PD) and healthy controls (HC) is performed considering a method for the characterization of the speech signals which is based on the estimation of the energy content of the unvoiced frames. The method is tested with recordings of three languages: Spanish, German, and Czech. Additionally, the signals are affected by two different reverberant scenarios in order to validate the robustness of the proposed method. The obtained results range from 85% to 99% of accuracy depending on the speech task, the spoken language, and the recording scenario. The method shows to be accurate and robust even when the signals are reverberated. This work is a step forward to the development of methods to assess the speech of PD patients without requiring special acoustic conditions.