Parkinson's Disease (PD) is a kind of neurodegenerative disorder. The disease causes communication impairment based on its progression. In general, identification of PD carried out based on medical images of brain. But it was recently identified that voice is acting as biomarkers for several neurological disorders. A review of speech features and machine learning algorithms is presented. This might be helpful for development of a non-invasive signal processing techniques for early detection of PD. Several models developed for disease detection is discussed, which are developed based on features like acoustic, phonation, articulation, dysphonia, etc. Machine learning algorithms like Logistic Regression (LG), Support Vector Machine (SVM), Boosting Regression Tree, bagging Regression, etc., and their performance accuracies in classification of Patient with PD (PWP) and Healthy Controls (HC) are reviewed. All these classification algorithms are trained and tested on several repository corpuses and customized datasets. The Spontaneous Speech (SS) is an efficient tool for the early detection of diseases like Parkinson's, Alzheimer's, Autism and several other dementia types in elderly people.