This study has investigated the efficiency of voice features in estimating the motor Unified Parkinson's Disease Rating Scale (UPDRS) score in Parkinson's disease (PD) patients. A total of 26 PD patients (mean age = 72) and 22 control subjects (mean age = 66.91) were recruited for the study. The sustained phonation /a/, /u/ and /m/ were collected in both off-state and on-state of Levodopa medication. The average motor UPDRS for PD off-state patients was 27.31, on-state was 20.42 and that of controls was 2.63. Voice features were extracted from the phonation tasks and were reduced to the most relevant 6 features for each phonation task using the Least Absolute Shrinkage and Selection Operator (LASSO) feature ranking method. The correlation between the reduced features and motor UPDRS was tested using the Spearman correlation coefficient test. AdaBoost regression learner was trained and used for automatically estimating the motor UPDRS score using the voice features. The results show that the vocal features for /m/ performed best by estimating the motor UPDRS score for PD off-state with the mean absolute error (MAE) of 3.52 and 5.90 for PD on-state. This study shows that assessment of voice can be used for day to day remote monitoring of PD patients.