Parkinson disease (PD) is the second most prevalent neurological disorder in the world. Although PD diagnosis is relevant due to possible misdiagnosis, disease monitoring is also a critical necessity, especially for patients who live in the countryside, where accessing neurologist experts is not feasible. To contribute to the process of designing strategies for remote monitoring of patients, this work evaluates the suitability of different dimensions of speech, namely articulation, prosody, and phonemic identifiability, to effectively model disease progression. In this study, two different cohorts of participants are considered. The PC-GITA corpus is used as a training set, and a cohort with fifteen patients recorded three times along a period of three years approximately is considered to perform the longitudinal test. Support vector machine classifiers (SVMs) are optimized in training. Afterward, the distance of each test sample to the SVM separating hyperplane is used as a biomarker to evaluate whether the model can effectively predict disease progression. Promising results are observed, indicating that articulation and phonological features are suitable to perform automatic evaluation of PD progression.