Parkinson’s disease patients develop several impairments related to the speech production process. The deficits of the speech of the patients include reduction in the phonation, articulation, prosody and intelligibility capabilities. Related studies have analyzed the phonation, articulation and prosody of the patients with Parkinson’s, while the intelligibility impairments have not been enough evaluated. In this study we propose two novel features based on the word accuracy and the dynamic time warping algorithm with the aim of assess the intelligibility deficits of the patients using an automatic speech recognition system. We evaluate the suitability of the features by the automatic classification of utterances of patients vs. healthy controls, and by predicting automatically the neurological state of the patients. According to results, an accuracy of up to 92% is obtained, indicating that the proposed features are highly accurate to detect Parkinson’s disease from speech. Regarding the automatic monitoring of the neurological state, the proposed approach could be used as complement of other features derived from speech or other bio-signals to monitor the neurological state of the patients.