The best speech task for differentiating between parkinsonian and healthy speech : Integrative literature review

Abstract

The aim of this study was to determine which speech task is the best for automatically identifying Parkinson's disease based on speech signal. The second aim was to examine whether different languages differ in which speech task is best used to identify Parkinson's disease. The study was conducted as an integrative literature review. Seven peer-reviewed scientific research articles were selected for the data based on predefined inclusion and exclusion criteria. The speech tasks, languages, classification algorithms, and results were extracted from the articles selected for the review. The results were examined both collectively for all languages studied and separately classified by language.

The articles selected for the review examined a total of six different speech tasks: prolonged vowel, rapid syllable repetition, reading single words, reading single sentences, reading text, and spontaneous speech. The results suggest that Parkinson's disease can best be identified based on spontaneous speech. The articles examined speech signals recorded from speakers of six languages: Spanish, Finnish, German, Czech, Lithuanian and English. Six of the seven articles examined Spanish speakers. The other languages were examined in one article each. Based on the results, the best speech task for Spanish and English speakers to identify Parkinson's disease was spontaneous speech. In contrast, the best results were obtained with prolonged vowel for Lithuanian speakers, with reading text for Finnish speakers, and for German and Czech speakers with rapid syllable repetition. It is noteworthy, however, that spontaneous speech was not examined in the data for Lithuanian, German and Czech speakers.

The results of the study can be utilized for the development of new cost-effective methods for diagnosing Parkinson's disease at an earlier stage.