Objective: To investigate the validity and robustness of an automatically generated intelligibility score in Parkinson’s disease (PD) across multiple languages.
Background: 90% of people with Parkinson’s disease (pwPD) develop speech disorders, significantly affecting intelligibility [1]. The gold standard for evaluating PD speech relies on subjective perceptual ratings. Some studies link reduced intelligibility to increased motor impairment, while others report only a weak association [2], [3]. These mixed findings drive interest in automated speech analysis for objective assessment. To be effective, such analyses must generalize across languages to serve the diverse PD population.
Method:
Acoustic speech samples from a reading task were collected from pwPD across three languages: Spanish (n = 54), German (n = 119), and Czech (n = 50). All pwPD were recorded in the medication-ON state, and motor impairment was assessed using the UPDRS III.
The data were processed using ki:elements SIGMA pipeline to calculate the ki: SB-M intelligibility score. This score is built automatically by i) transcribing speech using the Google Speech-to-Text API, ii) comparing the generated scripts to the corresponding reference text, and iii) calculation of word error rate (WER) between the transcribed words and the reference text. The lower the score, the better speech intelligibility. Spearman Rank-Sum correlations were computed between the intelligibility score and the overall UPDRS score in the overall sample and in each of the languages separately.
Results: Mean intelligibility scores per language as well as average UPDRS III scores are presented in [table1]. Correlation analyses revealed that the intelligibility score correlates with UPDRS III total score. Trend indicates that intelligibility score has a positive correlation in all cases. While the strongest correlations can be observed for Spanish, low correlations are present in the Czech data probably due to early PD stage not expecting pronounced motor impairment.
Conclusion: Increased motor impairment in Parkinson’s disease is associated with reduced speech intelligibility emphasizing the importance of regular speech evaluations in PD. Automated intelligibility scores offers a scalable and sensitive alternative to traditional assessments and holds promise for an objective evaluation of speech and motor status in PD.
Table1
References: [1] J. R. Duffy, Motor speech disorders: substrates, differential diagnosis, and management, Third edition. St. Louis, Missouri: Elsevier, 2013.
[2] T. Thies, Tongue Body Kinematics in Parkinson’s Disease. 2023. Accessed: Mar. 13, 2025. [Online]. Available: https://www.peterlang.com/document/1314093
[3] N. Miller, L. Allcock, D. Jones, E. Noble, A. J. Hildreth, and D. J. Burn, “Prevalence and pattern of perceived intelligibility changes in Parkinson’s disease,” J. Neurol. Neurosurg. Amp Psychiatry, vol. 78, no. 11, pp. 1188–1190, Nov. 2007, doi: 10.1136/jnnp.2006.110171.
To cite this abstract in AMA style:
T. Thies, F. Dörr, A. König, N. Linz, M. Barbe, J. Orozco-Arroyave, J. Rusz, J. Tröger. Automatic Intelligibility Rating in Parkinson’s Disease: A Multilingual Approach [abstract]. Mov Disord. 2025; 40 (suppl 1). https://www.mdsabstracts.org/abstract/automatic-intelligibility-rating-in-parkinsons-disease-a-multilingual-approach/. Accessed October 5, 2025.« Back to 2025 International Congress
MDS Abstracts - https://www.mdsabstracts.org/abstract/automatic-intelligibility-rating-in-parkinsons-disease-a-multilingual-approach/