Acoustic speech metrics that differentiate primary progressive aphasia variants depend on language typology

Abstract

AbstractBackgroundPrimary Progressive Aphasia (PPA) variants can be distinguished using acoustic speech measures. In particular, the syllabic pairwise‐variability‐index (syll‐PVI), which computes syllable duration differences between unstressed and stressed adjacent syllables, can differentiate between the logopenic (lvPPA) and non‐fluent (nfvPPA) variants in English.MethodWe analysed 5 repetitions of 5 multisyllabic words in a cohort of 37 patients (7 svPPA, 14 lvPPA, 16 nfvPPA) and 15 healthy controls. Data were annotated in Praat (manually revised) and metrics computed in R by means of Generalized Linear Mixed Models. The temporal parameters analysed were speech rate (number of sounds per second), average syllable duration (mean duration of each syllable for each word), and syll‐PVI (the normalized difference in duration between the stressed syllable of a word and the preceding one). Four speech language pathologists with extensive clinical experience with PPA performed blinded independent ratings of apraxia of speech severity in a subset of 31 participants.ResultSyllable duration behaves as expected, the control group had the smallest average syllable duration for the five polysyllabic words, followed by the lvPPA, nfvPPA and svPPA groups. The durations are longer in nfvPPA compared to lvPPA (Cohen’s d = 2.249, p = 0.02). Speech rate shows similar results; the largest difference is found between lvPPA and nfPPA, however it does not reach significance (Cohen’s d = 1.434, p = 0.07). Regarding syll‐PVI, nfvPPA tend to have the value closest to 0 (as expected due to greater presence of apraxia of speech), but the effect size does not reach significance for any word. Mean apraxia of speech severity showed higher correlation with syllable duration (spearman coef = 0.290, p = 0.11) than syll‐PVI (spearman coef = ‐0.073, p = 0.69).ConclusionSyllable duration, and not syll‐PVI, can distinguish lvPPA and nfvPPA in Spanish. This is probably due to a typological difference between English (a prototypical stress‐timed language) and Spanish (a prototypical syllable‐timed language) given that the syllabic‐PVI computes differences between duration of unstressed and stressed syllables and those differences are smaller in syllable‐timed languages. Our results highlight the necessity of considering language particularities in the development of speech and language clinical measures.