Minimal Perturbation-Based Segment Selection in Connected Speech Differentiates Healthy and Pathological Voice.
Owen P Wischhoff, Maiwand M Tarazi, Jakob R Holm and 3 others
A moving-window method applied to connected speech separated 41 healthy from 41 pathological voices across all three acoustic measure groups.
CPP showed the strongest separation. This suggests connected speech is a viable medium for acoustic voice assessment alongside sustained vowels.
Key findings
1All three acoustic measure groups (nonlinear dynamic, perturbation, and spectral) showed significant healthy-versus-pathological differences that survived within-group false discovery rate correction in connected speech.
2CPP produced the largest effect size of any measure in the study (d = 1.953), with pathological speakers showing markedly lower values than healthy speakers.
3The minimal perturbation method, previously validated in sustained vowel phonation, generalizes to connected speech and enables acoustic discrimination without controlled phonatory tasks.
to read the rest of this summary — three full summaries a month, no card.
Already have an account?
What it does not show
The pathological group is heterogeneous, spanning organic, neurological, and functional voice disorders, so the findings cannot be attributed to any single diagnosis. A direct comparison to sustained vowel recordings was not possible because the database cut disordered sustained samples at 1 second, limiting the ability to confirm whether connected speech adds value over the traditional task. Age and sex were not entered as covariates in the primary analysis, so a demographic contribution to the observed differences cannot be ruled out. The study compared only two discrete groups (healthy vs. pathological) and was not designed to determine whether any measure scales proportionally with dysphonia severity. Window parameters (0.8-second windows, 0.25-second shifts) were carried over from sustained-vowel work and have not been optimised for connected speech or for different passage types and speaker populations. All recordings used a single standardised passage (the Rainbow Passage), so generalisability to conversational speech or other texts is untested.
Declared interests
The authors declare no competing financial interests, no personal relationships that could have influenced the work, and no conflicts of interest. No funding source is named.
Summarised by AI from the full paper, without a clinician reviewing it. Check it against the source before it changes what you do. Read it on PubMed →