Minimal Perturbation-Based Segment Selection in Connected Speech Differentiates Healthy and Pathological Voice
Owen P Wischhoff1, Maiwand M Tarazi1, Jakob R Holm1
1Department of Otolaryngology-Head and Neck Surgery, University of Wisconsin School of Medicine and Public Health, Madison, Wisconsin.
Summary
This study shows that analyzing connected speech with a moving-window technique can effectively distinguish healthy from pathological voices. Spectral and nonlinear dynamic measures, particularly cepstral peak prominence (CPP), demonstrated the strongest voice disorder detection capabilities.
Area of Science:
- Speech and Hearing Sciences
- Acoustic Phonetics
- Medical Diagnostics
Background:
- Traditional voice analysis often uses sustained vowels, which may not reflect real-world speech patterns.
- The minimal perturbation method suggests that the most regular phonation in speech indicates laryngeal system function.
- Evaluating connected speech offers a more ecologically valid approach to voice assessment.
Purpose of the Study:
- To apply a moving-window segmentation technique to connected speech.
- To evaluate nonlinear dynamic, perturbation, and spectral acoustic measures for differentiating healthy and pathological voices.
- To assess the efficacy of the minimal perturbation method in running speech.
Main Methods:
- Eighty-two connected speech recordings (41 healthy, 41 pathological) were analyzed.
- A moving-window approach (0.8s windows, 0.25s shift) was used on the Rainbow Passage.
- Acoustic measures included nonlinear energy difference ratio (NEDR), weighted index of dysphonia (WID), voice-type component profile (VTCP), jitter, shimmer, cepstral peak prominence (CPP), and signal-to-noise ratio (SNR).
Main Results:
- Seven of ten acoustic measures significantly differentiated between healthy and pathological voices after correction for multiple comparisons.
- Cepstral peak prominence (CPP) showed the largest effect size (d=1.953).
- Other significant measures included VTC1, WID, NEDR, SNR, jitter, and shimmer, indicating their utility in voice disorder detection.
Conclusions:
- The moving-window method successfully classified voices without sustained vowels, validating its use in connected speech.
- Spectral and nonlinear dynamic measures were most effective, while segment-level optimization maintained the discriminative power of perturbation measures.
- Future research should explore optimal windowing strategies and composite acoustic analysis for enhanced voice assessment.