Voice pathology detection based eon short-term jitter estimations in running speech

Area of Science:

• Speech Science
• Biomedical Engineering
• Acoustic Analysis

Background:

• Voice pathology detection often relies on sustained phonations.
• Continuous speech analysis presents challenges due to natural variations.
• Jitter, a measure of pitch period variability, is a key indicator of voice dysfunction.

Purpose of the Study:

• To evaluate short-term jitter estimation using the spectral jitter estimator (SJE) for voice pathology detection in continuous speech.
• To establish thresholds and features for SJE in sustained vowels and reading tasks.
• To assess the performance of SJE-based methods in discriminating between healthy and pathological voices.

Main Methods:

• Utilized the spectral jitter estimator (SJE) for short-term jitter estimation.
• Cross-database validation was performed on sustained vowel recordings from healthy and pathological voices.
• Developed and validated new thresholds and features for continuous speech (reading text) analysis.
• Employed receiver operating characteristic (ROC) curves and area under the curve (AUC) to evaluate detection performance.

Main Results:

• SJE demonstrated robustness against pitch estimation errors, suitable for continuous speech.
• A threshold for sustained vowels showed good cross-database validation.
• New features and a second threshold for continuous speech yielded high discrimination scores (AUC ~95% for features, 87.8% for threshold).
• Confirmed inverse relationship between jitter and fundamental frequency, and higher jitter in pathological voices.

Conclusions:

• Short-term jitter estimation via SJE is effective for voice pathology detection in continuous speech.
• The proposed thresholds and features offer reliable tools for clinical voice assessment.
• Findings support the utility of acoustic analysis of continuous speech for understanding voice disorders.