Investigation of the Effects of Speech Signal Length on Vocal Disorder Sorting Done Via Dynamic Pattern Modeling

Area of Science:

• Medical Acoustics
• Speech Pathology
• Signal Processing

Background:

• Noninvasive diagnosis of vocal fold diseases is crucial for effective vocal analysis.
• Dynamic pattern modeling offers potential for detecting speech production defects due to signal time variations.

Purpose of the Study:

• To develop a noninvasive method for differentiating vocal fold diseases using dynamic pattern modeling.
• To investigate the impact of vocal signal length on the accuracy of disease classification.

Main Methods:

• Utilized the hidden Markov model (HMM), a state-space model, for classifying specific vocal fold diseases.
• Analyzed vocal signals of varying lengths (1, 3, and 5 seconds) to assess their effect on classification accuracy.

Main Results:

• Experimental results demonstrated that conditions like vocal fold cysts, false vocal cord issues, and masses are more discernible in longer, continuous voice productions.
• Classification accuracy for pathologic voice signals significantly improved with increased signal length when using dynamic modeling.

Conclusions:

• Dynamic modeling of longer vocal signals enhances the recognition accuracy for certain vocal fold pathologies.
• The hidden Markov model approach shows promise for noninvasive differentiation of vocal fold diseases based on signal length.