Exploiting nonlinear recurrence and fractal scaling properties for voice disorder detection

Area of Science:

• Acoustic analysis
• Speech science
• Biophysics

Background:

• Voice disorders significantly impact patients, necessitating objective acoustic measurement tools.
• Existing acoustic tools are limited, often failing to analyze complex, nonlinear, and non-Gaussian voice signals characteristic of disorders.
• Current methods cannot address key symptoms like nonlinear aperiodicity and turbulent randomness, limiting clinical utility.

Purpose of the Study:

• Introduce novel acoustic analysis tools: recurrence and fractal scaling.
• Overcome limitations of existing methods by directly addressing nonlinear and non-Gaussian properties of disordered voices.
• Develop a classifier for distinguishing normal from disordered voices using these new features.

Main Methods:

• Utilized recurrence and fractal scaling techniques for speech analysis.
• Developed a simple bootstrapped classifier using these two features.
• Applied quadratic discriminant analysis for classification.

Main Results:

• Achieved an overall correct classification performance of 91.8% +/- 2.0% on a large voice disorder database.
• Demonstrated high true positive (95.4% +/- 3.2%) and true negative (91.5% +/- 2.3%) classification rates.
• Outperformed combinations of popular classical acoustic analysis tools.

Conclusions:

• The new recurrence and fractal scaling techniques are simpler and computationally less complex than existing methods.
• These tools achieve clinically useful classification performance by exploiting inherent signal nonlinearity and randomness.
• The methods are widely applicable across the full spectrum of voice disorders, offering practical clinical benefits.