Exploiting Nonlinear Recurrence and Fractal Scaling Properties for Voice Disorder Detection

Abstract: Voice disorders affect patients profoundly, and acoustic tools can potentially measure voice function objectively. Nonetheless, existing tools are limited to analysing voices displaying near periodicity, and do not account for inherent biophysical nonlinearity and non-Gaussian randomness. They do not directly measure complex nonlinear aperiodicity, and turbulent, aeroacoustic, non-Gaussian randomness. Often these tools have limited clinical usefulness. This paper introduces two new tools to speech analysis: recurrence and fractal scaling, which overcome the range limitations of existing tools by addressing directly these two symptoms of disorder, and a simple bootstrapped classifier distinguishes normal from disordered voices to 91.8% overall accuracy on a large database of subjects with a wide variety of voice disorders. They are widely applicable to the whole range of disordered voice phenomena by design. These new measures could therefore be used for a variety of practical clinical purposes.

• The paper introduces recurrence analysis and fractal scaling methods to capture nonlinear patterns in voice signals, enhancing diagnostic accuracy.
• It applies time-delay embedding and detrended fluctuation analysis to quantify vibratory irregularities and breath noise in disordered voices.
• The approach achieves 91.8% classification accuracy with robust true positive and negative rates, demonstrating significant clinical potential.

Analyzing Voice Disorder Detection Through Nonlinear Recurrence and Fractal Scaling

In the domain of voice disorder detection, traditional linear signal processing methods have shown limitations, particularly when faced with disordered voices characterized by marked nonlinearities and turbulence. The paper "Exploiting Nonlinear Recurrence and Fractal Scaling Properties for Voice Disorder Detection" by Little et al. introduces an innovative approach to address these challenges by employing nonlinear dynamical systems theory.

Background and Motivation

Voice disorders manifest with increased vibrational aperiodicity and augmented breathiness due to complex physiological and psychological causative factors. Conventional acoustic tools struggle in diagnostic applications as they predominantly cater to near-periodic signals, leaving a substantial gap in effectively addressing the biophysical intricacies of disordered voices. This study is propelled by the need to develop methodologies that capture the inherent nonlinear and non-Gaussian characteristics of disordered voice signals, providing a more holistic evaluation.

Methodology

To counteract the inadequacies of existing approaches, the paper introduces two novel analytical tools in speech analysis: recurrence analysis and fractal scaling.

1. Recurrence Analysis: Utilizes the concept of state-space trajectories, identifying aperiodicity by examining recurrence time statistics through time-delay embedding techniques. This method adapts the concept of close returns, traditionally applied to deterministic systems, to manage both deterministic and stochastic signals. By assessing the recurrence period density entropy (RPDE), this tool allows normal and disordered voices to be differentiated along a spectrum based on the degree of irregularity present in their vibrational patterns.
2. Fractal Scaling: Employs detrended fluctuation analysis (DFA) to measure the statistical self-similarity of the signal. Fractal scaling evaluates breath noise, a key feature of disordered voices, by analyzing the scaling properties of random fluctuations apparent in the signal. The calculated scaling exponents provide a clear distinction between normal and disordered phonations, allowing further categorization of voice pathology.

Both methods conclude with outputs that can be projected onto a normalised scale ranging between zero and one, respectively representing perfect periodicity and extreme stochastic turbulence. The recurrence of standardised, reliable results from such measures is essential for clinical applicability.

Results

The analytical framework was tested on the extensive Kay Elemetrics disordered voice database, containing a wide variety of voice disorders. The RPDE and DFA measures achieved a robust correct classification performance of 91.8%, with a true positive rate of 95.4% and a true negative rate of 91.5%. Such performance surpasses that of conventional measures, which often fail to classify severely disordered voices accurately.

Discussion and Implications

The study's outcomes indicate that the proposed measures significantly reduce complexity while enhancing classification accuracy, offering a practical alternative to existing perturbation methods. By leveraging a minimal set of arbitrary parameters, these new measures deliver improved reproducibility and clinical utility. Nonetheless, certain constraints such as vocalization consistency requirements highlight potential areas for refinement. Future advancements might include optimizing the parameter choices for broader applicability and sensitivity.

The implications of this research signal a shift towards more integrated approaches in voice disorder diagnostics, potentially applicable beyond vocal pathophysiology to other nonlinear biosignal processing domains.

Conclusion

In summary, the paper makes substantial contributions toward advancing the detection of voice disorders through the integration of nonlinear dynamical tools. By addressing the biophysical complexity inherent in disordered phonation, the study establishes a foundation for future methods that are both clinically effective and theoretically sound. Further exploration may extend the applicability of these techniques across diverse speech-related and nonlinear signal processing challenges.