An Objective Parameter to Classify Voice Signals Based on Variation in Energy Distribution

Area of Science:

• Acoustics and signal processing
• Biomedical engineering
• Speech science

Background:

• Voice disorders present a diagnostic challenge.
• Objective voice analysis is crucial for diagnosis and treatment monitoring.
• Existing nonlinear methods for voice analysis can be computationally intensive.

Purpose of the Study:

• To introduce an iterative nonlinear weighted method for voice signal analysis.
• To differentiate between four distinct voice types based on spectral energy distribution.
• To establish the efficacy of the nonlinear energy difference ratio (NEDR) as a voice classification metric.

Main Methods:

• Collected 135 sustained vowel /a/ voice samples from the Disordered Voice Database.
• Classified voice samples into four types: nearly periodic, modulated/subharmonic, chaotic, and noise-dominated.
• Applied an iterative nonlinear weighted method using instantaneous frequency derivatives and Fourier transformations to compute spectral energy distributions and NEDR.

Main Results:

• The nonlinear energy difference ratio (NEDR) significantly differentiated between all four voice types (P < 0.001).
• NEDR demonstrated superior classification accuracy (AUC = 0.99) compared to spectrum convergence ratio and rate of divergence.
• NEDR calculations were nearly instantaneous, offering a significant computational advantage.

Conclusions:

• NEDR is an effective and objective metric for classifying voice signal types.
• The rapid calculation of NEDR represents a substantial improvement over previous nonlinear parameters.
• NEDR can aid clinicians in diagnosing voice disorders and monitoring treatment outcomes.