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

The purpose of this paper is to introduce an iterative nonlinear weighted method based on the variation in spectral energy distribution present in a voice signal to differentiate between four voice types: type 1 voice signals are nearly periodic, type 2 voice signals have strong modulations and subharmonics, type 3 signals are chaotic, and type 4 signals are dominated by stochastic noise. A total of 135 voice signal samples of the sustained vowel /a/ were obtained from the Disordered Voice Database and then individually categorized into the appropriate voice types based on the classification system described in Sprecher et al (2010). Voice samples were analyzed using the nonlinear methods of spectrum convergence ratio, rate of divergence, and nonlinear energy difference ratio (NEDR) to investigate classifier efficacy. An iterative nonlinear weighted method based on the derivative of instantaneous frequency and Fourier transformations is applied to calculate spectral energy distributions. The distribution is then used to calculate the NEDR to classify voice signal types. Statistical analysis revealed that NEDR effectively differentiated between all four voice types (P