Spectral energy bands for laryngeal pathologies discrimination in speech signals : Healthy and unhealthy voices discrimination, and pathology discrimination

Abstract

This work presents a model to discriminate between healthy and unhealthy voices with physiological laryngeal pathologies, between healthy and unhealthy voices with neuromuscular laryngeal pathologies, and between pathological voices with both types of mentioned pathologies. The model is based on the analysis of speech signal energy in different frequency bands, specifically in its mean value and variation over the signal. The accuracy rates obtained were 100% when discriminating between healthy and unhealthy voices with physiological laryngeal pathologies, 96.55% when discriminating between healthy and unhealthy voices with neuromuscular pathologies, and 93.48% when discriminating between physiological and neuromuscular laryngeal pathologies. The results demonstrate that certain frequency bands contain the information needed for the three discrimination processes performed.