Normalized noise energy as an acoustic measure to evaluate pathologic voice

Abstract

In order to evaluate noise components included in pathologic voice signals, a novel acoustic measure, normalized noise energy (NNE), is proposed and its effectiveness for the detection of laryngeal pathologies is investigated with 250 vowel samples spoken by 64 control (normal) subjects and 186 patients with various laryngeal diseases. The NNE is automatically computed from the voice signals using an adaptive comb filtering method performed in the frequency domain. Experiments with the voice samples show that the NNE is especially effective for detecting glottic cancer, recurrent nerve paralysis, and vocal cord nodules. Specifically, when glottic cancer is represented in terms of the T classification adopted by the UICC (Union Internationale Contre le Cancer), glottic T2-T4 cancer can be perfectly discriminated from normal samples, but 22.6% of patients with glottic T1 cancer are incorrectly classified as normal, with an error rate of 9.4% for normal subjects.