Application of Inverse Filtering for Detecting Laryngeal Pathology

Abstract

Inverse filtering analysis was performed on acoustic speech samples of the vowel sound “ah” produced by ten selected patients and ten normal adults. The inverse filter output signal is the residue obtained by automatically extracting the effects of the supraglottal structure from the acoustic speech signal. For normal speakers, the residue consists of a series of rather sharp periodic spikes having low amplitude noise between periods. For pathological voices, the residue consists of lower amplitude, less sharp spikes having higher amplitude noise between periods. Comparisons between the acoustic speech waveforms and the residue signals illustrate the superiority of the residue signal for detecting irregularity, and thus suggest the feasibility of applying inverse filtering as an aid to laryngeal diagnostics.