Inverse filtering applied to upper airway sounds

Abstract

Inverse filtering is a digital signal processing technique which may be applied to speech-like sounds to remove resonances introduced by upper airway cavities to leave a residual signal which is, in principle, spectrally flat and strongly related to the excitation source. The filter parameters, normally computed by a form of linear prediction analysis, are indicative of the frequencies and bandwidths of the resonances. This paper briefly outlines the principle of inverse filtering and describes two applications in the study of upper airway sounds for diagnostic purposes. The first application is concerned with the non-invasive measurement of variations in upper airway dimensions which occur with changes in posture. Results show that differences in the resonance frequencies caused by changes in posture can be measured, these being of the order of about 10% in normals. The measurement of such changes is known to be useful in the assessment of patients with sleep apnoea. The second application concerns the evaluation of vocal tract abnormalities resulting from infection in the larynx. Parameters derived from the residual are believed to be indicative of the existence and severity of a hoarse voice. Results have been obtained which support this theory.