Detection of hypernasal speech using a nonlinear operator

Abstract

Hypernasal speech is indicative of an underlying problem in the speech production system that results from either a physical defect, or a dysfunctional nervous system. Current clinical methods for determining whether speech is normal or hypernasal are invasive or awkward. A noninvasive system is proposed to classify speech as normal or hypernasal based on the nonlinear Teager Energy operator. The performance of the system is evaluated using normal and simulated hypernasal speech. Results show that the Teager Energy metric reliably detects the presence of hypernasality in a speech sample. >