Software for Characterizing Laryngeal Dynamics

Period of Performance: 04/01/1997 - 03/31/1999

Unknown

Phase 2 SBIR

Recipient Firm

Speech Technology/applied Research Corp.
Bedford, MA 01730
Principal Investigator

Abstract

AudioFile proposes to create a software "tool" to characterize aspects of the structure and performance of the larynx from analysis of speech. The analysis relies on the theory of nonlinear dynamical systems, or "chaos theory". It provides parameters (Lyapunov exponents, metric entropy, attractor dimensions) which cannot be derived from conventional reduction but which have already been shown to have value in characterizing the larynx as a dynamical system. The technique is non- invasive, and the software will accept continuous speech and even some unrequested vocalizations, as from neonates. The tool will be capable of detecting both unusual glottal periodicities and complete aperiodicity (chaos). There are both physiological and mathematical reasons to expect some distinct pathologies to excite distinct dynamical responses which the tool can distinguish. Phase I will: verify that the tool can reliably determine the exponents and the attractor dimensions for entries in an existing digital data base of vocalizations; show that the parameters identify distinct modes of stable vibration; and determine the extent to which the parameters vary with known laryngeal conditions. If successful, this software will provide a new set of clinical "measurements" for assisting in the diagnosis of laryngeal pathologies. It can then provide objective corroboration for clinical judgments.