Files in this item

FilesDescriptionFormat

application/pdf

application/pdfSERWY-DISSERTATION-2017.pdf (1MB)
(no description provided)PDF

Description

Title:Hilbert phase methods for glottal activity detection
Author(s):Serwy, Roger David
Director of Research:Hasegawa-Johnson, Mark
Doctoral Committee Chair(s):Hasegawa-Johnson, Mark
Doctoral Committee Member(s):Levinson, Stephen; Oelze, Michael; Viswanath, Pramod
Department / Program:Electrical & Computer Eng
Discipline:Electrical & Computer Engr
Degree Granting Institution:University of Illinois at Urbana-Champaign
Degree:Ph.D.
Genre:Dissertation
Subject(s):Hilbert phase
Analytic phase
Electroglottograph (EGG)
Glottal closure instant (GCI)
Abstract:The 2 pi discontinuities found in the wrapped Hilbert phase of the bandpass-filtered analytic DEGG signal provide accurate candidate locations of glottal closure instances (GCIs). Pruning these GCI candidates with an automatically determined amplitude threshold, found by iteratively removing from the full signal the inlier samples within a fraction of its standard deviation until converged, yields a 99.6% accurate detection system with a false alarm rate of 0.17%. This simpler algorithm, named Glottal Activity Detector For Laryngeal Input (GADFLI), outperforms the state-of-the-art SIGMA algorithm for GCI detection, which has a 94.2% detection rate, but a 5.46% false alarm rate. Performance metrics were computed over the entire APLAWD database, using an extensive, hand-verified markings database of 10,944 waveforms. A related proposed algorithm, QuickGCI, also makes use of Hilbert phase discontinuities, and does not require a thresholding post-processing step for GCI selection. Its performance is nearly as good as GADFLI. Both proposed algorithms operate using the electroglottographic signal or acoustic speech signal.
Issue Date:2017-04-07
Type:Thesis
URI:http://hdl.handle.net/2142/97304
Rights Information:Copyright 2017 Roger D. Serwy
Date Available in IDEALS:2017-08-10
Date Deposited:2017-05


This item appears in the following Collection(s)

Item Statistics