Automatic glottal segmentation using local-based active contours and application to glottovibrography

• Published in: Speech communication Vol. 54; no. 5; pp. 641 - 654
• Main Authors: Karakozoglou, Sevasti-Zoi, Henrich, Nathalie, d’Alessandro, Christophe, Stylianou, Yannis
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.06.2012; Elsevier : North-Holland
• Subjects: Active contours; Assessments; Computer Aided Engineering; Computer Science; Contours; Electroglottography; Engineering Sciences; Failure; Glottovibrogram; High-speed videoendoscopy; Humanities and Social Sciences; Library and information sciences; Linguistics; Representation; Representations; Segmentation; Signal and Image processing; Speech Physiology; Vibration; Vocal Folds; Vocal-fold vibration; Representation; Electroglottography; High-speed videoendoscopy; Vocal-fold vibration; Active contours; Glottovibrogram
• ISSN: 0167-6393; 1872-7182
• DOI: 10.1016/j.specom.2011.07.010

► Local-based active contour framework is applied so as to improve glottal segmentation. ► Segmentation’s discriminative power is investigated on 60 high-speed video recordings. ► Manual verification resulted in less than 1% difference on average glottal area. ► Glottovibrogram depicts time-varying distance of the vocal-fold edges. ► Glottovibrogram is effective while reducing error from glottal-axis detection. The use of high-speed videoendoscopy (HSV) for the assessment of vocal-fold vibrations dictates the development of efficient techniques for glottal image segmentation. We present a new glottal segmentation method using a local-based active contour framework. The use of local-based features and the exploitation of the vibratory pattern allows for dealing effectively with image noise and cases where the glottal area consists of multiple regions. A scheme for precise glottis localization is introduced, which facilitates the segmentation procedure. The method has been tested on a database of 60 HSV recordings. Comparisons with manual verification resulted in less than 1% difference on the average glottal area. These errors mainly come from detection failure in the posterior or anterior parts of the glottal area. Comparisons with automatic threshold-based glottal detection point out the necessity of complete frameworks for automatic detection. The glottovibrogram (GVG), a representation of glottal vibration is also presented. This easily readable representation depicts the time-varying distance of the vocal-fold edges.