Principal differential analysis for detection of bilabial closure gestures from articulatory data. (March 2016)

• Record Type: Journal Article
• Authors: Sattar, Farook; Rudzicz, Frank
• Is Part Of: Computer speech & language. Volume 36(2016)
• Page Start: 294
• Page End: 306
• Publication Date: 2016-03
• Subjects: Bilabial closure gesture -- Principal differential analysis (PDA) -- Articulatory data -- Tract variables -- Classification; Speech processing systems -- Periodicals; Automatic speech recognition -- Periodicals; Computers -- Periodicals; Linguistics -- Periodicals; Speech-Language Pathology -- Periodicals; Traitement automatique de la parole -- Périodiques; Reconnaissance automatique de la parole -- Périodiques; Automatic speech recognition; Speech processing systems; Electronic journals; Periodicals; 006.454
• DOI: 10.1016/j.csl.2015.07.002
• Languages: English
• ISSNs: 0885-2308

Abstract

Abstract : Highlights: Electromagnetic articulographic data represented in task dynamics. Dynamics of lip movement represented by linear differential equations. Linear differential equations learned by principal differential analysis (PDA). Bilabial/non-bilabial classification using parameters of PDA with high accuracy. Abstract: In this paper, a new statistical method for detecting bilabial closure gestures is proposed based on articulatory data. This can be surprisingly challenging, since mere proximity of the lips does not imply their involvement in a directed phonological goal. This segment-based bilabial closure detection scheme uses principal differential analysis (PDA) to extract articulatory gestures. The dynamic patterns of the tract variables (TVs) lip aperture, lip protrusion, and their derivatives, are captured with PDA and used to detect and quantify bilabial closure gestures. The proposed feature sets, which are optimized using sequential forward floating selection (SFFS), are combined and used in binary classification. Experimental results using the articulatory database MOCHA-TIMIT show the effectiveness of the proposed method demonstrating promising performance in terms of high classification accuracy (95%), sensitivity (95%), and specificity (95%).