Towards Precision Assessment of Dysphonic Speech: From Vocal Fold Physiology to Perception

点击翻译按钮获取中文摘要

Project Summary/Abstract The ways in which vocal fold (VF) vibratory behavior and resulting voice quality (VQ) perception differ across vowel categories and the co-articulatory variations of connected speech are unknown and largely unexplored. Furthermore, phonatory adjustments during connected speech (possible variations in voicing onsets and offsets and articulatory transitions) may provide important clinical information that can guide diagnostic protocols, may correspond closely with perceived handicap, and may represent functionally-relevant treatment targets. The overall goal of the proposed research is to discover and quantify physiological mechanisms underlying normal and abnormal VF behavior and establish their relationships to VQ perception in connected speech. This innovative proposal leverages the expertise of a multidisciplinary team and develops a comprehensive framework linking vocal physiology and perception with improved measurement approaches, methods, and analyses that account for the effects of co-articulation in connected speech. Abnormalities in pre-, post-, and peri-phonatory vibratory behavior are physiological hallmarks of voice disorders. Aim 1 will leverage precise, automated physiological measures of phonatory onset (pre-), offset (post-), and variation in VF phase asymmetry (peri-) to characterize VF vibratory behavior in uniform vowel- consonant-vowel (VCV) utterances with a controlled phonetic context. Aim 2 will establish relationship between these physiological measures and dimension-specific VQ perception in VCV utterances, using ratio- level matching tasks with physical units (e.g., dB) and biologically inspired computational models grounded in psychoacoustics and auditory-perception. These evaluative methods overcome technical and methodological limitations of conventional perceptual and acoustic methods. Aim 3 will evaluate and validate the three physiological measures in connected speech and discover new physiological signatures currently unknown through the use of novel and powerful machine-learning models that include as inputs physiological measures derived from high-speed videoendoscopy. Aim 4 will use automated, efficient dimension-specific computational models to evaluate VQ in connected speech and to discover physiological signatures that are related to VQ perception through machine learning. The unique combination of machine learning with computational models of VQ perception that are specific to VQ dimensions, rather than just overall severity, can effectively deal with the massive data associated with connected speech and high-speed videoendoscopy. Knowledge gained from this pre-translational research has the potential to improve our understanding of voice pathology and to substantially advance functional assessment and treatment outcomes for millions of people with hypo- and hyper-adductory voice disorders.