Thesis: Stationary and dynamic aerodynamic assessment of vocal hyperfunction using enhanced supraglottal and subglottal inverse filtering methods
Date
2018
Authors
Espinoza Catalan, Victor Manuel
Journal Title
Journal ISSN
Volume Title
Publisher
Universidad Técnica Federico Santa María
Abstract
Esta tesis describe los lineamientos, el diseño experimental y los resultadosiniciales de evaluaciones aerodinámicas estacionarias y dinámicas de lahiperfunción vocal. Este trabajo tiene como objetivo avanzar en la comprensióny evaluación clínica de la hiperfunción vocal, mediante el mejoramiento de losmétodos actuales de ltrado inverso para las señales de flujo oral y de aceleraciónde la piel del cuello, incorporando herramientas de análisis estadístico en estaestructura. Nuevos algoritmos de filtrado inverso, incluyendo uno basado ensegmentos de señal, son explorados y aplicados de manera automática paraestimar múltiples medidas aerodinámicas de la función vocal que luego sonutilizadas con métodos clínicos con el fi n de diferenciar la hiperfunción vocalrespecto a un comportamiento normal de la voz. Para lograr este objetivo, serealiza una evaluación clínica de la función vocal a través de grabacionesaerodinámicas, vibroacústicas y acústicas, en condiciones de laboratorio. Variosmétodos son explorados para mejorar la estimación de medidas aerodinámicas,incluyendo vocales sostenidas y habla continua. Una selección de técnicas de filtrado inverso fueron refi nadas para estimar el flujo de aire glotal en vocesnormales y patológicas, donde muchas de ellas constituyen las condiciones másdesafi antes para el filtrado inverso, como son las voces femeninas en hablacontinua. La hipótesis latente considera que los métodos de fi ltrado inversosupraglotal pueden ser perfeccionados en condiciones difíciles al limitar el anchode banda de la señal al primer formante. En este contexto, múltiples métricas deerror son propuestas y estudiadas con el fin de estimar la calidad del filtradoinverso, las cuales son fundamentadas en el comportamiento esperado de la señal filtrada y el contraste con simulaciones numéricas del impulso glotal en estadoestacionario. Agregado a lo anterior, el esquema de fi ltrado inverso basado en laimpedancia subglotal (del inglés impedance-based inverse fi ltering, IBIF) esexplorada en el contexto de habla continua. Las técnica de fi ltrado inversosupraglotal automática fue implementada en un enfoque basado en segmentos deseñal con el fi n de evaluar las incertezas de los parámetros del modelo IBIF y lasmedidas aerodinámicas derivadas de una señal de aceleración de la piel delcuello. Se utilizó un análisis estadístico robusto para reducir la influencia deeventos esporádicos y valores atípicos para la estimación objetiva de estosparámetros. Tomando en cuenta las incertezas de la señal del flujo de aire glotaly de aceleración de la piel del cuello, es posible mejorar el fi ltrado inversosubglotal, y en este sentido, se discuten las direcciones futuras de este enfoquepara avanzar en el monitoreo ambulatorio aerodinámico de la función vocal.Finalmente, se obtuvo un conjunto de datos actualizado de medidasaerodinámicas con los métodos supraglotal y subglotal propuestos, junto con unenfoque robusto para diferenciar pacientes hiperfuncionales de controlespareados, utilizando modelos estadísticos multivariados basados en grabacionesdel flujo de aire oral y de aceleración del cuello.
This thesis describes the guidelines, experimental design, and initial results forstationary and dynamic aerodynamic assessments of vocal hyperfunction. Thiswork aims to improve the understanding and clinical assessment of vocalhyperfunction by advancing current methods for the inverse ltering of both oralair flow and neck skin acceleration signals and by incorporating statisticalanalysis tools in this framework. New algorithms to perform inverse lteringincluding a frame-based approach are explored and applied in an automaticframework to estimate multiple aerodynamic measures of vocal function, andlater utilized with enhanced clinical methods to differentiate vocal hyperfunctionfrom normal vocal behavior. To achieve this goal, a clinical assessment isperformed through aerodynamic, vibroacoustic, and acoustic recordings of vocalfunction in laboratory conditions. Various methods to improve the estimation ofaerodynamic measures using different vocal gestures are explored, includingsustained vowels and continuous speech. Selected inverse ltering techniqueswere enhanced to estimate glottal air flow in normal and pathological voices,wherein many of them constitute the most challenging conditions for inverse ltering, namely female voices during running speech. The underlying hypothesisconsiders that supraglottal inverse ltering methods can be enhanced underchallenging conditions by limiting the signal bandwidth down to the firstformant. For estimating inverse ltering quality, several error metrics that arebased on the signal behavior and the contrast with glottal waveform simulations,are proposed and studied. In addition, the subglottal impedance-based inverse ltering (IBIF) scheme is explored in the context of running speech. The automated supraglottal inverse ltering technique was implemented in aframe-based framework to evaluate the uncertainties of the IBIF modelparameters and the derived aerodynamic measures from a neck skin accelerationsignal. Robust statistical analysis was used to reduce the influence of sporadicevents and outliers for the objective parameter estimation. Accounting for theuncertainties of glottal air flow from the neck skin acceleration signal would allowfor improving the subglottal inverse fi ltering, and future directions for thisapproach are discussed to advance the aerodynamic ambulatory monitoring ofvocal function. Finally, an updated dataset of aerodynamic measures was derivedwith the proposed supraglottal and subglottal methods along with a robustapproach to differentiate hyperfunctional patients with paired matched controls,using multivariate statistical models based on oral airflow and neck accelerationrecordings.
This thesis describes the guidelines, experimental design, and initial results forstationary and dynamic aerodynamic assessments of vocal hyperfunction. Thiswork aims to improve the understanding and clinical assessment of vocalhyperfunction by advancing current methods for the inverse ltering of both oralair flow and neck skin acceleration signals and by incorporating statisticalanalysis tools in this framework. New algorithms to perform inverse lteringincluding a frame-based approach are explored and applied in an automaticframework to estimate multiple aerodynamic measures of vocal function, andlater utilized with enhanced clinical methods to differentiate vocal hyperfunctionfrom normal vocal behavior. To achieve this goal, a clinical assessment isperformed through aerodynamic, vibroacoustic, and acoustic recordings of vocalfunction in laboratory conditions. Various methods to improve the estimation ofaerodynamic measures using different vocal gestures are explored, includingsustained vowels and continuous speech. Selected inverse ltering techniqueswere enhanced to estimate glottal air flow in normal and pathological voices,wherein many of them constitute the most challenging conditions for inverse ltering, namely female voices during running speech. The underlying hypothesisconsiders that supraglottal inverse ltering methods can be enhanced underchallenging conditions by limiting the signal bandwidth down to the firstformant. For estimating inverse ltering quality, several error metrics that arebased on the signal behavior and the contrast with glottal waveform simulations,are proposed and studied. In addition, the subglottal impedance-based inverse ltering (IBIF) scheme is explored in the context of running speech. The automated supraglottal inverse ltering technique was implemented in aframe-based framework to evaluate the uncertainties of the IBIF modelparameters and the derived aerodynamic measures from a neck skin accelerationsignal. Robust statistical analysis was used to reduce the influence of sporadicevents and outliers for the objective parameter estimation. Accounting for theuncertainties of glottal air flow from the neck skin acceleration signal would allowfor improving the subglottal inverse fi ltering, and future directions for thisapproach are discussed to advance the aerodynamic ambulatory monitoring ofvocal function. Finally, an updated dataset of aerodynamic measures was derivedwith the proposed supraglottal and subglottal methods along with a robustapproach to differentiate hyperfunctional patients with paired matched controls,using multivariate statistical models based on oral airflow and neck accelerationrecordings.
Description
Keywords
INVERSE FILTERING , VOCAL HYPERFUNCTION , VOICE ASSESSMENT
