Uncertainty of glottal airflow estimation during continuous speech using impedance-based inverse filtering of the neck-surface acceleration signal

Abstract

The aim of this work is to determine the uncertainty of non-invasive glottal aerodynamic measures that are obtained using subglottal impedance-based inverse filtering (IBIF) of the signal from a neck-placed accelerometer during continuous speech. Currently, we are studying the vocal behavior of individuals with typical voices and voice disorders by analyzing weeklong recordings using a smartphone-based ambulatory voice monitor. We extend on previously reported analyses of sustained vowel production using subglottal IBIF and move toward continuous speech applications where IBIF parameters are estimated in a frame-based approach. Selected voiced frames of both oral-airflow (baseline) and acceleration signal from the Rainbow Passage are used to build a probabilistic model of IBIF parameters to run multiple random realizations of the inverse-filtered neck-surface acceleration signal. Confidence intervals are estimated for both the glottal waveform and derived features. The probabilistic model is tested using data from patients with vocal hyperfunction and matched-control subjects with normal voices at a comfortable pitch and loudness in an acoustically treated sound booth. Results show that model parameters follow approximately normal distributions, and the confidence intervals for the estimates of glottal aerodynamic measures were <10%, which is in close agreement with previously reported IBIF performance using sustained vowels.