Comparing direct and indirect glottal flow waveform measurements in excised canine larynges and synthetic vocal fold models

Abstract

Despite being considered the gold standard technique for delineating the flow glottogram during phonation, the inverse filtering technique has never been validated against actual volume flow measurements. This signal-processingtechnique uses the speech signal with vocal tract estimates to calculate the glottal flow waveform. The lack of validation stems from the difficulty of obtaining in-vivo flow measurements at the glottal exit. The current study aims to assess the accuracy of the inverse filtering technique by comparing its flow glottograms with direct, simultaneous volume flow measurement. A vocal tract with varied constrictions that mimic its shape according to specific vowels was connected to either an excised canine larynx or a synthetic vocal fold model. Direct volume flow measurements at the glottal exit were done using tomographic particle image velocimetry. Indirect measurements were obtained by fitting the known Rothenberg mask to the vocal tract and estimating the glottal flow waveform using the Glottal Enterprise system. Preliminary results indicate that the accuracy of the inverse filtering technique relies on the degree of vocal tract constriction. The flow glottogram aligns well with front, lax, and unrounded lip vowels, but its accuracy diminishes with tense and round vowels, particularly noticeable with the excised canine larynx.