The “hot potato voice” (HPV), characterized by a muffled and thick vocal quality, is a critical indicator of severe upper airway obstruction stemming from conditions like acute epiglottitis and peritonsillitis. Recognizing the urgency associated with this vocal symptom, a novel diagnostic method leveraging the acoustic properties of HPV has been developed. This method employs a physical, articulatory speech synthesis model to analyze HPV’s acoustic characteristics and subsequently utilizes a support vector machine (SVM) for computerized recognition.
This research, employing a quasi-experimental design, meticulously examined alterations in the voice’s spectral envelope caused by upper airway obstructions. Using a hybrid time-frequency model of articulatory speech synthesis, the study pinpointed variations in formant structure and critical vocal tract area function thresholds that are hallmarks of HPV. A substantial dataset of 2,200 synthetic voice samples, generated via an articulatory synthesizer, was used to train SVMs. The efficacy of these SVMs was then rigorously tested using real patient voice datasets to classify voice pathologies.
The findings revealed that during the phonation of the Japanese vowel /e/, a notable shift occurs: the second formant’s frequency diminishes and merges with the first formant as the oropharynx area function constricts. Furthermore, changes in higher-order formants were observed to vary depending on the location of the airway constriction. Remarkably, the SVM classifier, trained on synthetic data, achieved a peak accuracy of 88.3% in diagnosing upper airway obstructions.
In conclusion, the study definitively demonstrates that HPV arising from upper airway obstruction possesses a distinctly identifiable spectral envelope. Capitalizing on this unique vocal signature, an SVM classifier, trained with meticulously synthesized data, can effectively and accurately diagnose upper airway obstructions. This innovative approach offers a promising avenue for enhancing diagnostic precision and timeliness in cases of upper airway compromise.
Level of evidence: 2c
Laryngoscope, 129:1301-1307, 2019.
