Automated Creak Identifies Laryngeal Dystonia During Conversational Speech.

Abstract

Creak is an acoustic feature that serves an umbrella term for irregularities in fundamental frequency. This study evaluated whether automated creak distinguished speakers with adductor laryngeal dystonia (AdLD), primary muscle tension dysphonia (pMTD), and those without voice disorders during conversational speech. Two stimulus types (reading passage and conversational speech) were produced by 50 speakers with AdLD, 50 speakers with pMTD, and 50 control speakers. An automated creak detector was used to calculate the percentage (%) of creak in both stimuli. The effects of group, stimulus type, and their interaction on % creak were assessed. Receiver operating characteristic curve analyses were conducted and the area under the curve (AUC) was used to evaluate the effectiveness of the creak detector in distinguishing among groups. We found a statistically significant effect of group on % creak, but no effect of stimulus type on % creak, suggesting that creak distinguished speakers with AdLD, pMTD, and controls during both a reading passage and conversational speech. AUC values indicated acceptable discriminative ability in distinguishing AdLD speakers from both pMTD speakers and controls across both stimuli types. Automated creak demonstrates promise as a discriminative feature for identifying differences between speakers with AdLD from pMTD and controls in both a reading passage and conversational speech.