Predicting voice quality of deaf speakers on the basis of glottal characteristics

Twenty profoundly deaf and 5 normal-hearing subjects produced 225 sustained vowels/a, i, u/at different pitch levels and 75 phonetically balanced sentences, while vocal fold vibration was recorded through an electrolaryngograph (ELG). The utterances recorded on audio tape were judged by 10 experienced listeners on general voice quality, breathiness, hoarseness, and laryngeal strain on a 5-point scale. Seven parameters describing time-domain characteristics of the ELG-signal were extracted online by a special purpose computer system. Measurements were made over 500 consecutive vibratory cycles (10-cycle window), yielding a mean and standard deviation for each parameter per utterance. All data were submitted to analysis of variance and multiple regression analyses. Multiple correlations between glottal parameters and judged voice deviations varied between .46 and .70 indicating that overall prediction cannot reliably be based on these parameters, although severe cases of deaf voice deviations may be detectable.