Automatic Evaluation of Voice Quality Using Text-Based Laryngograph Measurements and Prosodic Analysis

摘要: Due to low intra- and interrater reliability, perceptual voice evaluation should be supported by objective, automatic methods. In this study, text-based, computer-aided prosodic analysis measurements of connected speech were combined in order model the German Roughness-Breathiness-Hoarseness (RBH) scheme. 58 samples (43 women 15 men; 48.7 ± 17.8 years) containing version text “The North Wind Sun” evaluated perceptually 19 therapy students according RBH scale. For human-machine correlation, Support Vector Regression with vocal fold cycle irregularities (CFx) closed phases vibration (CQx) Laryngograph 33 features from a module used listeners' ratings. The best results for roughness obtained combination six CFx (r = 0.71, ρ 0.57). These correlations approximately same as agreement among human raters 0.65, 0.61). CQx was one substantial hoarseness model. breathiness, substantially lower. Nevertheless, method can serve basis meaningful objective support analysis.