To test a simple method for improving consistency among raters for the perceptual evaluation of pathological voice quality by providing visible speech (spectrogram) as additional information because, to date, the interrater variability still limits the widespread clinical use of the best available rating system.
Experimental comparison between 2 different ways (with and without the addition of visible speech) of perceptual rating by trained professionals of recorded pathological voices. Furthermore, the correlation between acoustical (jitter, shimmer, and noise-harmonic ratio) and perceptual parameters was investigated in both rating conditions.
Six experts evaluated 70 recorded pathological voices using the GIRBAS (grade, instability, roughness, breathiness, asthenicity, and strain) scale in 2 separate sessions: first, conventionally, without visible speech as additional information, and several months later, with visible speech as additional information.
Main Outcome Measures
The κ interrater agreement and the correlation coefficient between GIRBAS scores and acoustic measures.
We found a significant effect of visible speech on the agreement between the raters. The interrater agreement according to κ statistics was significantly stronger with the addition of visible speech than without for rating grade, roughness, and breathiness. The correlation between acoustical and perceptual parameters showed no significant effect of visible speech.
The addition of visible speech to the perceptual evaluation of pathological voices is an interesting clinical asset to enhance its reliability. The addition of visible speech to the clinical setting is feasible, since affordable computer programs are currently available that can provide the spectrogram in quasi–real time while conversing with the patient. The acoustical analysis might be applied in addition to perceptual rating in a multidimensional approach to assess voice quality.