Evaluating the Validity of Visual Analog Scales for Measuring Intelligibility in Spontaneous Speech

The purpose of this study was to investigate the relationship between orthographic transcription and visual analog scale (VAS) intelligibility ratings across read and spontaneous speech in speakers with Parkinson's disease (PD), speakers with multiple sclerosis (MS), and healthy controls.

Forty-four speakers (20 PD, 20 MS, four healthy controls) provided read speech samples from the Speech Intelligibility Test and spontaneous monologue samples. Speech samples were mixed with multitalker babble and presented to 131 crowdsourced naive listeners. Listeners were randomly assigned to orthographic transcription (n = 66) or VAS rating (n = 65) tasks. The relationship between measures was examined using linear regression and linear mixed-effects modeling.

Strong positive relationships between orthographic transcription and VAS were observed for both read speech (R² = .75) and monologue (R² = .72), with relationships remaining robust across all speaker groups (R² = .52-.76). Interactions between perceptual task and speech task or between perceptual task and speaker group were not significant. Read speech was approximately 4.3 percentage points more intelligible than monologue.

VAS and orthographic transcription yielded strongly associated intelligibility estimates across speech tasks and speaker groups, supporting VAS as an efficient alternative to orthographic transcription in diverse clinical populations and speech contexts. Clinicians should note that VAS does not capture qualitative error information and may require more listeners to achieve comparable reliability. Future work should examine speakers with more severe dysarthria and establish clinically meaningful change thresholds for spontaneous speech.

https://doi.org/10.23641/asha.33345129.