Comparing Two Methods for Measuring Speech Intelligibility in Two Different Environments

Abstract

Purpose: Speech intelligibility tests evaluate the proportion of a speech signal produced by a speaker that is understood by a listener. The touchstone for measuring speech intelligibility is orthographic transcription (TRA). This study evaluated the validity, reliability and efficiency of a subjective measure, visual-analog scaling (VAS) compared to TRA in two listening environments, in-lab setting (ILS) and outside-lab setting (OLS). This was accomplished using a small group of speakers with increasing levels of speech severity.

Methods: Sixty-five participants listened to recordings of 13 unique sentences from each of the six speakers. Each participant transcribed the sentence of 3 speakers and rated the sentences of 3 different speakers with VAS. Approximately half the participants were in the ILS environment and the other half in the OLS environment. Stimulus presentation and recording of responses was done with a specially designed browser-based application.

Results: Good agreement was observed in the aggregated TRA condition in the ILS and OLS environments. Although there was reasonable agreement between the aggregated TRA and VAS scores in both environments, there was much greater variability between participants using VAS. Administration of VAS was nearly 4 times faster than TRA.

Conclusion: Analysis of this small group of speakers suggests further study is required to determine if VAS may be a useful clinical tool in lieu of transcription. Internet-based transcription of disordered speech distributed across clinicians remains plausible.