Speech-in-Noise Testing in Virtual Reality

The potential of virtual reality (VR) in supporting hearing research and audiological care has long been recognized. While allowing the creation of experimental settings that closely resemble real-life scenarios and potentially leading to more ecologically valid results, VR could also support the current need for automated or remote assessment of auditory processing abilities in clinical settings. Understanding speech in competing noise is the most common complaint of patients with hearing difficulties, and the need to develop tools that can simplify speech-in-noise testing by reducing the time and resources required while improving the ecological validity of current assessment procedures is an area of great research interest. However, the use of VR for speech-in-noise testing has not yet been widely adopted because it is still unclear whether subjects respond to virtual stimuli the same way as they would in real-life settings. Using headphone-based binaural presentation, delivering visuals through head-mounted displays (HMDs), and using unsupervised (self-testing or remote) procedures are some aspects of virtualization that could potentially affect speech-in-noise measures, and the extent of this potential impact remains unclear. Before virtualization can be considered feasible, its effects on behavioral psychoacoustic measures must be understood. Thus, the ability to reproduce results from typical laboratory and clinical settings in VR environments is a major topic of current research. In this study, we sought to answer whether it is possible to reproduce results from a standard speech-in-noise test using state-of-the-art technology and commercially available VR peripherals. To this end, we compared the results of a well-established speech-in-noise test conducted in a conventional loudspeaker-based laboratory setting with those obtained in three different virtual environments. In each environment, we introduced one aspect of virtualization, i.e., virtual audio presentation in the first environment, HMD-based visuals with a visual anchor representing the target speaker in the second, and an alternative feedback- and scoring method allowing unsupervised testing in the last. Our results indicate that the speech-in-noise measures from the loudspeaker-based measurement and those from the virtual scenes were all statistically identical, suggesting that conducting speech-in-noise testing in state-of-the-art VR environments may be feasible even without experimenter supervision.