 Hi, my name is Brittany Jekyll and thank you for the opportunity to speak at VCCA 2023. Today I will be presenting our project, Mapping Measures of Vocal Reaction Time, Perceived Test Demand and Speech Recognition to understand the benefits of on-demand processing in hearing aids. Thank you to my co-authors who supported this work. We know that understanding speech and noise remains difficult for hearing aid users. One tool aiming to help users in these kinds of challenging listening environments is a hearing aid feature called on-demand processing, which can apply strong, rapid setting adjustments when listeners need it most. On-demand processing, when activated, prompts the hearing aid to classify the listening environment and then apply additional specialized setting changes specific to that environment. These classification and adaptation schemes were derived via machine learning, which was performed on a huge number of real-life sound recordings. The goal of on-demand processing is to improve user outcomes, like speech understanding, listening effort and sound quality, and a variety of challenging listening scenarios. Specifically, on-demand processing can apply changes to gain, noise reduction, microphone directionality and expansion settings, all in reference to the listener's environment and even in some iterations of this feature, their listening goal. This plot demonstrates how on-demand processing might alter the signals presented by the hearing aid when compared to default hearing aid settings. In this example, we see that gain is increased by 5 to 7 decibels and the frequency range is important for speech intelligibility. To understand the benefits of on-demand processing, we undertook the following study. We enrolled 21 experienced hearing aid users with sensory neural hearing losses and fit them with receiver in the canal devices with clinically appropriate coupling strategies. Fittings were verified via real-ear measurement. The average age of the group was 70 years with a range of 42 to 85 years. Participants were asked to repeat eye-triple-e sentences presented in a noisy restaurant background, either with default hearing aid settings or with on-demand processing enabled. The noise on average was 63 dBc and was presented from seven loudspeakers encircling the participant. The noise was recorded in a local restaurant and featured background talkers, music and brief intermittent noises like dishes rattling. The speech was presented from the loudspeaker directly in front of the participant at zero degrees and was presented at a level such that the participant achieved approximately 70% words correct in the default settings condition. We gathered data on three main outcomes. The first outcome was speech understanding scores or the number of keywords repeated correctly for each sentence. The second outcome was vocal reaction time or VRT. As participants completed the speech understanding task, their responses were voiced recorded. In an offline analysis, VRT was then measured as the time between the offset of each sentence stimulus and the onset of the participant's response. VRT is considered to be a behavioral listening effort measurement where faster participant responses indicate speedier processing and less cognitive resource consumption. The third outcome was a subjective listening effort measure, specifically a perceived mental task demand rating, which was measured using a scale adapted from the NASA task load index assessment tool. A rating was given for each listening condition. To analyze our results, we looked at percent change on each outcome from the default settings condition, that is, the baseline condition, to the experimental on-demand processing condition. All three outcome measures showed statistically significant changes in performance. VRT's decreased, that is, speeded up by 5.9%. Perceived task demand decreased by 22.2% and speech understanding increased by 12.5% with the activation of on-demand processing. Thus, participants' speech and effort outcomes were generally improved with on-demand processing. Here's a deeper dive into the results. We were interested in how the three outcome measures mapped onto one another. These data show how performance changed from the default condition to the on-demand processing condition in terms of percent change. In this grid, each circle represents one participant. Speech data are plotted along the X axis, with values greater than 0%, indicating that speech accuracy increased with on-demand processing. Most participants experienced an improvement in speech perception, falling into the upper right or lower right quadrants. VRT data are plotted along the Y axis, with values less than 0%, indicating that reaction time sped up with on-demand processing. This is indicative of fewer cognitive resources being consumed by the task. In other words, it is indicative of a reduction in listening effort. Many participants demonstrated this improvement, falling into the lower left or lower right quadrants. Circle color illustrates perceived mental demand rating. Blue circles indicate that the participant felt that mental demand decreased with on-demand processing. Yellow circles indicate the opposite. The participant felt that mental demand increased. Gray circles indicate that the participant perceived no change in mental demand across conditions. Interestingly, changes in perceived task demand did not neatly map onto changes in VRT. For example, the yellow circles are not consistently above 0% on the Y axis. Indeed, the correlation between these two outcomes was only weakly positively correlated. In addition, some participants, those falling into the upper right quadrant, did show improved speech outcomes with on-demand processing, but at a cost to cognitive resources. We wondered if demographic variables correlated with any of our outcomes. We found several correlations with perceived task demands specifically. With on-demand processing, older participants, participants with comparatively poorer hearing, and participants with higher perceived hearing handicap, as measured by the revised hearing handicap inventory, experienced the greatest decreases in mental task demand. No demographic variables correlated with changes in speech scores or changes in VRT. Overall, it's possible that changes in VRT, a behavioral effort measure, are less impacted by demographic factors than subjective measures of effort. To summarize, while on-demand processing in realistic environments may provide broad benefits for listening effort and speech understanding in hearing aid users, these results support using a multiplicity of measures when evaluating these outcomes to gain a fuller picture of each participant's listening experience. Thank you.