Word Recognition Clinical Testing of Personalized Deep Reinforcement Learning Compression

Current hearing aid fittings are carried out using prescriptive compression settings. These settings are derived from group averages but do not account for individual differences or preferences. In a previous work, we developed a human-in-the-loop deep reinforcement learning compression approach to set the compression ratios across a number of frequency bands. These compression ratios were compared to those of the widely used and accepted DSL-v5 hearing aid prescription to determine if incorporating user preference impacted word recognition performance. A pilot clinical study of this comparison for four participants is reported in this paper. The clinical testing results obtained strongly support the hypothesis that the personalized compression settings do not negatively impact word recognition or audibility compared to the prescriptive compression settings. The ability to provide a personalized amplification strategy with no degradation in hearing performance would be of benefit in hearing aids or other assistive listening technologies.