SCH: Shallow and Deep Personalization for Hearing Aids

根据最近的统计数据，美国约有4410万成年人患有听力损失。未经治疗的听力障碍会影响沟通，并可能导致社会孤立，抑郁，痴呆和生活质量下降。感音神经性听力损失和相关心理社会后果的主要干预措施是助听器（HA）放大。不幸的是，只有15-30%的人可以受益于HA使用它们。成功采用HAs的先决条件是有效的信号处理算法以及个性化方法，以配置其许多参数，从而改善语音理解、声音质量和用户的主观偏好。因此，本项目的重点是开发新的信号处理算法和配置方法，使听力损失患者能够满足其个性化的听力需求。本项目旨在开发两种方法，用于个性化HA，在个性化程度、用户找到满意配置所需的努力量以及处理不同严重程度听力损失的有效性方面进行不同的权衡。它将开发浅层个性化技术，用于配置现有HA信号处理管道的参数。这些方法通过根据听觉上下文使用不同的子带处理增益、压缩参数和降噪设置，提供了比现有技术的非处方HA更多的个性化选项。这些技术最适合轻度至中度感音神经性听力损失患者。我们还将开发深度个性化技术，用于训练和个性化使用深度神经网络放大声音的HA。这种方法的一个独特之处是使用脑电图信号与用户反馈相结合来驱动个性化过程。这些算法将使患有更严重听力损失或处于具有挑战性的听觉环境中的患者受益。该提案的智力价值是机器学习的新进展，这是使患者能够配置和有效使用其HA所必需的。该研究计划旨在帮助患者更多地参与听力保健，提高HA满意度，并丰富他们的社会交往。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Approximately 44.1 million adults in the US suffered from hearing loss, according to recent statistics. Untreated hearing impairment affects communication and can contribute to social isolation, depression, dementia, and reduced quality of life. The primary intervention for sensorineural hearing loss and related psychosocial consequences is hearing aid (HA) amplification. Unfortunately, only 15–30% of those who could benefit from HAs use them. A prerequisite for the successful adoption of HAs is effective signal processing algorithms coupled with personalization methods to configure their many parameters to improve speech understanding, sound quality, and users' subjective preferences. Therefore, this proposal focuses on developing new signal processing algorithms and configuration methods that empower people with hearing loss to meet their individualized hearing needs.This project aims to develop two approaches for personalizing HAs with different trade-offs in the degree of personalization, the amount of user effort required to find a satisfactory configuration, and their effectiveness in handling hearing losses of varying severity. It will develop shallow personalization techniques for configuring the parameters of existing HA signal-processing pipelines. These approaches provide more personalization options than state-of-the-art over-the-counter HAs by using different sub-band processing gains, compression parameters, and noise-reduction settings depending on the auditory context. These techniques are most suitable for patients with mild-to-moderate sensorineural hearing loss. We will also develop deep personalization techniques for training and personalizing HAs that use deep neural networks to amplify sounds. A unique aspect of this approach is using electroencephalogram signals combined with user feedback to drive the personalization process. These algorithms will benefit patients with more severe hearing loss or those in challenging auditory environments. The intellectual merit of this proposal is the new advancements in machine learning that are necessary to enable patients to configure and effectively use their HAs. The proposed research is anticipated to empower patients to become more involved in hearing care, improve HA satisfaction, and enrich their social interactions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.