Individualized Assessment and Prediction of Speech-Recognition Performance In Adults with Age-related Hearing Loss

年龄相关性听力损失成人的语音识别表现的个体化评估和预测

• 批准号: 10456939
• 负责人: Yi Shen
• 金额: $ 29.96万
• 依托单位: UNIVERSITY OF WASHINGTON
• 依托单位国家: 美国
• 财政年份: 2020
• 资助国家: 美国
• 起止时间: 2020-08-01 至 2024-07-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/10456939
• 关键词: Acoustics; Address; Adopted; Adult; Age; Algorithms; Americas; Assessment tool; Auditory; Clinical; Cognitive; Cognitive deficits; Communities; Complex; Cues; Custom; Dependence; Diagnostic Procedure; Diagnostic tests; Elderly; Environment; Face; Frequencies; Future; Goals; Guidelines; Health Personnel; Healthcare; Hearing; Hearing Aids; Hour; Individual; Individual Differences; Laboratory Finding; Laboratory Research; Link; Lipreading; Measures; Methods; Modeling; Noise; Patients; Performance; Persons; Presbycusis; Procedures; Research; Role; Services; Short-Term Memory; Signal Transduction; Source; Speech; Speech Audiometry; Speech Intelligibility; Standard Model; Standardization; Testing; Time; Translating; Visual; Weight; base; clinical practice; cognitive skill; diagnostic tool; experience; hearing impairment; improved; indexing; individual patient; predictive modeling; programs; speech recognition; temporal measurement; tool

The main complaint from listeners with age-related hearing loss is the difficulty in understanding speech in noisy environments. The sources of the speech-understanding difficulty involve auditory and cognitive factors and vary from one listener to another. Developing models of speech intelligibility that can account for these factors is necessary for predicting expected speech-recognition performance with or without the use of a hearing aid. Moreover, if such models can be efficiently fitted to individual hearing-aid users, then the amplification profile in the hearing aid can be customized to the users' specific needs. However, such efficient diagnostic procedures for fitting models of speech-intelligibility are not yet available. The proposed research program will address this issue directly. The long-term goal of the program is to establish an efficient diagnostic test to enable individualized hearing-aid fitting. As a first step toward this goal, a Bayesian adaptive procedure for fitting a widely-adopted model of speech intelligibility, i.e. the Speech Intelligibility Index (ANSI S3.5-1997), to individual listeners will be examined in detail. The Bayesian adaptive procedure uses a speech recognition task, similar to clinical speech audiometry, and it allows the estimation of the model parameters for the Speech Intelligibility Index using as few as 75 test sentences (approximately 12 minutes of testing time). These estimated parameters indicate (1) how much acoustic cues in various frequency bands are being used for speech recognition, (2) the signal-to-noise ratio required to reach a performance level of 50% correct recognition, and (3) the listener's benefits from contextual cues in speech. The relationship between these model parameters to listener's auditory and cognitive skills will be systematically evaluated using a group of older adults with diverse age and hearing status. The parameters will also be studied under two common listening conditions: speech recognition in temporally fluctuating backgrounds, and speech recognition with visual cues (i.e. the display of the talker's face). The dependencies of the model parameters for these commonly occurring listening conditions will be investigated. Additionally, the estimated model using the Bayesian adaptive procedure will be used to predict speech-recognition performance under aided and unaided conditions. Whether the individualized Speech Intelligibility Index provides additional predictive power compared to the standard model will be evaluated. The estimated model will also be used to optimize the amplification profiles for individual hearing-impaired listeners, and its relationship to the listeners' preferred amplification profiles will be examined. Upon the completion of the proposed research program, a model will be established to provide comprehensive profiling of listeners' speech-recognition performance. Moreover, a set of tools will be made available to efficiently fit the model to individual listeners and to optimize the amplification profile according to the estimated model parameters.

患有与年龄有关的听力损失的听众的主要抱怨是难以理解用英语表达的言语 嘈杂的环境。言语理解困难的根源涉及听觉和认知因素 并且因听众而异。开发可以解释这些问题的语音清晰度模型 因素对于预测预期的语音识别性能是必要的，无论是否使用 助听器。此外，如果此类模型可以有效地适合个人助听器用户，那么 助听器中的放大曲线可以根据用户的具体需求进行定制。然而，如此高效 用于拟合语音清晰度模型的诊断程序尚不可用。拟议的研究 程序将直接解决这个问题。该计划的长期目标是建立有效的诊断方法 测试以实现个性化助听器验配。作为实现这一目标的第一步，贝叶斯自适应过程 用于拟合广泛采用的语音清晰度模型，即语音清晰度指数 (ANSI S3.5-1997)， 将详细审查个别听众。贝叶斯自适应过程使用语音识别 任务，类似于临床语音测听，它允许估计语音的模型参数 可理解性指数仅使用 75 个测试句子（测试时间大约 12 分钟）。这些 估计参数表明 (1) 不同频段中的声学提示有多少被用于 语音识别，(2) 达到 50% 正确率的性能水平所需的信噪比 识别，以及（3）听者从言语中的上下文线索中获益。这些之间的关系 将使用一组系统地评估听众听觉和认知技能的模型参数 不同年龄和听力状况的老年人。这些参数也将在两种常见的情况下进行研究 听力条件：时间波动背景下的语音识别，以及 视觉提示（即说话者脸部的显示）。这些模型参数的依赖性 将调查常见的聆听条件。此外，估计模型使用 贝叶斯自适应过程将用于预测辅助和无辅助下的语音识别性能 状况。个性化语音清晰度指数是否提供额外的预测能力 将评估与标准模型的比较。估计模型也将用于优化 个别听障听众的放大曲线及其与听众偏好的关系 将检查扩增曲线。拟议的研究计划完成后，将建立一个模型 旨在提供听众语音识别性能的全面分析。此外，还有一套 将提供多种工具来有效地将模型适合各个听众并优化扩音 根据估计的模型参数进行剖面。