Modeling speech intelligibility in competing backgrounds by the hearing-impaired

描述（由申请人提供）：听力受损（HI）听众在嘈杂的环境中处于非常不利的地位。不到30%的助听器用户对其设备在噪音方面的表现感到满意，尽管在较低的不利条件下满意度要高得多（Kochkin, 2000）。这些困难在波动的背景下更加复杂。虽然正常听力（NH）的人能够利用掩蔽器水平的瞬间下降来获得相对于静止噪声的显著（5-10 dB）波动掩蔽器收益（FMB），但HI听众似乎无法做到这一点（Festen和Plomp， 1990）。提出的研究旨在阐明导致HI侦听器中FMB减少的机制，为针对这些特定机制的信号处理算法的发展奠定基础。为了解释HI听者有限的FMB，过去的研究主要集中在可听性降低，频谱或时间分辨率降低，或目标-源分离线索有限。本提案探讨了这样一个假设，即HI和NH听者在测试信噪比（SNR）时的差异导致了FMB的差异，而一些波动的掩蔽物可能是HI听者FMB减少的主要原因。如果假设调制掩模的有效语音动态范围比先前估计的静止噪声更窄，则可以通过现有的语音可理解度模型（Rhebergen等人，2006）预测信噪比相关的FMB。实验1将直接测量该有效动态范围，以改进模型并提高跨信噪比的FMB预测精度。初步结果表明，在控制信噪比差异后，HI和模拟HI （HISIM）听众在某些波动掩码下表现出与NH听众相似的FMB。实验2和3将根据听者组之间信噪比和可听性均衡后FMB仍在减少的程度来区分波动掩蔽器类型。这一提议有可能对提高噪声环境下语音清晰度的研究工作产生实质性影响。对于波动掩蔽器，其中信噪比效应不能解释FMB差异的全部幅度，本文开发的方法可以控制信噪比差异，以更直接地追求负责限制FMB的损伤相关扭曲。对于波动掩蔽器，在信噪比差异得到控制后，HI听众和NH听众一样从掩蔽器波动中受益，未来的工作将寻求(a)提高目标语音的可听性，例如通过快速压缩，可以在波动背景中选择性地放大低水平目标；(b)确定噪声中限制可听性的因素，一般来说，研究结果也应扩展到波动掩蔽器。此外，改进的语音清晰度模型有潜力通过以下途径改善HI听者的临床管理：(a)在信号处理算法的开发中使用，以提高语音清晰度；(b)在临床应用中识别可能遭受超出可听性的扭曲的个体，这些扭曲限制了波动背景下的语音清晰度。

DESCRIPTION (provided by applicant): Hearing-impaired (HI) listeners are severely disadvantaged in noisy situations. Fewer than 30% of hearing-aid users are satisfied with the performance of their devices in noise, even though satisfaction levels are considerably higher for less adverse conditions (Kochkin, 2000). These difficulties are compounded in fluctuating backgrounds. While normal-hearing (NH) individuals are able to take advantage of momentary dips in the level of a masker to receive a significant (5-10 dB) fluctuating-masker benefit (FMB) to speech intelligibility relative to stationary noise, HI listeners seem unable to do so (Festen and Plomp, 1990). The proposed research aims to elucidate the mechanisms responsible for the reduced FMB in HI listeners, setting the stage for the development of signal processing algorithms to target these specific mechanisms. Attempting to explain the limited FMB in HI listeners, past studies have focused on reduced audibility, reduced spectral or temporal resolution, or limited cues for target-source separation. This proposal explores the hypothesis that differences in the signal-to-noise ratio (SNR) at which HI and NH listeners are tested contribute to FMB differences, and for some fluctuating maskers may account for most of the reduction in FMB for HI listeners. An SNR-dependent FMB is predicted by an existing model of speech intelligibility (Rhebergen et al., 2006), if the effective speech dynamic range is assumed to be narrower for modulated maskers than previously estimated for stationary noise. Experiment 1 will directly measure this effective dynamic range to refine the model and improve the accuracy of FMB predictions across SNRs. Preliminary results indicate that after SNR differences are controlled, HI and simulated HI (HISIM) listeners show a similar FMB to NH listeners for certain fluctuating maskers. Experiments 2 and 3 will differentiate fluctuating-maskers types based on the extent to which the FMB is still reduced after SNR and audibility are equalized between listener groups. This proposal has the potential to substantially impact research efforts to improve speech intelligibility in noise. For fluctuating maskers where SNR effects do not account for the full magnitude of FMB differences, the methods developed here could control SNR differences to more directly pursue impairment-related distortions responsible for limiting FMB. For fluctuating maskers where HI listeners are shown to benefit from masker fluctuations as much as NH listeners after SNR differences are controlled, future work would seek to (a) improve target speech audibility, e.g. via fast compression, which could selectively amplify a low-level target in a fluctuating background and (b) identify factors limiting intelligibility in noise, generally, with the idea that the findings should also extend to fluctuating maskers. Furthermore, the refined speech intelligibility model has the potential to improve the clinical management of HI listeners via (a) its use in the development of signal processing algorithms to improve speech intelligibility and (b) its clinical application in identifying individuals likely to suffer distortions beyond audibility that limit speech intelligibility in fluctuating backgrounds. PUBLIC HEALTH RELEVANCE: Hearing-impaired listeners experience the most difficulty when trying to listen in noisy environments, particularly those environments with masking sounds that fluctuate in intensity, like interfering speech. This proposal seeks to understand and model the underlying causes of these particular difficulties. The knowledge gained and the computational model developed over the course of the project could significantly impact the direction of research and rehabilitation efforts aimed at alleviating the problems experienced by impaired listeners in noisy environments.