Improving intelligibility in noise for hearing-impaired listeners

提高听力障碍听众的噪音清晰度

• 批准号: 9759651
• 负责人: ERIC W HEALY
• 金额: $ 31.42万
• 依托单位: OHIO STATE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 2016
• 资助国家: 美国
• 起止时间: 2016-09-01 至 2021-08-31
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/9759651
• 关键词: Address; Algorithms; American; Attenuated; Cochlear Implants; Data; Decision Making; Ensure; Foundations; Frequencies; Goals; Hearing Aids; Hearing problem; Individual; Knowledge; Lifting; Masks; National Institute on Deafness and Other Communication Disorders; Noise; Predisposition; Process; Quality of life; Scheme; Seminal; Signal Transduction; Speech; Speech Intelligibility; Speech Perception; Techniques; Time; Weight; Work; attenuation; base; design; experimental study; hearing impairment; hearing loss treatment; improved; indexing; microphone; normal hearing; novel; preservation; sound; speech in noise; speech recognition

Project Summary The primary complaint of hearing-impaired (HI) listeners is poor speech understanding when background noise is present (see Dillon, 2012). This problem can therefore be considered the most significant for the estimated 37.5 million Americans with hearing loss (NIDCD, 2015). Accordingly, a solution to this problem has commonly been considered a “holy grail” of our field. One proposed solution involves a single-microphone algorithm to extract speech from background noise. This may be considered an ultimate goal, because it is the algorithm that performs the task that the listener cannot. But despite 50 years of effort by groups around the world, an algorithm capable of improving intelligibility, especially for HI listeners, has remained elusive. We have recently provided the first demonstration of an algorithm capable of improving intelligibly in noise for HI listeners (Healy et al., 2013b, 2014, 2015). Not only is this work seminal, but the intelligibility improvements are substantial. Prior to algorithm processing, most of our HI listeners were able to understand roughly 1 in every 3 words within noisy sentences, and some scores were as low as 0-10%. Following algorithm processing, intelligibility for many of our HI listeners improved to roughly 90%. The long-term goal of the currently proposed study is to advance our ability to remedy the speech-in-noise problem for HI listeners. The first aim establishes basic information essential to our understanding of speech recognition in noise. During this aim, we establish what we have termed “noise susceptibility” for each individual frequency region of speech. We argue here that current efforts confound noise susceptibility with speech band importance, so that noise susceptibility is not known. We then provide direct and immediate application of this knowledge through a correction factor to incorporate noise susceptibility into the Speech Intelligibility Index (ANSI, 1997). During the second and third aims, we provide translational significance by advancing our algorithm in fundamental and important ways. During Aim 2, we establish a novel advancement that maximizes speech information while minimizing noise. We accomplish this by incorporating our understanding of noise susceptibility and speech band importance into our algorithm. During Aim 3, we compare the intelligibility and sound quality resulting from three different foundational schemes for our algorithm. One scheme is novel and will be introduced here. It promises to offer the advantages of both schemes we have already implemented. Overall, the current study has the potential to transform our basic understanding of speech recognition in noise and improve the ANSI standard used to predict it. Further, the proposed study is translational and addresses the primary limitation of HI listeners. We address this highly significant issue by advancing our algorithm in important and fundamental ways, thus progressing closer to our ultimate goal of implementation into hearing aids and cochlear implants. The contributions described here have the potential to substantially impact quality of life for millions of Americans and transform our treatment of hearing loss.

项目摘要 听力受损者的主要抱怨是当背景噪声时言语理解能力差 存在（见狄龙，2012）。因此，这个问题可以被认为是最重要的估计， 37.5 100万美国人患有听力损失（NIDCD，2015）。因此，该问题的解决方案通常 被认为是我们领域的“圣杯”。一种提出的解决方案涉及单麦克风算法， 从背景噪声中提取语音。这可能被认为是一个最终目标，因为它是算法 执行监听器无法执行的任务。但是，尽管世界各地的团体努力了50年， 能够提高可懂度的算法，特别是对于HI收听者，仍然是难以捉摸的。我们有 最近提供了一个算法的第一个演示，能够改善噪音的HI 听众（Healy等人，2013 b，2014，2015）。这项工作不仅是开创性的， 都很重要在算法处理之前，我们的大多数HI听众能够理解大约1英寸的声音。 在嘈杂的句子中，每3个词，有些分数低至0- 10%。以下算法 处理，我们的许多HI听众的可懂度提高到大约90%。的长期目标 目前提出的研究是为了提高我们的能力，以补救语音噪声问题的HI听众。的 第一个目标是建立基本信息，这对我们理解噪声中的语音识别至关重要。期间 为了达到这个目的，我们为每个单独的频率区域建立了我们所谓的“噪声敏感性”， 演讲我们认为，目前的努力混淆了噪声敏感性与语音波段的重要性，所以， 噪声敏感性是未知的。然后，我们通过以下方式提供这些知识的直接和即时应用： 将噪声敏感度纳入语音清晰度指数（ANSI，1997）的校正系数。期间 第二个和第三个目标，我们通过在基础和基础上改进我们的算法来提供转化意义 重要的方式。在目标2中，我们建立了一个新的进步，最大化语音信息， 最小化噪音。我们通过结合我们对噪音敏感性和语音的理解来实现这一点 把重要性加入到我们的算法中在目标3中，我们比较了结果的可懂度和音质 我们的算法的三种不同的基本方案。其中一个方案是新颖的，将在这里介绍。 它承诺提供我们已经实施的两个计划的优点。总的来说，目前的研究 有可能改变我们对噪声中语音识别的基本理解，并提高ANSI 此外，拟议的研究是翻译和解决的主要限制， 嗨，听众。我们解决这个非常重要的问题，通过推进我们的算法在重要和基本的 我们的目标是在助听器和人工耳蜗植入物中实现这一最终目标。 这里所描述的贡献有可能对数百万人的生活质量产生重大影响。 美国人和改变我们对听力损失的治疗。