Application of Ideal Binary Masking to disordered speech

项目摘要/摘要 背景噪音的存在通常会对听者的理解能力产生重大的负面影响 演讲。不幸的是，在嘈杂的环境中破译语音对听者来说是司空见惯的。作为我们最近的 初步数据显示，当语音信号本身退化时，例如由于神经疾病 构音障碍的言语障碍，加上背景噪音只会进一步降低听者的 理解这篇演讲(Yoho&Borrie，2018)。一种非常有希望的技术被开发来修复大部分 这种噪声中语音的困难被称为理想二进制掩码(IBM)。IBM和IBM估计， 对有感觉神经的听者来说，在噪声中理解语言的能力有显著的提高 听力损失。然而，到目前为止，对这项技术的研究主要集中在健康、完整的语音上。这个 提出的研究涉及IBM的一种创新应用-克服噪声中的语音困难 用退化的语音。此应用程序对于听力正常或听力损失的听众很重要，因为 即使是正常听力的听者也很难理解噪音中退化的语音。这项建议 旨在演示IBM对降级语音信号的应用的概念验证，从 构音障碍的测试案例。在目标1中，我们量化了IBM恢复构音障碍语音清晰度的能力 在安静的情况下，噪音将达到性能水平。我们将对噪声、语音中的语音条件进行比较 在由IBM处理的噪声中，以及在安静的基线语音中，对于健康和构音障碍的语音。一个 预测模型将量化IBM处理对听力正常的听众的好处，并可能 这项福利的版主。在目标2中，我们研究了IBM为听众带来的好处 感觉神经性听力障碍。数据分析将遵循目标1，模型结果将是 与目标1中的相比。该项目的成功完成将为R01的开发提供信息 评估IBM处理对各种严重程度和类型的降级或 不完美的讲话，并解释了诸如听力努力等认知过程。了解IBM如何 处理不完美的语音信号将支持当前IBM-估计的精细化 算法，在现实的收听环境中优化功能。这一提议可能产生的影响 研究是深刻的-基于IBM的处理不仅具有改善听众生活的潜力，而且 听力损失，但也包括构音障碍和其他言语障碍的说话者的对话伙伴。这个 这项工作的长期目标是提供临床上有意义的工具来改变这两个重要的 人群--言语障碍患者和听力损失患者。这项合作研究 计划是概念验证的第一步，展示了这一前景看好的几种可能的新应用 新的临床治疗技术。

Project Summary/Abstract The presence of background noise often has a substantial, negative impact on a listener’s ability to understand speech. Unfortunately, deciphering speech in noisy environments is commonplace for listeners. As our recent preliminary data demonstrate, when the speech signal itself is degraded, such as due to the neurological speech disorder of dysarthria, the addition of background noise only further decreases a listener’s ability to understand that speech (Yoho & Borrie, 2018). A highly promising technique developed to remediate much of this speech-in-noise difficulty is termed the Ideal Binary Mask (IBM). The IBM, and IBM estimation, has demonstrated significant improvements in understanding speech in noise for listeners with sensorineural hearing loss. However, research on this technique to date has focused on healthy, intact speech. The proposed research involves an innovative application of the IBM—to overcome the speech-in-noise difficulties with degraded speech. This application is important for listeners with either normal hearing or hearing loss, as even normal hearing listeners struggle considerably to understand degraded speech in noise. This proposal aims to demonstrate proof-of-concept for the application of IBM to a degraded speech signal, starting with the test case of dysarthria. In Aim 1, we quantify the ability of the IBM to restore intelligibility of dysarthric speech in noise to performance levels in quiet. Comparisons will be made between conditions of speech in noise, speech in noise processed by the IBM, and baseline speech in quiet, for both healthy and dysarthric speech. A predictive model will quantify the benefit of IBM processing for normal-hearing listeners, and possible moderators on this benefit. In Aim 2, we examine the benefit achieved from the IBM for listeners with sensorineural hearing impairment. Data analysis will follow that of Aim 1, and model outcomes will be compared to those from Aim 1. Successful completion of this project will inform the development of a R01 proposal evaluating the effects of IBM processing on a wide range of severities and types of degraded or imperfect speech, and account for cognitive processes such as listening effort. Knowledge of how IBM processing works with an imperfect speech signal will support the refinement of current IBM-estimation algorithms, optimizing functionality in realistic listening environments. The possible impact of this proposed research is profound—IBM-based processing has the potential not only to improve the lives of listeners with hearing loss, but also the conversational partners of speakers with dysarthria and other speech disorders. The long-term goal of this work is to provide clinically-significant tools to transform treatment of these two important populations—individuals with speech disorders and individuals with hearing loss. This collaborative research plan is the first proof of concept step to demonstrate several possible novel applications of this promising technique for new clinical treatments.