Physiologically-Based Signal Processing Schemes

描述（由申请人提供）：尽管助听器技术取得了重大进展，但目前的信号处理方案还没有解决听力受损听众的几个问题。该项目将开发新颖的、基于生理的信号处理策略，以解决助听器用户面临的两个主要问题：嘈杂环境中的听力困难和响度失真，这限制了听障听众在舒适听力水平下的动态范围。我们最近的生理学研究提出了背景噪声和水平编码中信号掩蔽检测的神经编码和处理机制。这些研究已经产生了定量模型，成功地预测了人类听众在与掩蔽检测和水平歧视相关的心理物理任务中的表现。在这个项目中，我们将利用这些模型背后的基本概念，并将它们转换为信号处理算法。这项工作不仅为我们的模型提供了额外的测试，而且还提供了一个机会，将我们的生理和模型研究提出的想法应用于听力受损听众的实际问题。本项目将探讨两种策略。(1)基于掩蔽神经模型的降噪。我们将使用掩模检测模型来识别存在背景噪声的信号。以信号能量为主的频段将被放大，而其他信道将被衰减。我们的模型在波动噪声中检测信号的成功是该方法的一个重要方面。(2)基于层次编码神经模型的响度感知补偿。我们将引入在健康耳蜗中产生的电平相关的交叉频率相位差，利用模拟听觉神经调谐的非线性滤波器。目标是增加舒适的音量范围，并通过向受损的耳朵提供这些非线性线索来改善语音识别。

DESCRIPTION (provided by applicant): Despite significant advances in hearing-aid technology, several problems for hearing-impaired listeners have not been solved by current signal-processing schemes. This project will develop novel, physiologically-based signal processing strategies to address two major problems faced by hearing-aid users: difficulty listening in noisy environments and loudness distortion, which limits hearing-impaired listeners' dynamic range of comfortable listening levels. Our recent physiological studies have suggested neural encoding and processing mechanisms for masked detection of signals in background noise and for level coding. These studies have resulted in quantitative models that successfully predict the performance of human listeners on psychophysical tasks related to masked detection and level discrimination. In this project, we will take advantage of the basic concepts behind these models and convert them into signal-processing algorithms. This effort not only provides additional tests for our models, but also provides an opportunity to apply ideas suggested by our physiological and modeling studies to real problems for hearing-impaired listeners. Two strategies will be explored in this project. (1) Noise reduction based on a neural model for masking. We will use our masked-detection model to identify signals in the presence of background noise. Frequency bands that are dominated by signal energy will be amplified, and other channels will be attenuated. The confirmed success of our model in detecting signals in fluctuating noises is an important aspect of this approach. (2) Compensation of perceived loudness based on a neural model for level coding. We will introduce into the signal the level-dependent cross-frequency phase differences that are created in the healthy cochlea, taking advantage of nonlinear filters that simulate auditory-nerve tuning. The goal is to increase the comfortable range of levels and to improve speech recognition by providing these nonlinear cues to the impaired ear.