Microscopic modeling of hearing impaired listeners speech recognition using a physiologic auditory model

使用生理听觉模型对听力受损听众语音识别进行微观建模

• 批准号: 209735833
• 负责人: Professor Dr. Tim Jürgens
• 金额: --
• 依托单位: Department of Computer Science
• 依托单位国家: 德国
• 项目类别: Research Fellowships
• 财政年份: 2011
• 资助国家: 德国
• 起止时间: 2010-12-31 至 2012-12-31
• 项目状态: 已结题
• 来源: https://gepris.dfg.de/gepris/projekt/209735833?language=en
• 关键词: Microscopic; modeling; hearing; impaired; listeners

Degraded speech intelligibility is one of the most frequent complaints of sensorineural hearing-impaired listeners (the most common form of hearing impairment) in noisy situations. The aim of this project is to gain a better understanding of the impact of different types of sensorineural hearing impairment on human speech recognition. The knowledge obtained will be beneficial for the design of new hearing aids. I intend to use a computational modeling approach to model the speech recognition of hearing-impaired listeners in order to understand this problem. Mixtures of speech and noise will be processed by a model of the auditory periphery, which supplies the input to an automatic speech recognition (ASR) system. This approach can be termed a microscopic speech recognition model, because it operates on digitized time-domain signals (rather than operating on long-term spectra). This simulates the processing that takes place in the human auditory system. Furthermore, I intend to model the recognition of single phonemes rather than whole sentences. The proposed experiments will replace the previously in my dissertation used auditory model with a model that mimics individual physiological stages in the auditory system much more closely. Hitherto, human auditory processing was only modeled in an "effective" (phenomenological) way. The new, more physiological auditory model, developed at the University of Essex, reproduces auditory nerve firing patterns, i.e. those patterns that the central nervous system has access to.The choice of Essex for this research will allow me to do additional speech recognition experiments with hearing-impaired listeners, for whom a very precise diagnosis of their hearing has already been generated using recently developed new techniques. A second advantage of this location is a close cooperation with the „Speech and Hearing“ group at the University of Sheffield (worldwide leading in ASR research).

言语清晰度下降是感音神经性听力受损的听众（最常见的听力障碍形式）在嘈杂环境中最常见的投诉之一。该项目的目的是更好地了解不同类型的感音神经性听力障碍对人类语音识别的影响。所获得的知识将有益于新助听器的设计。我打算使用一个计算建模的方法来模拟听力受损的听众的语音识别，以了解这个问题。语音和噪声的混合物将由听觉外围的模型处理，该模型将输入提供给自动语音识别（ASR）系统。这种方法可以被称为微观语音识别模型，因为它对数字化的时域信号进行操作（而不是对长期频谱进行操作）。这模拟了在人类听觉系统中发生的处理。此外，我打算建模单个音素的识别，而不是整个句子。所提出的实验将取代以前在我的论文中使用的听觉模型与模型，模仿个人的生理阶段在听觉系统更密切。然而，人类的听觉处理只是以一种“有效的”（现象学的）方式建模的。埃塞克斯大学开发的新的、更符合生理学的听觉模型再现了听觉神经的放电模式，即中枢神经系统可以接触到的模式，选择埃塞克斯进行这项研究将使我能够对听力受损的听众进行额外的语音识别实验，对他们来说，已经使用最近开发的新技术对他们的听力进行了非常精确的诊断。这个位置的第二个优势是与谢菲尔德大学的"言语和听力”小组（全球领先的ASR研究）密切合作。