PROCESSES OF SPEAKER NORMALIZATION IN SPEECH PERCEPTION

言语感知中说话人标准化的过程

• 批准号: 2126635
• 负责人: Keith A Johnson
• 金额: $ 9.4万
• 依托单位: OHIO STATE UNIVERSITY
• 依托单位国家: 美国
• 财政年份: 1992
• 资助国家: 美国
• 起止时间: 1992-07-01 至 1997-06-30
• 项目状态: 已结题
• 来源: https://reporter.nih.gov/project-details/2126635
• 关键词: auditory stimulus; hearing; human subject; identity; mathematical model; perception; psychoacoustics; psychomotor reaction time; short term memory; sound frequency; speech; speech synthesizers; stimulus /response; visual stimulus

This project will investigate the listener's ability to understand speech produced by different talkers. This perceptual ability (which is called speaker normalization) is one example of the adaptiveness of speech perception, and so a better understanding of how listeners adapt to different talkers may lead to a better understanding of the adaptive perceptual processes which take place in response to hearing impairment. In spite of its intrinsic and practical importance, speaker normalization has not been a focus of interest for most speech perception researchers. The project will compare two models of speaker normalization. The mediated normalization model holds that the perception of speech is cognitively mediated by information about the speaker, whereas the immediate normalization model holds that information in the speech signal resulting from acoustic differences between speakers is auditorily integrated with linguistic acoustic cues in such a way that differences between speakers are eliminated. These two models will be contrasted in 13 experiments involving tests of (1) the role of familiarity with particular speakers in speech perception, (2) the role of secondary acoustic cues for speaker identity (such as breathiness) and visual information about the speaker, and (3) speaker normalization effects for plosive consonants and nasals. The proposed experiments make use of synthetic speech or digitally processed natural speech and involve manipulating acoustic cues for speaker identity, interstimulus interval, and presentation type (randomly mixing speakers or blocking stimuli by speaker) while observing the impact of these manipulations on the perception of vowels, fricatives, stops, and nasals. In addition to the experiments, the proposed program of research involves the construction of mathematical models of speech perception. Exemplar-based implementations of both the mediated and immediate normalization models will be constructed and tested against the data generated by the experiments.

这个项目将调查听者理解语言的能力 由不同的演讲者制作。这种感知能力(称为 说话人标准化)是语音适应性的一个例子 感知，从而更好地理解听众如何适应 不同的说话者可能会更好地理解适应性 对听力受损作出反应的知觉过程。 尽管说话人标准化具有内在的和实际的重要性， 并不是大多数言语知觉研究人员感兴趣的焦点。 该项目将比较两种说话人归一化模型。这个 中介归一化模型认为，言语知觉是 由说话人的信息进行认知调节，而 立即归一化模型保持语音信号中信息 由于说话者之间的声学差异而产生的 以这样一种方式与语言声音线索相结合，即差异 免去了扬声器之间的干扰。这两个型号将作对比。 13个实验涉及测试(1)熟悉度的作用 特定说话人在言语感知中的作用，(2)第二语言的作用 用于说话人身份识别(如喘息)和视觉的声学提示 有关说话人的信息，以及(3)说话人归一化效果 塞音辅音和鼻音。拟议中的实验利用了 合成语音或数字处理的自然语音，并涉及 操纵用于说话人身份、刺激间间隔 和演示类型(随机混合说话者或通过以下方式阻止刺激 演讲者)，同时观察这些操作对 对元音、擦音、塞音和鼻音的感知。除 实验中，提出的研究方案涉及到建设 语言感知的数学模型。基于样本的 中介和即时归一化模型的实现 生成的数据进行构造和测试 实验。