Mathematical modeling of the temporal acpects of speech based on human perceptual and cognitive mechanisms

这个项目的目的是寻找线索，人类检索语言的时间结构，了解他们的用法，并建立一个定量的方法来评估一个给定的语音的时间充分性或自然性，可以复制人类判断的表现。为此，我们实施了三个基本的调查任务：心理物理层面的研究、语言层面的研究和评价模型的构建。这个项目的一个显著特点是，尽管它的主要对象是口语，但它对心理物理方面的强调几乎和语言方面一样多。由于一个人可以很容易地识别快速的言语，即使是在意义未知的外语的情况下，人们假设言语的时间方面的处理无疑涉及非语言的，因此是独立于语言的活动。通过专注于这种独立于给定语言的处理，我们开发了一种基本技术，用于语言处理的开销很小，并且可以简单地扩展到多种语言。主要结果如下。(1)心理物理层面：开发了一种算法，通过复制人类听觉处理的功能来预测给定语音中的时间参考点。这个算法最重要的好处是它几乎适用于无限的语言变化，因为它不需要任何语言知识。(2)语言层面：一项实证研究表明，影响韵律单位感知的因素因特定语言的单位选择而异。这一发现为设计有效的外语训练方法时韵律因素的重要性提供了实际意义。（三）建模：通过整合心理物理层面调查得出的听觉功能，实现数学模型，自动评价英语学习者的语音自然度。该模型的表现与母语为英语的教师的主观评价非常接近。这一结果不仅表明心理物理因素在言语的适当性或自然性评价中的重要性，而且还暗示了所提出的模型在多种语言中的潜在可扩展性。少

The purpose of this project was to find clues by which humans retrieve the temporal structure of speech, to understand their usage, and to establish a quantitative method to evaluate the temporal adequateness or naturalness of a given speech sound that can replicate the performance of human judgment. For this purpose, three fundamental investigative tasks were implemented a study at the psychophysical level, a study at the linguistic level, and the construction of an evaluation model. One distinguishing feature of this project is that it emphasized the psychophysical aspects nearly as much as the linguistic, even though its primary object was spoken language.Since a person can easily recognize fast speech, even in the case of a foreign language where the meaning is unknown, it is assumed that the processing of the temporal aspects of speech undoubtedly involve non-linguistic and therefore language-independent activities. By concentrating on such processing that is independent of a give … More n language, we developed the basic technology toward a system with small overhead for language processing as well as simple extensibility to multiple languages. The major results follow.(I) Psychophysical level : An algorithm was developed to predict temporal reference points in a given speech by replicating the function of human auditory processing. This algorithm's most important benefit is its applicability to virtually unlimited language variations since it doesn't require any linguistic knowledge.(II) Linguistic level : An empirical study levealed that factors, which affect the perception of prosodic units, vary depending on the particular language's choice of units. This finding provides practical implications concerning how much weight should be placed on prosodic factors when designing effective foreign-language training methods.(III) Modeling : By integrating auditory functions derived from investigation at the psychophysical level, a mathematical model was implemented to automatically evaluate the naturalness of the speech of the English learners. The model's performance closely approximated the subjective evaluation of a native-speaking English instructor. This result not only suggests the importance of psychophysical factors in the adequateness or naturalness evaluation of speech but it also implies potential extensibility of the proposed model to multiple languages. Less