Doctoral Dissertation Research:Modeling temporal coordination in speech production using an artificial central pattern generator neural network

博士论文研究：使用人工中央模式生成器神经网络对语音产生中的时间协调进行建模

• 批准号: 1155592
• 负责人: Jennifer Cole
• 金额: $ 0.53万
• 依托单位: University of Illinois at Urbana-Champaign
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2012
• 资助国家: 美国
• 起止时间: 2012-08-01 至 2014-01-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=1155592&HistoricalAwards=false
• 关键词: Doctoral; Dissertation; Research; Modeling; temporal

This project investigates how timing is coordinated in speech production and how this coordination can be computationally modeled using artificial neural networks. Research has shown that speech rhythm reflects a multi-tiered, hierarchical organization of speech units (syllables, accentual units, phrases) that are essentially cyclical in nature. One type of neural network which shares these properties, the central pattern generator (CPG), has been hypothesized to underlie speech timing. This project presents two speech production experiments designed to investigate temporal coordination within and among three levels of speech units in French and English, languages with distinct rhythm types, and to discover which timing properties they share and which are language-specific. A three-level CPG-type artificial neural network is presented which is used to model the results of both experiments in order to test the ability of such a model to simulate the timing behavior of two rhythmically distinct languages. Experiment 1 focuses on the coordination between phrases, accentual units, and syllables through a comparison of the durational effects of lengthening due to phrasal stress and phrase-final syllable lengthening in both languages. Experiment 2 compares the overall temporal coordination of spoken phrases and their constituent parts in the two languages by measuring the changes in articulatory timing of phrases which are repeated multiple times. The main goals of the proposed studies and modeling work are to investigate temporal coordination between the hierarchically structured levels of speech units, specifically how that coordination varies between languages, and to test the ability of a biologically inspired CPG-type neural network model to model this coordination and the speech patterns that result from it.This work contributes to an understanding of the coordination and timing of rhythm units in speech and how that coordination may be generated by the brain. In its broader impact, it will enable the integration of theories of speech production with other motor behavior in biological and cognitive models, and help to forge links between research on speech, linguistic prosody, and more broadly, neural models of animal behavior. This work will also contribute to the construction of a unified model of rhythm and prosody which is sufficient to explain both the underlying universalities and the variations of human language. The project will also enrich the training of the graduate student co-PI.

该项目研究了语音产生中如何协调时间以及如何使用人工神经网络对这种协调进行计算建模。研究表明，语音节奏反映了语音单元（音节、重音单元、短语）的多层次、分层组织，本质上是循环的。一种具有这些特性的神经网络，即中央模式生成器 (CPG)，已被假设为语音定时的基础。该项目提出了两个语音生成实验，旨在研究法语和英语（具有不同节奏类型的语言）三个级别的语音单元内部和之间的时间协调，并发现它们共享哪些计时属性以及哪些是特定于语言的。提出了一种三级 CPG 型人工神经网络，用于对两个实验的结果进行建模，以测试该模型模拟两种节奏不同语言的计时行为的能力。实验1通过比较两种语言中短语重音和短语结尾音节延长造成的持续时间效果，重点关注短语、重音单位和音节之间的协调。实验2通过测量多次重复的短语的发音时序的变化来比较两种语言中口语短语及其组成部分的整体时间协调性。所提出的研究和建模工作的主要目标是研究语音单元的层次结构级别之间的时间协调，特别是这种协调在不同语言之间如何变化，并测试受生物启发的 CPG 型神经网络模型对这种协调和由此产生的语音模式进行建模的能力。这项工作有助于理解语音中节奏单元的协调和时间安排，以及这种协调如何发挥作用。 由大脑产生。 就其更广泛的影响而言，它将使得言语产生理论与生物和认知模型中的其他运动行为相整合，并有助于在言语、语言韵律以及更广泛的动物行为神经模型研究之间建立联系。这项工作还将有助于构建一个统一的节奏和韵律模型，足以解释人类语言的潜在普遍性和变化。该项目还将丰富研究生co-PI的培训。