DHB/Collaborative Research: Using Machine Learning to Model the Interplay of Production Dynamics and Perception Dynamics in Phonological Acquision

DHB/协作研究：使用机器学习对语音习得中的产生动态和感知动态的相互作用进行建模

• 批准号: 0729140
• 负责人: Jan Edwards
• 金额: $ 30.31万
• 依托单位: University of Wisconsin-Madison
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2008
• 资助国家: 美国
• 起止时间: 2008-01-15 至 2011-12-31
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=0729140&HistoricalAwards=false
• 关键词: DHB; Collaborative; Research; Using; Machine

This project adapts acoustic modeling techniques for robust Automatic Speech Recognition to a large multi-language database of adult and child speech recordings. It explores how cognitive representations relevant to speech production and perception in any given speech community come to be internalized by normally developing children.Current models of language acquisition cannot account for fine-grained phonetic differences between speech communities and between younger and older speakers within a community because they are based on research that uses alphabetic transcription as the primary observational tool. Because this project is a collaboration among specialists in speech engineering, linguistics, and speech and hearing science it enables child-language researchers to develop finer-grained measures to begin to answer several key questions about the dynamics of language acquisition, including the following three: How do adult speakers from different speech communities perceive the acoustic patterns in words pronounced by speakers of different ages? How do adult stereotypes about children's productions shape speech addressed to children? How do adults' perceptual responses to children's (mis)articulations shape the course of children's mastery of the community-specific patterns? These questions are a key part of understanding what similarities and differences exist across cultures and what these mean for our understanding of the human capacity for speech evolved. In addition, by linking together advances in linguistic and engineering, the modeling results should provide a basis for developing a new approach to speech technology, with broader impacts in several areas such as assistive technology for the disabled and more democratic access to electronic databases.

本项目将声学建模技术用于强大的自动语音识别，以适应成人和儿童语音记录的大型多语言数据库。 它探讨了在任何特定的言语社区中，与言语产生和感知相关的认知表征是如何被正常发育的儿童内化的。目前的语言习得模型不能解释言语社区之间以及社区内年轻人和老年人之间的细粒度语音差异，因为它们是基于使用字母转录作为主要观察工具的研究。 因为这个项目是语音工程，语言学，语音和听力科学专家之间的合作，它使儿童语言研究人员能够开发更细粒度的措施，开始回答有关语言习得动态的几个关键问题，包括以下三个：来自不同语言社区的成年人如何感知不同年龄的说话者发音的声学模式？ 成年人对儿童作品的刻板印象如何塑造针对儿童的演讲？成年人对儿童发音（错误）的知觉反应如何影响儿童掌握社区特定模式的过程？ 这些问题是理解不同文化之间存在的相似性和差异的关键部分，以及这些对我们理解人类语言能力进化的意义。 此外，通过将语言学和工程学的进步联系在一起，建模结果应该为开发语音技术的新方法提供基础，并在残疾人辅助技术和更民主地访问电子数据库等几个领域产生更广泛的影响。