Durational Patterns in Connected Discourse

关联话语中的持续模式

• 批准号: 9983106
• 负责人: Caroline Smith
• 金额: $ 12.78万
• 依托单位: University of New Mexico
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2000
• 资助国家: 美国
• 起止时间: 2000-05-01 至 2004-04-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=9983106&HistoricalAwards=false
• 关键词: Durational; Patterns; Connected; Discourse

The speech synthesis systems now available can produce speech whosesegmental properties mimic human speech remarkably well. However, syntheticspeech still often sounds unnatural, primarily because of the relativelyless successful modeling of the prosodic properties of speech, such asintonation and timing. This project focuses on the prosodic properties oflonger stretches of speech - sequences of sentences and paragraphs. Thegoal is to explore how prosody, specifically durations, is organized overlarger units of speech.The approach adopted here is to compare different languages, with knowndifferences in timing at segmental, word or phrase level, to see how theydiffer in structuring timing over an entire discourse. The languages thatwill be compared are English, French and Japanese. Improved understandingof discourse-level timing will help to explain the relation betweendurational patterns and linguistic structure.The project is organized into four phases: (1) a production study toinvestigate what factors influence durational patterns in longer stretchesof speech; (2) modeling of the results of the production study; (3)analysis-synthesis incorporating the model of durations into synthesizedspeech; and (4) testing the resulting synthesis through evaluation by humanlisteners. The durational properties to be examined include amount oflengthening at sentence boundaries, duration of pauses, and changes inspeech rate before and after a boundary. Comparisons among the threelanguages will identify the factors that are important in influencingdurations, and help explain how these factors relate to the differentlinguistic structures of the languages.

现有的语音合成系统可以产生分段特性非常好地模仿人类语音的语音。然而，合成语音听起来仍然经常不自然，主要是因为对语音的韵律属性(如语调和定时)的建模相对较不成功。这个项目关注的是较长语段的韵律特性--句子和段落的序列。我们的目标是探索韵律，特别是时长是如何在较大的言语单位上组织的。这里采用的方法是比较不同的语言，在分段、单词或短语层面上已知的时间差异，看看它们在整个语篇的时间结构上有什么不同。将进行比较的语言是英语、法语和日语。本项目分为四个阶段：(1)生产研究，以考察在较长语段中哪些因素影响持续模式；(2)对生产研究的结果进行建模；(3)分析-合成，将持续时间模型融入合成语音中；(4)通过听者的评估来检验合成的结果。要检查的持续时间属性包括句子边界处的增强量、停顿的持续时间以及边界前后的语速变化。三种语言之间的比较将确定影响持续时间的重要因素，并有助于解释这些因素与语言的不同语言结构之间的关系。