SGER: Loudmouth - Toward Intelligible Speech Synthesis in Everyday Noise Situations

合成语音已经走出了科幻小说的想象领域，它提供了从HAL到C-3 PO的机器人的声音，进入了我们日常生活中的电器，电话应用和个人通信设备。 在辅助技术领域，有严重沟通障碍的人通常依靠语音合成来代表他们说话。 然而，任何试图在嘈杂的火车站或机场收听PA系统的人都知道，语音合成在这些日常噪音情况下会失败。 这些正是用户必须最依赖该技术的通信场景类型，因此突出了开发能够适应用户的情景噪声背景的说话风格的合成器的重要性。 在这个探索性的项目中，PI将朝着设计和开发一个语音合成器迈出第一步，该语音合成器可以监听环境噪声水平，并修改其韵律，以类似于人类的方式补偿背景噪声。 PI将开始通过检查由人类制造的噪声中的语音修改来学习如何在机器中实现这些改变。 她假设人类对语音的修改对于语义突出（即内容）的词与非突出（即功能）的词是不同的;这是一个重要的区别，但没有得到太多的关注。 如果假设被证明是正确的，自适应合成器将需要有区别地修改其说话风格，以考虑所讲单词的语义作用以及背景噪声水平。更广泛的影响：这项研究将影响所有使用口语对话界面的人（例如，在电话和虚拟交易应用中），以及必须依赖语音合成作为其唯一通信模式的残疾人。 大多数交流互动发生在有一定程度环境噪音的活跃空间。 当前的语音合成技术不能科普这样的条件，因此它不能在诸如教室、餐馆、旅馆大厅、汽车、公共汽车站和许多其它情况的普通情况下用作使能技术。

Synthetic speech has moved out of the imaginary realm of science fiction, where it provided the voices of robots from HAL to C-3PO, into our everyday world of appliances, telephony applications and personal communication devices. In the domain of assistive technology, individuals with severe communication disorders often rely on speech synthesis to speak on their behalf. Yet, anyone who has tried to listen to the PA system in a noisy train station or airport knows that speech synthesis fails in these everyday noise situations. These are precisely the types of communication scenarios in which users must rely most on the technology, thus highlighting the importance of developing a synthesizer capable of accommodating its speaking style to the user's situational noise context. In this exploratory project the PI will take first steps toward designing and developing a speech synthesizer that listens to the ambient noise level and modifies its prosody to compensate for background noise in human-like ways. The PI will begin by examining the modifications to speech in noise made by humans in order to learn how to implement these changes in machines. She posits that human modifications to speech will differ for semantically salient (i.e. content) words compared to non-salient (i.e. function) words; this is an important distinction that has not received much attention. If the hypothesis proves to be correct, the adaptive synthesizer will need to differentially modify its speaking style to account for the semantic role of the words spoken as well as the background noise level.Broader Impacts: This research will impact all people who use spoken dialog interfaces (e.g., in telephony and virtual transaction applications), as well as individuals with disabilities who must rely on speech synthesis as their sole mode of communication. Most communicative interactions occur in lively spaces that have some degree of ambient noise. Current speech synthesis technology cannot cope with such conditions and thus it fails to serve as an enabling technology in commonplace situations such as classrooms, restaurants, hotel lobbies, cars, bus stations, and numerous other situations.