EXP: Collaborative Research: Perception and Production in Second Language: The Roles of Voice Variability and Familiarity

许多来自非英语国家的高技能专业人士在高等教育、医疗保健和技术领域工作，推动了美国经济的发展。这些专业人士通常有扎实的英语知识，但与年龄有关的非母语发音会损害他们的可理解性。高技能的职业需要高级的口头沟通技巧，包括高度清晰的发音；因为这些专业人士来自不同的母语，他们的发音需求是高度个性化的。因此，传统的课堂教学不容易满足不同学习者的需求，也不允许大量的一对一教学。为了解决这个问题，研究人员提出了一种创新的计算机辅助发音训练（CAPT）技术，允许学习者按照自己的节奏练习，并根据自己的意愿，使用个性化的语音模型，增加提高可理解性的可能性。具体来说，这种方法建立在研究的基础上，研究表明第二语言学习者在模仿类似的声音时更有可能成功，这就是所谓的黄金演说家。建议的CAPT技术为每个学习者提供了一个理想的黄金扬声器：他们自己的声音，但带有母语口音，通过数字学习环境，允许第二语言学习者通过选择不同的美国地区口音，逐步和互动地建立他们的黄金扬声器。这一愿景是由美国国家科学基金会资助的一项正在进行的口音转换算法研究实现的，该算法可以通过参考母语人士的超分词和分词特征来数字化地改变第二语言的话语。这个网络学习项目将口音转换技术以基于网络的交互式学习工具的形式引入发音教学。该项目使用基于设计的研究范式来(1)了解人们如何在与计算黄金扬声器互动时学习第二语言发音技能；(2)确定在第二语言学习环境中实施该技术的有效方法。该项目最初侧重于西班牙语英语学习者，后来将该方法推广到其他母语人士。

The U.S. economy is boosted by many highly skilled professionals from non-English speaking countries, who work in higher education, healthcare, and technology. These professionals typically have solid knowledge of English but age-related non-native pronunciations that can impair their intelligibility. Highly-skilled professions require advanced oral communication skills, including highly intelligible pronunciation; because such professionals come from a variety of first languages, their pronunciation needs are highly individualized. Thus, traditional classroom instruction cannot easily meet the needs of diverse learners nor allow for significant amounts of one-on-one instruction. To address this issue, the investigators propose an innovative genre of technology for computer-assisted pronunciation training (CAPT) that allows learners to practice at their own pace, and as often as they wish, with a personalized voice model, increasing the likelihood of improved intelligibility.Specifically, the approach builds on research showing that second-language (L2) learners are more likely to succeed when imitating a similar voice, a so-called golden speaker. The proposed CAPT technology provides an ideal golden speaker for each learner: their own voice but with a native accent through a digital learning environment that allows L2 learners to build their golden speakers incrementally and interactively by selecting among different regional US accents. This vision is enabled by on-going NSF-sponsored research on accent-conversion algorithms to digitally alter L2 utterances with suprasegmental and segmental features from a reference native speaker. This Cyberlearning project brings accent-conversion techniques into pronunciation instruction in the form of a web-based interactive learning tool. The project uses a design-based research paradigm to (1) gain knowledge of how people learn L2 pronunciation skills while interacting with computational golden speakers; and (2) identify efficient ways of implementing this technology in L2 learning contexts. The project focuses initially on Spanish learners of English, later generalizing the approach to other first language speakers.