Using Speech Recognition to Enhance Communication Capabilities for Individuals with Physical Disabilities

这个项目建立在PI以前的工作基础上，特别强调以沟通为导向的活动和身体残疾的个人。该研究将集中于改善用户与语音识别系统的交互。此前，PI开发了一种新的错误规范技术，该技术使用置信度分数来促进识别错误纠正过程。虽然其他人早先使用信心分数来达到这一目的的努力在很大程度上是不成功的，但PI对他的技术的初步评估产生了有希望的结果；早期的工作依赖于识别特定的错误单词，而PI的技术即使只能识别可能存在错误的区域，也可以证明是有益的。在新项目中，PI将把置信度分数与自然语言处理技术相结合，以促进纠正识别错误的过程。他将开发新的交互技术和模型，支持新的交互场景，并允许更好地理解用户和语音识别系统之间的关系。需要解决的问题包括：当纠错活动明确地与听写分离时，我们能否扩展现有模型以有效地描述用户与语音识别系统之间的交互？当这些活动暂时与听写分离时，我们如何支持发现和纠正识别错误的过程？当完成涉及延迟错误纠正的任务时，个人（例如，传统用户和有各种残疾的个人）的能力和限制如何影响他们的策略？更广泛的影响：这项研究扩展了UMBC信息系统部门和劳动力和技术中心（WTC）之间现有的合作。语音识别对身体残疾的人来说是一项重要的技术。虽然与传统计算机用户相比，这些人采用了不同的策略，但有限的研究集中在这一重要的潜在用户群体使用语音识别技术上。因此，通过明确调查这些个人在使用语音识别技术时遇到的挑战，该项目将直接影响这一代表性不足的用户群体，并将进一步促进他们更大的独立性和新的就业机会。由于这个项目将包括范围广泛的用户，其结果也将使没有身体残疾的传统计算机用户受益。通过与WTC的合作，PI将通过适当的出版物在康复界以及HCI和ASR界传播他的研究结果。

This project builds on the PI's previous efforts, placing a special emphasis on communications-oriented activities and individuals with physical disabilities. The research will focus on improving user interaction with speech recognition systems. Previously, the PI developed a novel error specification technique that uses confidence scores to facilitate the recognition error correction process. While earlier efforts by others to use confidence scores for this purpose had largely been unsuccessful, the PI's initial evaluation of his technique has yielded promising results; whereas earlier efforts relied on identifying specific words that were incorrect, the PI's technique can prove beneficial even if it is only possible to identify the regions where errors may exist. In the new project, the PI will integrate confidence scores with techniques from natural language processing, to facilitate the process of correcting recognition errors. He will develop novel interaction techniques and models, that support new interaction scenarios and allow for better understanding of the relationship between users and speech recognition systems. Questions to be addressed include the following: Can we expand existing models to effectively describe interactions between users and speech recognition systems when error correction activities are explicitly separated from dictation? How can we support the process of finding and correcting recognition errors when these activities are temporally separated from dictation? How do the abilities and limitations of individuals (e.g., traditional users and individuals with various disabilities) affect their strategies when completing tasks that involve deferred error correction?Broader Impacts: This research extends an existing collaboration between UMBC's Information Systems Department and the Workforce and Technology Center (WTC). Speech recognition can be a critical technology for individuals with physical disabilities. While these individuals have been shown to employ different strategies as compared to traditional computer users, limited research has focused on the use of speech recognition technologies by this important group of potential users. Therefore, by explicitly investigating the challenges these individuals encounter when using speech recognition technologies, this project will directly impact this underrepresented group of users, and will furthermore facilitate greater independence and new employment opportunities for them. Because a broad range of users will be included in this project, the results will also benefit traditional computer users with no physical disabilities. Through the collaboration with the WTC, the PI will distribute his results via appropriate publications within the rehabilitation community, in addition to the HCI and ASR communities.