A Study on Production of Personalized Programs with a Virtual Human Agent on the WWW

本研究的目的是开发一个环境，使我们能够产生类似电视节目，其中一个虚拟人代理作为一个演示者，在万维网（WWW）上。我们首先扩展了我们的虚拟人演示者系统，其动画语言是语音文本的形式与手势相关的嵌入式命令，使其能够与日语在万维网上工作。其次，我们尝试了自动生成动画天气报告，其中虚拟人扮演气象预报员，作为自动化电视类节目制作的示例。我们的原型系统的输入是一组（位置，时间，天气）类型的天气预报数据。然后，系统通过聚类对数据进行汇总，并生成其案例框架表示。一组模板将这些表示转换为日语文本，并嵌入手势候选项。通过适当地选择手势候选者，获得用于天气报告的命令嵌入式语音文本。扩展后的虚拟人演示系统从生成的语音文本中，使虚拟人Agent扮演天气预报员的角色，并以实时三维动画的形式呈现天气预报图，与语音输出同步，实验结果表明，该系统能够很好地制作出WWW上的动画天气预报节目，证明了该方法的有效性。这项研究的成果在国际多媒体建模会议（MMM2000）上发表。

The purpose of this research project is to develop an environment which enables us to produce TV-like programs, where a virtual human agent acts as a presenter, on the World Wide Web (WWW).We first extended our virtual human presenter system, whose animation language is in the form of speech texts with gesture-related embedded commands, so that it works with Japanese language on the WWW.Secondly, we tried automatic generation of animated weather reports, where a virtual human performs as a weatherman, as an example of automated TV-like program production. An input to our prototype system is a set of (location, time, weather)-type weather forecast data. The system then summarizes the data by clustering and generates its case frame representations. A set of templates translates these representations into Japanese texts with embedded candidates of gestures. By selecting the gesture candidates appropriately, command-embedded speech texts for the weather reports are obtained. From the generated speech texts, the extended virtual human presenter system makes the virtual human agent act as a weatherman with weather charts in real-time 3D animation synchronized with speech outputs.Our experiments showed that the system well produces animated weather report programs on the WWW, and proved the validity of our approach. The results of this research were presented at the International Conference on Multimedia Modeling (MMM2000).