Research on Database Access by Speech

语音是人类最自然、最确定的交流方式之一，近年来语音识别技术取得了巨大的进步。数据库是各类企业的综合信息资源，随着社会的计算机化，数据库的价值越来越大。然而，在大多数情况下，数据库访问是通过使用传统的关键字来完成的。这种情况令人遗憾，因为现在的社会正在向移动和无处不在的社会发展，人们很难使用键盘来获取信息。本研究试图通过建立一个与关系数据库系统的语音接口来解决这个问题。我们实现了一个由Sun工作站、Oracle 7关系数据库管理系统、DS200语音识别系统和FINE TALK语音合成系统组成的系统。我们开发了四个软件模块来组合这些系统组件。对原型进行了评价，揭示了需要解决的同音问题以及如何容纳日语句子不完整的演讲。在本研究中，我们附加了第一个问题，并通过引入“实例键”给出了一个解决方案，实例键是使用功能依赖定义的传统键的泛化。实验结果表明，采用该方法可以实现更复杂的语音接口。

Speech is one of the most natural and certain ways of human communication, and the recent progress of speech recognition technology is tremendous. Databases are the integrated information resources of various types of enterprises, and their values are increasing depending on the computerization of societies. However, in most cases, the database accesses are done by using the traditional keywords. This situation is deplorable since the society is now moving to mobile and ubiquitous one, where people are hard to use keyboard to obtain information. This research tried to resolve this problem by building a speech interface to a relational database system. We have implemented a system which consists of a Sun workstation, Oracle 7 a relational database management system, DS200 a speech recognition system, and FINE TALK a speech composition system. We have developed four software modules to combine such system components. The evaluation of the prototype was done, and it was revealed out that homonym problem should be resolved as well as how to accommodate incomplete Japanese sentence speeches. In this research, we have attached the first problem, and showed a solution by introducing the "instance keys, " which are the generalization of the traditional keys defined by using functional dependencies. It was shown that a more sophisticated speech interface can be realized by pursuing the proposed approach.