Study on Development of Language Learning System for Japanese Speech Using Acoustic Analysis and CD-ROM

我们利用声学分析和CD-ROM开发了两个计算机辅助日语语音学习系统（CALL）。这两个系统被称为LESSON/J（日语语音点播教育系统）和PAL/J（便携式日语高级学习系统）。点播网络系统已于1997年6月在万维网服务器上提供，网址为http：//sp.cis.iwate-u.ac.jp/sp/lesson/j/index.html。三年来，80个国家访问了该网络系统约60 000次。国外访问率约为70%。访问次数最多的前5个国家依次是美国、加拿大、意大利。澳大利亚和韩国。便携式系统由一台普通笔记本电脑、一个音频耳机和CD-ROM组成。便携式系统供外国学生在教室、学生家里和其他地方使用。一些学习者在教室里为自己的错误答案感到羞愧，但私下里却没有使用便携式系统。本研究亦利用声学分析方法，开发出日本口音发音评估系统及特殊发音评估系统。在特殊停顿系统中，利用语音识别技术中的动态规划方法自动估计学习者特殊莫拉的持续时间。然后，使用持续时间长度的新的归一化函数来计算特殊莫拉的质量水平。利用重音的相对位置和基频的衰减来自动识别重音的类型。

We have developed two systems of computer assisted language learning (CALL) for Japanese speech using acoustic analysis and CD-ROM. The systems are called as LESSON/J (Japanese Language Education System for Speech on an On-demand Network) and PAL/J (Portable Advanced Learning System for Japanese).The on-demand network system is now available on the World Wide Web server, the URL is http://sp.cis.iwate-u.ac.jp/sp/lesson/j/index.html, since in June 1997. The network system was accessed about 60,000 times from 80 countries for three years. The rate of the accesses from foreign countries is about 70%. The top 5 countries for accesses in order are U.S.A, Canada, Italy. Australia and Korea.The portable system is consisting of a general notebook computer, an audio headphone and CD-ROM. The portable system is used by foreign students in a lecture room, student's home and other places. Some learner is ashamed of his wrong answer in the lecture rooms, but is not in private using the portable system. It is one of advantages for the portable language learning system.The evaluation systems of pronunciation for Japanese accent and special morae are also developed using acoustic analysis method. In the system for special morae, the duration length of special mora of the learner was automatically estimated using a method of dynamic programming in the speech recognition technology. Then, quality level of the special mora is calculated using a new normalized function for the duration length. The type of accent is automatically recognized using relative position of accent and decay of fundamental frequency.