字音转换策略介绍及性能代价评估

Grapheme-to-Phoneme Conversion method and cost performance analysis

许多面向应用的方法研究主要关注的是该方法的技术的性能,而相应的软件开发则要顾及整个系统的性能代价比,即不但要考虑到技术的性能,还要考虑到系统开发过程中的人力投入以及系统实现运行时的存储占用等方面的因素。这就对学术研究提出了另外一个要求,即对方法的实现做出性能代价的评估。探讨了字音转换技术,采用高频默认方法和语境约束方法,给出在不同分词词表背景下的准确率、存储空间占用及人力消耗等统计数据,从而为标音软件的开发提供一定的参考。

In computational linguistics,methods research for application pays main attention to the precision,or in other words, the performance of the technology,while the corresponding software development has to weigh and balance. It is necessary for a developer to take into account the cost of manpower and storage as well as the performance. So research papers should not only introduce a method and its result but also make a cost performance analysis. Taking the example of the technology of Grapheme-to-Phoneme Conversion,this paper provides some statistical data about precision,storage cost and manpower cost of two methods,high frequency default method and context restriction method. These reference datas are useful for the development of automatic phonetic notation system.