Development for speech interface for form -based in formation access services on Web

虽然已经开发了一些用于访问Web资源的语音接口系统，但它们仅限于访问某些特定内容，并且它们没有为WWW上的任意信息检索服务提供通用接口。我们提出了一个交互式语音用户界面系统，它可以应用于许多基于形式的信息检索服务的WVVW。特别是，我们的系统是基于客户端-服务器，一个Web代理为中心的体系结构，并采用了信息提取和HTML文档的语言处理提供了一个通用的接口，许多基于表单的WWW内容。我们还进行了一些实验，由12名受试者在不同的使用条件下的可用性进行比较。结果表明，在语音识别性能理想的情况下，该系统获得了与笔触输入法相当的、更高的可用性指标，可以预期达到预期的效果。 ...更多信息 s或优越于笔触摸界面的可用性，因为他们的使用条件接近现实的PDA使用条件。一个基于语音识别的姓名输入接口，使用基于音节的N-gram和单词字典，这是经常需要输入到基于表单的网页。用户首先说出一个名字，然后选择正确的单词/音节从语音识别获得的单词/音节候选人的笔触摸。由于词汇量大，很难准确识别姓名发音，因此系统使用基于音节的N-gram的连续音节识别和包含频繁词的字典的孤立词识别。用户可以以82- 86%的比率从单词候选者或音节序列候选者中找到正确答案，并且可以以94-96%的比率从音节网格中选择音节来输入正确姓名。一些受试者使用了这个界面，并认为它很有用。少

While some speech interface systems have been developed for accessing Web resources, they are limited for accessing some specific contents and they don't provide a universal interface for arbitrary information retrieval services on the WWW. We propose an interactive speech user interface system, which could be applied to many form-based information retrieval services of the WVVW. In particular, our system was implemented based on a client-server, a Web proxy-centered architecture and employed an information extraction and language processing of HTML documents for providing a general-purpose interface for many form-based WWW contents. We also performed some experiments by 12 subjects for the comparison of the usability under different usage conditions. As a result, the proposed system attained comparative and higher expected usability measures over the pen-touch input method under the condition of an ideal speech recognition performance, and could be expected to achieve the effectivenes … More s or the superiority over a pen touch-only interface in terms of the usability as their usage condition approaches to a realistic PDA usage condition.We also proposed an. interface for a name input based on speech recognition using syllable-based N-gram and a word dictionary, which was frequently required to input into form-based web pages. User first utters a name and then chooses the correct word/syllables by pen touch from word/syllable candidates which were obtained from speech recognition. Name utterance is hard to recognize accurately because of the large vocabulary size, so the system uses continuous syllable recognition with syllable-based N-gram and isolated word recognition with a dictionary containing frequent words. The user can find the correct the answer from word candidates or syllable sequence candidates at a rate of 82-86%, and can input correct name at a rate of 94-96% with syllable selection from the syllable lattice. Some subjects used this interface and felt that it was useful. Less