利用背景知识提高web语音浏览中的识别精度的方法

Improve the Accuracy of Recognition in Web Speech Browsing Using Context Knowledge

语音识别的精度不够高一直是阻碍语音技术得以广泛应用的瓶颈,在具体的应用中充分利用背景知识是解决此问题的一种有效方法.在web语音浏览中,用户的语音输入为某个有限集的元素之一,本文利用这个特点,首先定义了一种文本字符串之间的相似度,利用相似度对识别引擎的识别结果进行后处理,进而给出更准确的识别结果.实验结果表明,采用这种方法,语音识别的正确率能够达到95％以上,为真正实现语音上网提供了有力支持.

The accuracy of speech recognition is still a bottleneck to baffle the application of speech technology. The accuracy of recognition may be improved greatly by using context knowledge efficiently. In web speech browsing, user's speech input is usually one element of a finite set. Based on these observations, this paper first defines a kind of similarity between two Chinese text strings, then processes the recognition results of engine to acquire more accurate results. Experiments show that our approach is mostly efficient: the accuracy is improved from less than 60% to more than 95%.