An interactive way to acquire Internet documents for language model adaptation of speech recognition systems
In this paper, a new method for language model adaptation based on users feedback in the field of speech recognition is described. Different from other methods, the proposed method conducts corpus collection and language model adaptation in an interactive way. The user can input a small quantity of texts to describe the topic or the basic idea of the speech and evaluate some of the obtained texts as good or useless. The system can learn from the interaction information and acquire textual corpus which is more relevant to the topic of the speech. Experimental results show that for a given speech recognition system using this approach the recognition accuracy is increased by 7 percentage points compared to the same system using traditional adaptation method without interaction.
speech recognition corpus acquiring users feedback language model adaptation
Hong Zhang Xiangdong Wang Yueliang Qian Shouxun Lin
Institute of Computing Technology, Chinese Academy of Sciences Beijing, China, 100190
国际会议
杭州
英文
97-100
2011-08-26（万方平台首次上网日期，不代表论文的发表时间）