一种新的关键词确认方法

本文提出了一种新的基于模型距离矩阵的关键词确认算法,并给出模型距离的定义及其训练方法,利用模型相对距离矩阵对语音识别结果进行确认.对于关键词库较大的关键词检出系统,通过对关键词分段,得到扩展的模型距离矩阵确认算法,使得大词表确认问题得到很好的解决,并能够获得和小词表系统一样的确认效果.为了对关键词库进行方便的操作,模型距离矩阵的更新算法使得用户可以很方便地修改关键词库内的关键词,而不必重新训练整个模型距离矩阵.

A Novel Keyword Verification Algorithm

This paper proposes a novel utterance verification (UV) algorithm based on model-distance matrix.It gives the definition of model-distance and training algorithm for model-distance matrix which is used to verify the recognition result.By using expanded model-distance matrix verification algorithm,it makes impossible to verify the result of a big keyword spotting system,and gets the same performance as on a small one.An updating algorithm to modify model-distance matrix is also introduced when vocabulary is modified.