一种基于随机段的固定音频检索方法

Method of specific audio retrieval based on randomly segment

在固定音频检索的整体检索方法中,当检索目标较长时,检索时间会变得很长。为了减小检索时间,提出了一种基于随机段的音频检索方法。把整个检索过程分成随机段检索和整体匹配两个阶段:随机段检索是从参考模板中随机选择一段(随机段)作为检索目标进行检索;整体匹配是在随机段检索出的基础上,判断潜在目标信号是否为参考模板。把这种随机检索的方法应用到计算特征距离和直方图交集方法中,结果证明该检索方法的准确率可以达到90%以上,而且平均检索时间可以降低到随机段与参考模板的比值和整体检索时间的积。

This paper proposed a specific audio retrieval method based on random segment in order to decrease the retrieval time for relatively long object in the total retrieval of audio retrieval. The whole retrieval process was composed of random segment retrieval and total matching： the first was to select a segment from template model as object to retrieve in stored signal and the second was to judge whether the potential object signal was the template model based on random segment. Then using this method in computing feature distance directly and histogram intersection retrieval, the experimental results show retrieval accuracy over 90% and average retrieval time declined to the ratio of random segment to template model multiplying total retrieval time.