摘要： 根据音频文件数据量大、数据间存在一定相关性的特点，提出一种基于K-L距离的两步固定音频检索方法。该方法采用基于可变门限的直方图检索方法快速筛选出相似度较高的语音文件，利用特征矩阵的K-L距离对剩余语音进行精确比较，取得较好的效果。实验结果证明，该方法能使检索准确率达到90%左右。

关键词: 固定音频检索, 过零率, 直方图, 美尔频率倒谱系数, K-L距离

Abstract: Due to the huge amount of audio data, and some relation among them, this paper proposes a two-stage specific audio retrieval method based on K-L Distance. The method uses histogram retrieval method based on variable threshold to choose audio file of high similarity, compares precisely with residual audio using K-L distance of feature matrix, and obtains good effect. Experimental results show that the retrieval accuracy is over 90%.

Key words: specific audio retrieval, Zero Crossing Rate(ZCR), histogram, Mel Frequency Cepstral Coefficient(MFCC), K-L distance

中图分类号: 

• TP18