基于过零率及频谱的语音音乐分类算法

Speech music classification algorithm based on zero-crossing rate and spectrum

针对语音/音乐分类过程中由于特征参数提取过多造成的维数灾难及分类准确率有待提高等问题,提出了一种基于过零率及频谱的语音/音乐分类算法.该算法在对语音及音乐2类信号进行端点检测及分段预处理后,结合每一音频段的过零率和频谱幅值特性进行分类识别处理,最后通过计算被判别为语音或音乐的概率实现分类.实验结果表明,此算法在音频分类中较同样最多只提2个音频特征且未用分类器算法的准确率平均提高约7.9%,较提取多个音频特征且采用分类器算法的准确率平均提高约5.7%.证明了该算法不仅计算量小,且分类准确率也有所提高.

Due to the problems of dimensionality disaster caused by excessive extraction of feature parameters and the need to improve the classification accuracy in speech/music classification process,this paper proposes a speech/music classification algorithm based on zero-crossing rate and spectrum.After endpoint detection and segmented preprocessing of speech and music signals,the algorithm classifies and recognizes each audio segment by combining the zero-crossing rate and spectral amplitude characteristics,and finally realizes the classification by calculating the probability of being identified as speech or music.Experimental results show that the accuracy of this algorithm in audio classification is about 7.9%higher on average than that of the same algorithms which only mention two audio features at most and do not use the classifier,and about 5.7%higher than that of the algorithms which extract multiple audio features and use the classifier.It proves that this algorithm not only has a small amount of calculation,but also improves the classification accuracy.