基于MFCC和HMM的音乐分类方法研究

Research of Music Classification Based on MFCC Feature and HMM Model

采用基于Mel倒谱系数特征的隐马尔可夫模型对音乐进行分类.对音乐通过有监督的学习方式进行聚类,分类时将测试样本归入似然值最大的类别,对同一音频抽取若干样本,对样本识别结果采用投票法判定该音频的音乐类别,使分类的准确率得到进一步的提高.仿真实验对4种分类器在有干扰和无干扰的环境下的分类性能进行了比较,实验结果表明该方法具有更好的抗干扰能力和正确率.

In this paper, we use hidden Markov Model based on Mel-frequency cepstrum coefficients to classify the mu- sic. Classification divides the test samples into categories according to the largest likelihood value. We draw several sam- ples of the same music frequency, identify the results of the samples using the voting method, and thus determine the category of the audio to further improve classification accuracy. We make a simulation experiment to compare the per- formance of four different classifications in the environments of disturbance and nodistabance. The results show that HMM classification has more advantages on performance and is less sensitive to disturbance.