摘要： Mel频率倒谱系数(MFCC)等传统声学特征不能精确地体现出不同乐器信号间的差别。为此，提出一种基于匹配追踪(MP)稀疏分解原子参数的乐器分类方法。利用MP算法提取各类乐器信号的稀疏分解原子，将得到的原子参数作为特征，通过支持向量机进行分类。实验结果表明，该方法的分类正确率达到89.17%，相对于MFCC提高了17.37%。

关键词: 乐器分类, 参数提取, 稀疏分, 匹配追踪算法, 原子参数, 支持向量机

Abstract: In order to solve the problem that the traditional features such as Mel Frequency Cepstral Coefficients(MFCC) can not reflect the difference between instruments accurately and intuitively. Musical instrument classification based on atomic parameters with Matching Pursuit(MP) sparse decomposition is proposed. In the simulation experiments, the atomic parameters are extracted with MP method from the instrument signals as the features to be classified using Support Vector Machine(SVM). Simulation results show that the musical instrument classification based on sparse decomposition atomic parameters with MP is effective and the recognition rate reaches 89.17%, 17.37% higher than MFCC.

Key words: musical instrument classification, parameter extraction, sparse decomposition, Matching Pursuit(MP) algorithm, atomic parameter, Support Vector Machine(SVM)

中图分类号: 

• TN911.72