基于最小二乘支持向量机的乐器音乐分类

提出使用最小二乘支持向量机LS—SVM（Least Squares Support Vector Machines）算法进行乐器音乐分类，从而实现乐器的辩识。在对Ls—sVM理论进行深入探讨的基础上，选择乐器音乐clip作为样本，进行特征提取，提取的特征包括频谱特征，短时自相关系数和MFCC等，然后用最小二乘支持向量机算法进行分类。对古琴、古筝、箜篌和琵琶音乐采取样本进行仿真实验，求得分类准确率和运行时间，同时使用逻辑回归（Logistic Regression）算法进行对比试验，其中最小二乘支持向量机和逻辑回归分类的准确率分别为96．5％和92．5％，且LS—SVM的运行时间比Logist的少。实验结果表明最小二乘支持向量机具有更为优越的分类性能和非线性处理能力，可以推广用于解决其它实际分类问题。

Classification of Music Instruments Based on LS-SVM

The algorithm of least squares support vector machines （ISSVM） is proposed for classification of music instruments to recognize them. The corresponding music clips are chosen as experimental samples, and feature selection including spectrum features, short-time auto-correlation coefficients, MFCC, etc., is implemented, and then LS- SVM is used in the categorization. The simulating experiments of four traditional musical instruments are conducted to obtain classifying accuracy and running time. Meanwhile, logistic regression is also used in experiments as comparison. Accuracy rate of KS - SVM and logistic regression are respectively 96.5 % and 92.5 %. Running time of KS - SVM is less than that of Logist. The results demonstrate that LS - SVM possesses better performance, and can be generalized to cope with other practical classification problems.