基于核Logistic回归的乐器音乐辨识

基于统计学习的音频分类具有理论基础坚实,实现机制简单等特点受到广泛关注并被很多音频分类系统所采用。本文对核Logistic回归算法(KLR)进行了深入分析,提出基于KLR的音频分类器设计方法,应用其解决同类型的乐器音乐分类问题。结合所采集的小提琴中提琴和大提琴的音乐信号样本进行了分类仿真试验,并与传统的Logistic回归(LR)和支持向量机(SVM)进行对比。结果表明,核Logistic回归模型具有较为优越的分类性能和非线性处理能力。

Musical Instrument Audio Identification Based on Kernel Logistic Regression

Audio classification based on statistical learning has attracted widespread attention and been widely used in audio classification system, because of better theoretical foundation and simple implementation mechanism. Based on exploration on theory about kernel logistic regression （KLR）, a novel approach for audio classifier is put forward with the help of KLR in this paper. It is used to handle music form the same type of musical instrument. Music signals are collected with violin, viola and cello, and all the signals are preprocessed to extract features. The processed samples are used in experiments, the classification performances are compared with traditional LR and Support Vector Machines （SVM）. Simulation results show that KLR performs better on classification accuracy and non-linear processing ability.