幅度调制信息对乐器识别的影响  

The effects of amplitude modulation cues on musical instrument recognition

作　　者：孟庆林[1,2] 原猛[1] 夏洋[1] 冯海泓[1] 

机构地区：[1]中国科学院声学研究所东海研究站,上海200032 [2]深圳大学信息工程学院,深圳518060

出　　处：《声学学报》2015年第2期300-306,共7页Acta Acustica

基　　金：国家自然科学基金青年科学基金(11104316);上海市自然科学基金(11ZR1446000);中国科学院声学研究所所长择优基金(Y154221701);深圳市科技研发资金(JCYJ20120614085245889);深圳市重点实验室提升项目(CXB201105060068A)资助

摘　　要：通过乐器音自动识别实验研究了幅度调制(amplitude modulation,AM)信息对于人类乐器识别的影响。具体步骤为:依据听觉模型,提取乐器音信号中若干频带中的AM信息,再基于所得到的AM信息计算统计学特征,采用逐对比较的支持向量机法进行乐器音的机器识别。采用了5种分频带数目(2,4,8,16和32)和4种AM计算方法。结果表明,频带数的增加有助于识别效果的提高,但从16频带到32频带效果趋近平稳;不同的AM提取方法也会对识别结果产生影响,其中解析信号法产生的AM信息提供了最好的乐器识别效果。分析发现自动识别结果高于采用相似的AM信息的人类识别结果。该自动识别系统为人工耳蜗或声码器仿真声模型的乐器识别提供了一个计算模型,对人工耳蜗乐器识别实验和训练具有参考价值。The effects of amplitude modulation （AM） cues on human musical instrument recognition were investigated. An automatic musical instrument recognition experiment was carried out. The following stages were manipulated： calculation of the statistical features from AM of the musical signals extracted based on the auditory model; a support vector machine with pairwise classification strategy was used as the classifier. Totally five band conditions （2, 4, 8, 16 and 32） and four AM extraction methods were evaluated. Results showed that recognition performance improved with the increasing of the band number, and 16 bands were sufficient for reaching asymptotic performance. The AM extraction methods also had significant effects on the performance. The analytic signal based method showed better performance than the rectification-low-pass -filter methods. The automatic recognition system showed better performance than human subjects using similar AM information. The automatic recognition system provided a computational model for predicting the performance of musical instrument recognition with cochlear implants and vocoder simulations and can be used as a reference in relevant experiments and training about musical instrument recognition with cochlear implants.

关 键 词：幅度调制 自动识别 人工耳蜗 计算统计学 听觉模型 声码器 识别实验 机器识别 解析信号 频带信号 

分 类 号：J62[艺术—音乐]