幅度调制信息对乐器识别的影响

发布时间：2018-03-22 13:16

本文选题：幅度调制　切入点：自动识别　出处：《声学学报》2015年02期 　论文类型：期刊论文

【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。
[Abstract]:The effects of amplitude modulation (AM) information on the recognition of human musical instruments are studied through the experiment of automatic recognition of musical instruments. The steps are as follows: according to the auditory model, the AM information in several frequency bands of musical instrument sound signals is extracted. Then, based on the calculated statistical characteristics of AM information, a pair by pair comparison support vector machine (SVM) method is used to recognize musical instrument sounds. The increase of frequency band number is helpful to improve the recognition effect, but the effect from 16 band to 32 band tends to be stable, and different AM extraction methods will also affect the recognition results. The result of automatic recognition is higher than that of human with similar AM information. This automatic recognition system is a cochlear implant or vocoder. The instrument recognition of the simulated sound model provides a computational model. It has reference value for the experiment and training of cochlear instrument recognition.
【作者单位】：中国科学院声学研究所东海研究站;深圳大学信息工程学院;
【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
【分类号】：J62

【共引文献】