端点检测在开心果碰撞声分类研究中的应用  

Endpoint detection in pistachio sorting research based on impact sound

作　　者：王涛[1] 冯涛[1] 钟晨玉 

机构地区：[1]北京工商大学材料与机械工程学院,北京100048

出　　处：《计算机应用》2016年第A01期196-200,共5页journal of Computer Applications

摘　　要：针对基于碰撞声的开心果分类研究中,用手工方式从实验录制的音频文件中截取样本信号工作效率低的问题,提出了一种使用基于能熵比的语音端点检测算法自动获取样本信号的方法。首先,分析音频文件中碰撞声信号与背景噪声信号在能量和谱熵参数上的差异特点,推导使用能熵比作为端点检测算法参数的可行性;其次,根据实验中开心果与地面初次碰撞产生目标样本信号后可能会出现因为开心果的反复弹起而产生紧随目标样本信号的干扰信号的情况,对目标样本信号与干扰信号的分布特点进行分析;最后以实验中产生相邻目标样本信号之间的最小时间差作为排除干扰信号的参数,改进了现有的端点检测算法。实验录制了共51个通道的音频文件,在Matlab中使用基于能熵比的端点检测算法通过编程自动获取了969个样本信号,其中样本信号的有效率为99.38%、遗失率为0.62%。相对于手工截取样本信号的方法,所提方法执行速度快,并且获得的样本信号具有幅值归一化和长度统一的优点,这有助于加快基于碰撞声的开心果分类研究进程,同时,该方法在适当修改参数后可以应用到其他研究项目的样本信号获取上。A sample signal extracting method of endpoint detection for pistachio sorting based on the impact sound was proposed to solve inefficiency of the artificial signal extracting method from recorded sound file.Firstly,the differences of energy and entropy between impact sound and background noise signal were analyzed in the sound file for the feasibility speculating of using energy to entropy ratio as the parameter of endpoint dictation algorithm.Secondly,according to the disturbed signal because of the repeatedly rebound of the pistachio adjacently to the target sample signal with the first impact,the distribution characteristics of the target signal and disturbed signal were analyzed.Finally,the minimum difference of time among adjacent sample signals was used as the parameter to improve endpoint detection algorithm for excluding disturbed signal.In the experiment,files with total of 51 channels were recorded,and 969 sample signals were acquired by Matlab programming,in which the efficiency ratio of the sample signals was 99.38% and the losing ratio was 0.62%.In comparison with the artificial signal acquiring method,the proposed method can execute quickly and it has the advantage of getting normalized signal of the same length,so it can accelerate the research progress of pistachio sorting based on impact sound,at the same time,the method can also be used in other research project of the sample signal acquiring after adjusting the relevant parameters.

关 键 词：端点检测 能熵比 开心果 碰撞声 信号获取 

分 类 号：TN912.34[电子电信—通信与信息系统]