基于多维特征联合的鸟类鸣声识别方法研究  

Research of birds call recognition method based on multi-feature fusion

作　　者：陈海兰[1] 孙海信[2] 齐洁[2] 高春仙[2] 颜佳泉[2] 

机构地区：[1]集美大学理学院物理系,厦门361021 [2]厦门大学信息科学与技术学院通信工程系,厦门361005

出　　处：《南京大学学报（自然科学版）》2015年第6期1234-1239,共6页Journal of Nanjing University（Natural Science）

摘　　要：湿地不但具有丰富的资源,还有巨大的环境调节功能和生态效益.随着人类社会的发展,湿地生态保护日益受到人们的重视,其中鸟类监测经常作为湿地环境质量的有效指标.目前国内外对于鸟类监测技术主要通过大量人力统计手段,严重浪费宝贵的人力资源.基于此种状况,且传统的语音识别方法主要采用基于时域特征或频域特征的识别方法,根据不同种鸟类鸣声的特点,提出一种将不同音节长度特征与多段式平均频谱法相结合的多维特征联合的鸟类鸣声识别方法.实验结果发现,将时域音节长度特征分类与频域多段式平均频谱法相结合的识别方法比只采用多段式平均频谱法的识别方法将辨识正确率由92.38%提高到100%,实验采用了湿地常见的17种鸟类,辨识效果非常准确,表明本文所提出的多特征结合识别方法对于提高识别正确率具有重要的参考价值.Wetland has rich resources,huge environmental regulation function and ecological benefit.With the development of human society,the problem of wetland ecological protection is being paid more and more attention, bird monitoring is often used as an effective indicator for the evaluation of wetlands’environmental quality.At home and abroad,bird monitoring technology mainly takes a large number of manpower statistics method,and this leads to a serious waste of valuabale human resources.Based on this condition,the traditional method of speech recognition took the method of exacting the time domain characteristics or the frequency domain characteristics,and according to the characteristics of bird call,this paper presents a method which combining the syllables’length and multi segment average spectrum.The experimental result shows that comparing the method of combining characteristics of syllable length and multi segment average spectrum with the method of using only characteristic of multi segment average spectrum,the recognition rate is increased from 92.38% to 100%,the experiment adopted 17 kinds of common wetland birds,the identification effect is very accurate which indicates that the multi feature recognition method proposed in this paper has an important reference value for improving the accuracy of recognition.

关 键 词：语音辨识 模板匹配 短时傅立叶变换 多段式平均频谱法 

分 类 号：TN912[电子电信—通信与信息系统]