【作者】 李楠；

【导师】 王双维；

【作者基本信息】 东北师范大学 ， 电路与系统， 2014， 硕士

【摘要】 语言是不仅人类沟通的重要桥梁，也是蝙蝠进行沟通的重要工具。不同国家和地区的语言有所不同，我们通常称之为方言，然而在蝙蝠界也是如此，不同种群和地域的蝙蝠其发声方式也是不同的。本文主要内容就是关于蝙蝠“方言”的研究，以马铁菊头蝠为例，对其声音信号不同类型进行区分，从而简单了解蝙蝠的语义。马铁菊头蝠的声音样本分为五类，由生命科学学院进行分类与提供，分别是AFM、BNBs、DFM、SFM和UFM，每一类表示同种含义，且声音样本是来自全国的不同地区。本文在前人研究人类语音的基础上，对蝙蝠的声音进行分析，提出了蝙蝠“方言”信号的语谱图图像处理的方法。根据语谱图固有的特点，应用数学形态学图像处理技术对马铁菊头蝠五种声音类型进行特征提取，将语音识别技术与图像处理技术相结合，从而达到区分蝙蝠“方言”的目的。本文主要是通过MATLAB7.0软件对算法进行研究、编程、仿真和实现的。处理过程如下：首先，把录制好的蝙蝠“方言”信号用Cool Edit Pro2.1软件进行预处理，对其量化标准进行统一，生成的语谱图图像用抓图软件进行抓图；其次，应用直方图图像处理技术对蝙蝠“方言”语谱图进行处理；再次，应用数学形态学图像处理技术对蝙蝠“方言”语谱图进行二值化、形态学滤波和骨架化处理；最后，对蝙蝠“方言”信号的骨架化语谱图进行角度处理，用BP神经网络对实验结果进行检测，判断此结果是否合理。更多还原

【Abstract】 Language is not only an important bridge of human communication, but also animportant communicate tool to bat. Different countries and regions of the language isdifferent, we call it dialect, however, the same as the bat community. The sound stylesof different populations and geographical bat are different. The key of my paper isabout the research of the bat “dialect” we take the greater horseshoe bat as an example,and distinguish between the different types of sound signals so that we simplyunderstand the semantics of the bat.The greater horseshoe bat sound samples are divided into five categories,provided by the College of Life Sciences, and classify them, respectively, AFM,BNBs, DFM, SFM and UFM, each class represents the same kind of meaning, and thesound samples are in different regions from China. In this paper, we analyze the soundof bat based on the previous studies on the human speech, and put forward the bat"dialects" signal spectrogram method for image processing. According to the inherentcharacteristics of spectrogram, we do the feature extraction of the greater horseshoebat five sound types by applying the mathematical morphology image processingtechnology, combining the speech recognition technology and image processingtechnology, so as to distinguish “dialect” purposes.We study on the algorithm by introducing the MATLAB7.0software. Theprocess is as follows. The algorithm firstly introduces the Cool Edit Pro2.1softwarefor recorded “dialect" bat signal preprocessing, unifies the quantitative standard,achieves the resulting spectrogram image by capture software. Secondly, process thebat "dialect" spectrogram by applying the histogram image processing technology.Thirdly, do the binarization, morphological filtering and skeleton processing by usingthe mathematical morphological image processing technology. Finally, do the angleprocessing about the results of the bat "dialect" signals spectrogram. To test theexperimental results and determine rationality, the BP neural network is detected.更多还原