基于LVQ的普米语语谱图识别  

Primi Language Spectrogram Recognition Based on Learning Vector Quantization

作　　者：杨花 江涛[1] 和丽华 陈绍雄 潘文林[1] 

机构地区：[1]云南民族大学,数学与计算机科学学院,云南昆明

出　　处：《计算机科学与应用》2018年第12期1850-1856,共7页Computer Science and Application

基　　金：国家自然科学基金项目(61363022);云南民族大学研究生创新基金(2018YJCXS228);云南省教育厅科学研究基金项目(2017YJS056).

摘　　要：本文使用基于学习向量量化算法实现了对普米语语谱图的识别。该算法首先通过傅立叶变换绘制出每条普米语语料的语谱图;再次,提取每张语谱图的图像灰度共生矩构建特征向量;最后,运用学习向量量化实现对普米语孤立词的分类。基于学习向量量化算法的分类准确率达到了80.0%。In this paper, we proposed a Primi language speech recognition algorithm based on Learning Vector Quantization (LVQ). Firstly, the algorithm uses the Fourier transform to plot each Primi language pragmatics;secondly, it extracts the gray level cooccurrence moments of each spectrogram to construct feature vectors;and finally, it uses learning vector quantization to realize the classification of Primi isolated words. The classification accuracy based on the Learning Vector Quantization (LVQ) is 80.0%.

关 键 词：普米语 灰度共生矩阵 学习向量量化(LVQ) 

分 类 号：TP39[自动化与计算机技术—计算机应用技术]