基于小波特征的维吾尔语方言识别  

Uyghur Dialect Recognition Based on Wavelet Features

作　　者：孙杰 王斌 王宏 吾守尔·斯拉木[4] SUN Jie;WANG Bin;WANG Hong;Wushour Silamu(School of Air Transportation,Changji University,Changji 831100,Xinjiang,China;Xinjiang Electric Power Research Institute of State Grid,Urumqi 830000,China;Department of Information Engineering,Xinjiang Institute of Technology,Aksu 843100,Xinjiang,China;College of Information Science and Engineering,Xinjiang University,Urumqi 830046,China)

机构地区：[1]昌吉学院航空学院,新疆昌吉831100 [2]国网新疆电力有限公司电力科学研究院,乌鲁木齐830000 [3]新疆理工学院信息工程学院,新疆阿克苏843100 [4]新疆大学信息科学与工程学院,乌鲁木齐830046

出　　处：《实验室研究与探索》2023年第4期148-152,共5页Research and Exploration In Laboratory

基　　金：新疆自然科学基金面上项目(2022D01C03)。

摘　　要：为进一步提高维吾尔语方言识别准确率,从特征域出发提出基于小波包变换的方言语音声学参数提取算法,利用小波包变换对语音信号进行精细分解,使用Bark滤波器提取低频子带和高频子带系数与人耳基底膜的不同部分对应,用局部线性嵌入的流形学习算法进行降维,提升信号的分类性能。分别以i-vector和x-vector两个主流系统进行方言识别实验,结果表明,小波特征声学参数的方言识别准确率比MFCC特征和FBank特征识别准确率均有提高。In order to further improve the accuracy of Uyghur dialect recognition,a dialect speech acoustic parameter extraction algorithm based on wavelet packet transform is proposed from the feature domain.The algorithm is used wavelet packet transformation to finely decompose the speech signal,so that the low-frequency sub-band and high-frequency sub-band correspond to different parts of human ear basement membrane.The algorithm uses manifold learning to reduce the dimension and improve the classification performance.The experimental results of two baseline dialect recognition systems of i-vectorandx-vector show that the dialect recognition accuracy of wavelet feature acoustic parameters is higher than that of MFCC feature and FBank feature.

关 键 词：小波包变换 流形降维 x-vector向量 维吾尔语方言 识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]