当前位置>主页 > 期刊在线 > 信息技术 >

信息技术2019年6期

基于数据压缩与改进的概率神经网络的贵州方言辨识
艾虎¹,李菲²
(1. 贵州警察学院 刑事技术系,贵州 贵阳 550005;2. 香港教育大学,香港 999077)

摘  要:为了判断犯罪嫌疑人的方言归属地,从而为案件的侦破提供重要线索,本研究从贵州6 个不同的地区采集到600份不同年龄和性别的语音样本,并提取梅尔频率倒谱系数MFCC,采用主成分分析与本研究所提出的数据压缩方法对MFCC 进行降维处理,得到用于概率神经网络训练的数据集,然后对概率神经网络进行改进,并构建贵州地区方言辨识模型。仿真结果表明,方言模型辨识结果与实际结果的相关系数R 为90%,该模型能有效地对贵州地区方言进行辨识。


关键词:汉语方言辨识;梅尔频率倒谱系数;主成分分析;概率神经网络



中图分类号:TP391.4         文献标识码:A        文章编号:2096-4706(2019)06-0005-05


Identification of Guizhou Dialect Based on Data Compression and
Improved Probabilistic Neural Network
AI Hu1,LI Fei2
(1.Department of Criminal Technology,Guizhou Police College,Guiyang 550005,China;
2.The Education University of Hong Kong,Hong Kong 999077,China)

Abstract:In order to judge the location of the suspect’s dialect,it provides important clues for the detection of the case. In this study,600 phonetic samples of different ages and sexes were collected from 6 different regions of Guizhou and the Mel frequency cepstrum coefficient MFCC was extracted from the samples. The Principal Component Analysis (PCA) and the data compression method proposed in this study are used to reduce the dimensionality of the MFCC to get the data set used in the training of probabilistic neural network. Then the probabilistic neural network is improved,and then it is used to construct the identification model of Guizhou dialect. The simulation results show that the correlation coefficient R between the dialect model identification result and the actual result is 90%. This model can effectively identify the dialects in Guizhou.

 Keywords:Chinese dialect identification;mel frequency cepstrum coefficients;principal component analysis;probabilisti probabilisticneural network


参考文献:

[1] BAKER W,EDDINGTON D,NAY L. Dialect identification:The effects of region of origin and amount of experience [J].American Speech,2009,84(1):48-71.

[2] 贾晶晶,顾明亮,朱恂,等. 基于流形学习与特征融合的汉语方言辨识 [J]. 计算机工程与应用,2015,51(7):233-237.

[3] 顾明亮,张世形,张浩,等. 基于联合多样性密度的汉语方言辨识 [J]. 计算机工程与应用,2016,52(10):161-166.

[4] 景亚鹏,郑骏,胡文心. 基于深层神经网络(DNN)的汉语方言种属语音识别 [J]. 华东师范大学学报(自然科学版),2014(1):60-67.

[5] 崔瑞莲,宋彦,蒋兵,等. 基于深度神经网络的语种识别[J]. 模式识别与人工智能,2015,28(12):1093-1099.

[6] 张毅,黎小松,罗元,等. 基于人耳听觉特性的语音识别预处理研究 [J]. 计算机仿真,2015,32(12):322-326.

[7] Pearson K. On lines and planes of closest fit to systems of points in space [J]. The London,Edinburgh,and Dublin Philosophical Magazine and Journal of Science,1901,2(6):559-572.

[8] Abdi H,Williams LJ. Principal component analysis [J]. Wiley Interdisciplinary Reviews Computational Statistics,2010,2(4):433-459.

[9] SPECHT DF. Probabilistic neural networks for classification,mapping,or associative memory [C]// Neural Networks,1988.,IEEE International Conference on. S.l.:s.n.,1988:525-532.

[10] Specht DF. Probabilistic neural networks [J]. Neural  Networks,1990,3(3):109-118.

[11] 董长虹.Matlab 神经网络与应用(第2 版) [M]. 北京:国防工业出版社,2007.


作者简介:艾虎(1974-),男,汉族,江西弋阳人,博士,副教授,研究方向:声音与图像。