基于SOFMNN模型的新疆方言语音识别研究  

Research on Speech Recognition of Xinjiang Dialect Based on SOFMNN Model

作　　者：刘彪 房锐林 邓美娟 赵文文 LIU Biao;FANG Ruilin;DENG Meijuan;ZHAO Wenwen(No.69230 Troops of PLA,Wusu 833000;Lanzhou University,Lanzhou 730000)

机构地区：[1]解放军69230部队 [2]兰州大学

出　　处：《计算机与数字工程》2018年第7期1405-1409,共5页Computer & Digital Engineering

摘　　要：针对新疆方言的特点,建立自组织特征映射神经网络(Self-Organizing Feature Mapping Neural Network,SOFMNN)模型,对经过梅尔倒谱系数(Mel-scale Frequency Cepstral Coefficient,MFCC)量化的语音特征进行训练,实现对新疆方言的语音识别。实验结果表明,采用该方法比采用传统隐马尔可夫模型的方法在识别正确率上占有明显优势。Based on the characteristics of Xinjiang dialect,a self-Organizing Feature Mapping Neural Network(SOFMNN)model is established to quantify the speech features of Mel-scale Frequency Cepstral Coefficient(MFCC)training and to achieve the Xinjiang dialect speech recognition.The experimental results show that the proposed method has a significant advantage in identifying the correctness rate by using the traditional hidden Markov model.

关 键 词：SOFMNN 新疆方言 语音识别 神经网络 

分 类 号：TP312[自动化与计算机技术—计算机软件与理论]