徽州方言孤立字识别的实现  

The Implementation of Isolated Word Recognition of Huizhou Dialect

作　　者：石瑛[1] 王雪飞[2] 汪欣[2] 沈来信[1] 

机构地区：[1]黄山学院机电与信息工程学院,安徽黄山245021 [2]黄山学院现代教育中心,安徽黄山245021

出　　处：《蚌埠学院学报》2014年第2期9-12,20,共5页Journal of Bengbu University

基　　金：安徽省教育厅自然科学研究项目(KJ2011B164);黄山学院教学研究项目(2009JXYJ10);安徽省文物局文化遗产保护领域科学和技术研究课题(2013-YB-HT-015)

摘　　要：徽州方言和普通话差异非常大,用声学研究的方法对徽州方言进行识别,先把休宁话数字语音采样后,将过零率作为参数进行端点检测,提取美尔频率倒谱系数(MFCC)声学特征后,把动态时间规整(DTW)作为识别的模型,对休宁话的孤立字进行识别,平均识别率较低。然后将短时能量也作为参数进行端点检测,把MFCC特征的一阶导数也作为识别参数,把DTW识别算法进一步改进,平均识别率有所提高。Huizhou dialect is a regional dialect of great differences compared with Mandarin .By the meth-od of acoustic research ,the recognition of Huizhou dialect was studied .Xiuning dialect was sampled firstly and the zero rate was used as a parameter for endpoint detection .After Mel Frequency Cepstrum Coeffi-cient（MFCC）acoustic features was extracted,the Dynamic Time Warping（DTW） act as the identification model to recognizing isolated words of Xiuning dialect .But the average recognition rate is low .Then the short-time energy used as a parameter for endpoint detection and the derivative of the MFCC feature is al-so as recognition parameters .The recognition algorithm of DTW is further improved .The average recogni-tion rate is increased at last .

关 键 词：美尔频率倒谱系数 动态时间规整 端点检测 

分 类 号：H102[语言文字—汉语] TN912[电子电信—通信与信息系统]