CAJ | 학술논문

语种识别就是计算机能够识别出语音段所属的语言的过程。借鉴其它语种识别库的建立，设计了三种少数民族语的文本语料，并构建了语种识别库。在此语种识别库的基础上，提取了加权MFCC语音特征参数，采用高斯混合模型来训练语种识别模型，并得出识别率。通过实验得出，在现有的样本下16阶的GMM得出的识别率最高，达到了75%以上。
어충식별취시계산궤능구식별출어음단소속적어언적과정。차감기타어충식별고적건립，설계료삼충소수민족어적문본어료，병구건료어충식별고。재차어충식별고적기출상，제취료가권MFCC어음특정삼수，채용고사혼합모형래훈련어충식별모형，병득출식별솔。통과실험득출，재현유적양본하16계적GMM득출적식별솔최고，체도료75%이상。
Language identification is a process that the computer can identify the language of its phonetic segment. Learning from the establishment of other language identification library, this paper designs the three minority-language text corpus and builds the language identification library. On the basis of this language identification library, it extracts the weighted MFCC speech charac-teristic parameters, trains the language identification model using Gaussian mixture model, and obtains the recognition rate. Through the experiment, the recognition rate of the 16-step GMM is highest in the existing sample, reaching more than 75%.