多位朝鲜语数字语音识别方法的研究  

Recognition of Multi-digit Korean Number Speech

作　　者：朴明姬[1] 芦世丹[1] 崔荣一[1] 

机构地区：[1]延边大学工学院计算机科学与技术系智能信息处理研究室,吉林延吉133002

出　　处：《延边大学学报（自然科学版）》2011年第4期337-341,共5页Journal of Yanbian University（Natural Science Edition）

摘　　要：针对朝鲜语数字串语音识别中单位符号容易被误判为数字的问题,提出了基于朝鲜语数字串结构特点的识别方法。首先,对事先采集好的语音样本提取特征并存储成数字模板和单位符模板;其次,将数字串测试样本切分成孤立的数字或单位符并提取特征参数;最后,根据朝鲜语数字串发音特点,即若前1个孤立词已被识别为数字,那么后1个被默认为单位符并且只与存储单位符号的模板进行匹配。实验结果表明,该方法不仅可以获得较高的正确识别率,同时提高了识别效率。According to the problem that Korean numbers speech recognition unit symbol is mistaken for a number, a recognition method based on characteristics of Korean digital string is proposed. First of all, the feature of speech samples were extracted and stored into a digital template and digits template; Second, cutting the string of numbers into an isolated symbol and extracting feature parameters; Finally, according to the characteristics of Korean pronunciation of the number string, that is, if the previous word has been identified as an isolated number, the unit is a default symbol for the next one and matching with only the unit template. Experimental results show that the method can not only obtain a higher recognition rate, but also improve the recognition efficiency.

关 键 词：语音识别 朝鲜语数字语音 美尔频率倒谱系数 动态时间规整 

分 类 号：O235[理学—运筹学与控制论]