摘要
作为语言学习的重要组成部分之一,发音学习是人与人之间互相交流的基础。该文分析了传统语音教学及其评价方式的不足,以梅尔频率倒谱系数作为参数进行特征提取,用隐马尔可夫声学模型和音素后验概率算法将测试语音与标准语音进行比对,从而找出二者之间的差异程度,通过评分机制得到分数。实验结果表明,该系统能提高说话者学习语音的积极性,其评分结果与教师的主观感觉相一致,具有一定的实用性和可操作性。
As one of the important components of language learning, the pronunciation learning is the basis for people to communicate with others. This paper analyzes the shortage of the traditional phonetics teaching and assessment methods, and extracts Mel Frequency Cepstral Coefficients (MFCC) in feature selection. It compares the original speech with the standard one and finds out the differences between them by using the HMM acoustic model and the Viterbi algorithm. Then the evaluation mechanism makes the appraisal. The experimental results indicate that the assessment made by the system has a good consistence with that of a teacher and the system which can increase the enthusiasm of studying has certain practicability and operability.
出处
《绵阳师范学院学报》
2012年第5期88-92,共5页
Journal of Mianyang Teachers' College
基金
福建省教育厅B类项目(JB11265)
福建师范大学福清分校校级立项科研项目(KY2011022)
关键词
语音评分
特征提取
梅尔频率倒谱系数
隐马尔可夫模型
维特比算法
Speech evaluation
feature extraction
Mel - Frequency Cepstral Coefficients ( MFCC )
Hidden Markov Model (HMM)
Viterbi algorithm