基于MFCC和HMM的腭裂语音辅音省略识别算法  

Recognition algorithm of consonants omission for people with cleft palate based on MFCC and HMM

作　　者：袁亚南[1] 何凌[1] 龚晓峰[1] 尹恒[2] 李杨[2] 

机构地区：[1]四川大学电气信息学院,四川成都610041 [2]四川大学华西口腔医院,四川成都610041

出　　处：《计算机工程与设计》2014年第2期615-619,共5页Computer Engineering and Design

基　　金：国家自然科学基金青年基金项目(30900391)

摘　　要：为了弥补国内外在腭裂语音辅音发音错误识别方面的漏缺,以及为临床腭裂语音类型的诊断提供一种非主观的辅助措施,提出了一种基于语音识别系统的腭裂语音辅音省略识别算法。对腭裂语音辅音发音错误特点进行了语谱图上的能量研究,建立了改进参数的基于美尔频率倒谱系数和隐马尔科夫模型的腭裂语音识别系统。实验结果表明,该语音辅音省略识别算法取得了较高的正确率,实现了对腭裂语音清晰度的自动量化评估,可以用于临床辅助诊断。To compensate for the gaps of mispronunciation of consonants by speakers of cleft palate recognition, and to provide an non-subjective complementary measures for the clinical diagnosis of cleft palate speech types, a recognition algorithm of conso- nants deletion based on speech recognition system is proposed. cleft palate on the energy which is reflected by the spectrogram, By studying the features of consonants deletion by speakers of an improved automatic recognition system of consonants deletion based on MFCC and HMM is established. Experimental results show that the proposed automatic recognition algorithm of mis- pronunciation of consonants achieves a high accuracy and realizes the automatic quantitative analysis of intelligibility of speech by speakers with cleft palate, and can be used for the clinical diagnosis.

关 键 词：腭裂语音 辅音省略 美尔频率倒谱系数 隐马尔科夫模型 语音识别 

分 类 号：TN912.34[电子电信—通信与信息系统]