腭裂语音高鼻音等级自动识别算法研究
Automatic Hypernasal Detection Based on Acoustic Analysis in Cleft Palate Speech
作者：何凌(四川大学 电气信息学院)；袁亚南(四川大学 电气信息学院)；尹恒(四川大学 华西口腔医院)；张桠童(四川大学 电气信息学院)；张劲(四川大学 电气信息学院)；刘奇(四川大学 电气信息学院)；李杨(四川大学 华西口腔医院)
Author：He Ling(School of Electrical Eng. and Info.,Sichuan Univ.)；Yuan Ya'nan(School of Electrical Eng. and Info.,Sichuan Univ.)；Yin Heng(West China Hospital of Stomatology,Sichuan Univ.)；Zhang Yatong(School of Electrical Eng. and Info.,Sichuan Univ.)；Zhang Jing(School of Electrical Eng. and Info.,Sichuan Univ.)；Liu Qi(School of Electrical Eng. and Info.,Sichuan Univ.)；Li Yang(West China Hospital of Stomatology,Sichuan Univ.)
收稿日期：2013-06-14 年卷（期）页码：2014,46(2):127-132
期刊名称：工程科学与技术
Journal Name：Advanced Engineering Sciences
关键字：腭裂语音；高鼻音；香农能量；Mel倒谱系数；高斯混合模型识别器
Key words：cleft palate speech；hypernasality；Shannon energy；Mel frequency cepstrum coefficient；Gaussian mixture model classifier
基金项目：国家自然科学基金青年基金资助项目(30900391)
中文摘要
为了对腭裂语音的高鼻音进行等级区分,提出基于声学特征参数分析的腭裂语音高鼻音等级自动识别算法，提取基于香农能量和Mel倒谱系数（Mel frequency cepstrum coefficient,MFCC）的S-MFCC作为声学特征参数，结合高斯混合模型（Gaussian mixture model,GMM）分类器实现对腭裂语音4类高鼻音等级（正常、轻度、中度和重度）的自动识别。实验结果表明，提出的自动识别算法取得了较高的高鼻音类别正确识别率，对4类高鼻音的平均识别率达到79%以上，其中,提出的S-MFCC参数取得了85%的平均正确识别率,优于传统的香农能量算法、MFCC算法，具有较高的临床应用价值。
英文摘要
In order to detect hypernasal automatically for cleft palate patients, based on Shannon energy and Mel frequency cepstrum coefficient acoustic features and by combining with Gaussian mixture model classifier, an automatic hypernasal detection algorithm was proposed. The experiment results showed that the presented method achieved a good performance on the detection of four levels of hypernasal, such as normal, low-level, moderate-level and high-level. The average classification accuracies for four levels of hypernasal were over 79%. Moreover, the correct recognition accuracy using energy plus Mel frequency cepstrum coefficient feature set reached up to 85%. The classification of hypernasal levels has important clinical applications.
【关闭】
