基于非线性参数的腭裂患者高鼻音自动识别

Automatic hypernasal detection for left palate patients based on non-linear parameters

为了实现对腭裂患者高鼻音的自动识别,提出了基于语音信号非线性能量算子及K-最近邻分类器的高鼻音自动识别算法。实验对腭裂语音数据库中非高鼻音及高鼻音信号提取基于香农能量和非线性能量算子的语音特征参数,结合模式识别分类器,实现了对高鼻音语音的自动检测。实验结果表明,应用非线性能量算子,能实时跟踪语音信号瞬时能量变化,实现了对高鼻音较高的判别正确率,其分类器正确识别率在90%以上,且优于传统的香农能量算法,具有较高的临床应用价值。

To detect hypernasal automatically for cleft palate patients,a hypernasal identification algorithm in speech signals is proposed based on Shannon energy and non-linear energy operator combined with k-nearest neighbor classifier.The experiment is implemented for non-hypernasal and hypernasal speech of cleft palate patients.The non-linear energy operator parameters are extracted as acoustic features,then pattern recognition classifier is applied to classify non-hypernasal and hypernasal speech.The experiment results indicate that the non-linear energy operator tracked instantaneous energy change of speech signals in time.The proposed algorithm achieves high classification accuracies for hypernasal detection,the correct rates are over 90％,which is higher than that achieved by Shannon energy feature.The system has high potential in clinical applications.