自动检测正常嗓音和病理嗓音的关键是选出有效的特征参数,并对其进行优化得到简单易实现的参数。同时选择合适的识别模型对正常嗓音和病理嗓音进行识别以得到最好的识别率。为了能实时、便利地检测正常嗓音和病理嗓音,这里提出了线性预测倒谱系数(LPCC)和MEL频率倒谱系数(MFCC)声学特征参数,采用动态时间规整(DTW)算法进行识别,实验结果表明该模型的识别率可达到90%以上,且MFCC方法优于LPCC。 The key to automatically detect the normal voice and pathological voice is to select the valid characteristic parameters and optimize them to get the simple and easy to implement parameters. At the same time select the appropriate recognition model to normal voice and pathological voice recognition to get the best recognition rate. In order to detect normal voice and pathological voice in real time and in real time, acoustic parameters of Linear Predictive Cepstral Coefficients (LPCC) and MEL Frequency Cepstral Coefficients (MFCC) are proposed, and dynamic time warping (DTW) algorithm is used to identify the experimental results The recognition rate of this model can reach more than 90%, and the MFCC method is superior to LPCC.