论文部分内容阅读
目的:探讨临床病态嗓音的特征及计算机自动识别病态嗓音的可行性。方法:选择129例声带息肉患者为病态嗓音组,同期选取125例社区正常嗓音人群为对照组。应用Praat软件采集分析2组病例获得相关声学参数值,包括基频微扰、振幅微扰、谐噪比、信噪比、声门噪声。采用该病态嗓音组与对照组病例作为神经网络检测的训练集和测试集。同样方法另外收集140例病态嗓音及正常嗓音数据作为验证集。应用SPSS Modeler软件进行人工神经网络建模,计算模型对病态嗓音的识别率。结果:本研究根据不同性别分组计算,病态嗓音组在基频微扰、振幅微扰、声门噪声方面数值比对照组增大(P<0.05),谐噪比、信噪比方面数值比对照组减少(P<0.05)。人工神经网络模型对病态嗓音的识别率为75.7%。结论:客观嗓音分析有助于病态嗓音的鉴别,人工神经网络在病态嗓音的识别上准确率较高,有很好的临床应用价值。
Objective: To investigate the characteristics of clinically pathological voice and the computer to automatically identify the feasibility of morbid voice. Methods: 129 patients with vocal cord polyps were selected as pathological voice group, and 125 normal voice groups were selected as control group. Praat software was used to collect and analyze two groups of patients to obtain relevant acoustic parameter values, including fundamental frequency perturbation, amplitude perturbation, harmonic ratio, signal to noise ratio and glottal noise. The pathological voice group and the control group were used as the training set and the test set of the neural network. In the same way, another 140 sick and normal voice data were collected as validation sets. The application of SPSS Modeler software for artificial neural network modeling to calculate the recognition rate of pathological voice model. Results: According to the different gender grouping, the values of phonetic disturbance, amplitude perturbation and glottal noise in pathological voice group were higher than those in control group (P <0.05), and the values of harmonic ratio and signal to noise ratio were higher than those of control group Group decreased (P <0.05). Artificial neural network model for pathological voice recognition rate of 75.7%. Conclusion: Objective voice analysis is helpful for the identification of morbid voice. Artificial neural network has a high accuracy in identification of morbid voice and has good clinical value.