基于语音特征与MFCC谱图融合模型的抑郁症检测

Detection of depression based on fusion model of speech feature with MFCC profiles

据《2022年国民抑郁症蓝皮书》显示,我国的抑郁症患者人数接近1亿。为更好地实现抑郁症辅助检测,文章首先利用MFPH端点检测方法分离了语音信号的有声段、无声段,其次提取了停顿时长、停顿次数、短时过零率等语音特征及MFCC特征谱图。对比分析发现,基于语音特征与MFCC特征谱图的融合模型在测试集上的准确率可以达到76.4%。

According to the 2022 National Depression Blue Book,the number of depression patients in China is close to 100 million.To better achieve depression assisted detection,the article first used the MFPH endpoint detection method to separate the voiced and voiceless segments of the speech signal.Secondly,speech features such as pause duration,pause frequency,and short-term zero crossing rate were extracted,as well as MFCC feature spectrograms.Comparative analysis shows that the fusion model based on speech features and MFCC feature spectrograms can achieve an accuracy of 76.4%on the test set.