Clinical Studies On Voice Signal Features Of Bipolar Mania

Objective Given the lack of objective usable biological markers for early identification of bipolar mania,and the sufferers' voice has obvious fluctuation when the transition into different state.The purpose of this study is to investigate the voice signal features of patients with bipolar mania and healthy controls.Methods 30 manic patients(MP)and 30 healthy controls(HC)were recruited.The voice signal features of the recruited subjects were collected by Voice Collecting System and then were analyzed.Bech-Rafaelsdn Mania Rating Scale(BRMS)and Clinical impression rating scale(CGI)were used to assess the illness.We compared the differences of voice signal features between manic state and normal state among MP and between MP and HC.Then making correlational analysis between voice signal features and BRMS.Results(1)We found the first Formant(F1)(p = 0.000),the second Formant(F2)(p = 0.036),and the Linear Prediction Coefficient(LPC)(p = 0.000)were significantly higher in MP than HC.In addition,the BRMS score has significant positive correlation with LPC(r = 0.398;p = 0.040).(2)As compare the different state in patients with bipolar disorder,we found the significant differences in the fourth Formant(F4)(p = 0.003),Formant Bandwidth(p = 0.040),Formant Amplitude(p = 0.004)and the Linear Prediction Coefficient(LPC)(p = 0.001)between manic state and normal state.Among them,F4,Formant Amplitude and LPC in manic state were higher than normal state,Formant Bandwidth in manic state was lower than normal.(3)The ratio using single LPC for classification was 4.86,which was obviously higher than Formant and MFCC.(4)In designing the classifier to distinguish the different state,we found that the recognition rate of SVM classifier for a single patient were up to 90% and multi patients' rate of 50%,we also found that the GMM classifier yields the best performance with a classification rate of 72% for multi patients.Conclusions(1)F1,F2 and LPC in MP were higher than HC,it showed that F1,F2,LPC had some certain value for early identification of manic disease.(2)The correlation analysis showed positive correlation between clinical symptoms and LPC,it prompted that LPC was likely to be objective biological indicators in early warning,disease severity and treatment during the clinical process control.(3)F4,Formant Amplitude and LPC in manic state were higher than normal,and Formant Bandwidth was lower than normal,it suggested that single vocal acoustic measure also could reflected manic symptoms variations in some degree.(4)Support Vector Machine(SVM)had a good performance for a signal patient,Gaussian Mixture Model(GMM)had a better performance than SVM for multi patients' classification.and LPC was the most important features than the others.In the research,LPC played an important role in bipolar mania,it may be more sensitive to judge clinical severity and treatment efficacy,even early recognition and warning.