An Analysis Of Acoustic Features In Reading Speech From Chinese Patients With Depression

In the present study,we described and analyzed in detail the voice quality features and formant features of speech of patients with depression,in order to investigate whether there is a difference in voice acoustic between depression and non-depression,and to verify whether these speech features can effectively distinguish depression from non-depression and accurately predict the depression levels.Through experiments,we hope to further confirm whether speech can be used as an objective marker for diagnosing depression patients.Annotation and Methodology: The analysis is based on 3 reading passages of per subject.(1)Voice quality analysis is a segments analysis based on large sample.Voice quality features include jitter,shimmer,harmonic noise ratio(HNR),fundamental frequency and cepstrum peak prominence(CPP).The analysis of variance and multiple regression analysis are applied in the analysis of voice quality.(2)Formant analysis is an experiment based on a small sample.And subjects include 5 patients with depression(as experimental group)and 5 healthy people(as control group).The 8 vowels which are as nucleus(/ a /,/ i /,/ u /,/ y /,/ ? /,/ ? /,/ ? /,/ ? /)and 6 diphthong vowels which are as finals(ai,ei,ao,ou,ua,uo)are annotated.Formant features include formant frequency and formant bandwidth.The independent sample t-test is applied in formants analysis.Results:(1)Voice quality analysis.The analysis of variance shows that patients with depression and non-patients has significant differences in jitter,shimmer,frequency,HNR,and CPP.However,the results of descriptive statistical did not show obvious rules,and only some of them were in line with expectations.First,jitter and shimmer only increase with the level of Beck Depression Inventory when they are used as the speech features of patients with depression.The increase in jitter and shimmer indicates an increase in speech roughness and hoarseness in patients with depression.Second,in analyzing all the participants,the mean frequency and standard deviation of the fundamental frequency decreased with the increase of the depression degree.Third,the maximum HNR of BDI level four is lower than that of the first three levels,which indicates that the speech of patients with major depression is most hoarse.Fourth,multiple regression models based on voice quality characteristics have a certain ability to predict BDI scores.(2)Formant analysis.The results of independent sample show that there is no significant difference in formant features between the experimental group and the control group,but descriptive statistics show that patients with depression reduce vowels and have a lower opening degree.Firstly,the F1,F2,and F1 standard deviations and F2 standard deviations of the experimental group are lower than those of the control group.Secondly,the bandwidth of the three formants of the experimental group is larger than that of the control group,which indicates the spectral energy of patients is lower than that of the healthy people.Thirdly,the nucleus of experimental group has a lower HNR than control group.Fourthly,in the analysis of diphthong vowels,it was found that the mean of F1 and F2 in the experimental group were lower than those in the control group.Fifthly,the changes of the F1 and F2 of the experimental group are smaller than those of the control group.Conclusions: This paper proves that the jitter,shimmer and HNR have the ability to distinguish the degree of depression in patients.Additionally,the fundamental frequency,the standard deviation of the fundamental frequency and HNR have the ability to distinguish between patients with depression and non-patients.This paper also describes the differences in formant characteristics between patients with depression and non-patients,indicating that formant characteristics can be used as an objective marker for the recognition of depression.There are 50 figures,32 tables and 42 references in this paper.