ISSN (Online) 2581-9429

IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 2, April 2022

Impact Factor: 6.252

Review of Identification of Depression through

Speech Analysis
Prof. Nilesh Shelke1, Ms. Ruchika Jadhav2, Ms. Nikita Aldak2, Ms. Neha Moon2,
Ms. Gayatri Gajbhiye2, Ms. Namrata Patil2
Assistant Professor, Department of Computer Science Engineering1
Research Scholar, Department of Computer Science Engineering2
Priyadarshini College of Engineering, Nagpur, Maharashtra, India.
[email protected] and [email protected]

Abstract: Depression has developed into a crucial worldwide public health problem. This is a common
psychological disturbance which affects the individuals physically and psychologically which leads to
neurological illness. This effects on people in any age category. These makes researchers to work on this field
so much. Traditionally, depression identification is performed by using semi-structured interviews of an
individual and additional personality inventories that makes detection of depression is heavily depend on
individual’s response. Early treatment and identification of depression is needed to promote remission,
prevention of relapse and decreasing an emotional tension of the disorder. It is difficult to detect depression
at early stage of it using traditional processes. Studies in improvement of computational objective approaches
indicates that speech signal of a speaker shows valuable relationship between depression and speech. Hence
these acoustic features are used for diagnosis of depression. Enhancement in machine learning and deep
learning techniques makes understanding of depression characteristics more rapid and convenient way which
reduces the changes of clinical mistakes and labour costs. This paper shows study of various depression
detection system or feature selection methods used by researchers in this field. This makes to detect
depression at early stage and can be cured faster.

Keywords: Speech, Depression Detection, Voice Quality Features, Machine Learning, Emotion, Deep
Learning.

I. INTRODUCTION
Depression, a psychiatric illness which happens due to many factors like social, physical factors, etc. This expresses loss
of interest, fatigue, irritability, psycho motor retardation, mood swings in daily every joyful activity and affects individual’s
work productivity and thinking power. By World Health Organization (WHO), over 350 million people worldwide are
recognized with depression. Nearly one of the five women and one of the twelve men are affected by these major depressive
disorders [1]. It has developed into a crucial worldwide public health problem. During COVID-19 pandemic, depression
and anxiety show generality in population worldwide [2,3]. By 2030, depression will become second most affecting disorder
in general population [4]. Suicide is big result of depression which makes over 8,00,000 of people passed every year [5].
Symptoms for depression may be seen as
1. Deficit of commitment in day-to-day actions.
2. Grief, emptiness or thinking downward.
3. Discouragement.
4. Annoyed and lacking spirit.
5. Little confidence, ego assessment or sensing impotent.
6. Disturbance concentrating and struggling to make judgments.
7. Irritation or enormous exasperation.
Traditionally, Diagnostic Statistical Manual (DSM) was the depression detection standard protocol which includes
Patient Health Questionnaire Depression (PHQ) [6], Montgomery-Asberg Depression Rating Scale (MADRS) [7] and
Beck’s Depression Inventory (BDI) [8] plus by doctor’s judgment. According to their answers, scores are assigned
automatically. But all these are subjective measures are bias and requires physician’s experience. These all are patients self-
Copyright to IJARSCT DOI: 10.48175/568 186
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 2, April 2022

Impact Factor: 6.252
report symptoms and which makes to identify only half of the depression [9]. Most of the people don’t admit their emotional
instability and conditions of mental illness and public health. But depression cab be cured. Now mental health disorder
identification and recognition have put their step in artificial intelligence and computer vision community. WHO states that,
even though having treatments for depression, barriers in effective treatments and cure for depression is due to inaccurate
assessment. Hence the development in easily accessible tools which enable reliable treatment of depressive symptomatology
is needed. There are several systems have been involved to learn human-machine interactions which automatically access
mental state and emotions of person objectively [10].
It is shown that voice signals carry significant information about the mental health disorder in speaker’s speech [11-13].
The continuous changes in emotion state also affects acoustic properties of voice. Depression diagnosis using speech
characterises is more effective and convenient way. It is different from normal people’s speech characteristics. Distortions
in path of speech, increase in number of pauses monotonous pitch and lower intensity of loudness are shown to be
dependable indication of the depression [14-15]. These all are relatively objective features. The spectral feature and features
related to energy of voice can be used for detection in depression [16]. The fundamental frequency (F0), intensity, speed,
energy distribution and cepstral features are also considered to be good features for detection of depression. The variance
in audio amplitude, bandwidth, and energy is also reduced in person. Along with this, speech signal of patient having
depression do not show strong emotions like happiness and anger but they commonly show sadness, calm emotions. The
scale of pitch, volume get decreased in depression. These are possibly also measured in cheaply, remotely and non-
intrusively way [17]. This paper represents the study of related works done on depression identification and recognition of
depression through speech. The various diagnosis or recognition systems and selection of relevant features methods which
predicts the appropriate results estimated by researchers have discussed. This also shows works on different kind of datasets.

II. LITERATURE REVIEW

Physically how you appear is not just sufficient to say that you are physically well but also it is depended on how is your
mental state. Mental state is also a very big problem to deal with such as Depression. Health care treatments by professionals
are available but they are time consuming. There are several scientific literature research papers for depression detections
which uses different ways according to nature of data like speech, handwringing and drawing which focuses on shape of
the lines drawn, video analysis, contents of words written and spoken, electroencephalogram (EEG) and multimodality.
Here focusing on relevant features extracted from speech only. If anyone suffering from Depression, there is changed can
be found in speaking, pitch of voice, expressions of an individual. Studies have shown that speech show different voice
characteristics of depressed person for depression detection diagnosis. Emotions indicated through speech may be analyzed
in three distinct levels as follows:
 Physiological level: shows impulse of nerves or muscles consciousness in speech production process.
 Phonatory-articulatory level: shows movements and position of vocal folds or cords in voice of a person.
 Acoustic level: describes characteristics in the audio signals produced by speech.
There are many literatures which used acoustic and prosodic characteristic extracted from speech for depression detection
[18, 19, 20, 21]. There are many works done on automatic depression recognition and analysis [22]. The sizeable cross-
label differences but the compact inner-label differences must be manifested by feature. Most of the methods in literatures
are regarding automatic depression detection (ADD) methods. They examine two important sources of knowledge as either
combined or independent such as visual and audio methods. Even though here visual features are not considered for this
research but audio features for depression diagnosis is considered. A heterogeneous symbol-used method for detecting the
depression in speech was used by Zhaocheng Huang [23]. Sudden changes and acoustic regions are uniquely and inclusively
figured out to meet among various embedding ways.
The Gaussian Mixture Model (GMM) and Support Vector Machine (SVM) algorithm are most used Machine Learning
methods applied for depression detection model because of their potential to manage sparse dataset dynamically with
comparably small cost and presence in more free libraries (Cummins et al., 2015 [24]). GMM gain 77% accuracy by using
the Mel Frequency Cepstral Coefficients (MFCCs) only as feature, while SVM gain more accuracy than GVM but might
gain worst accuracy if unsuitable kernel is taken. Most of the research workers have depended on support vector machine
(SVM) classification to identify person with mental illness or depression except to find severity of depression with
regression classifiers (Schwartz et al., 2014). They researched essential characteristics of paralinguistic (study of vocal
Copyright to IJARSCT DOI: 10.48175/568 187
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 2, April 2022

Impact Factor: 6.252
signals beyond the normal verbal speech) speech. By this the patterns for regression and classification issues were analysed.
They have also discussed on current challenges and limitation in speech diagnosis.
The depression diagnosing through only speech has observed in Interspeech Conference held on 2018 in Hyderabad,
India. There are many machine learning methods which are multimodal in nature like text, video, audio, etc. But depression
detection using speech has also seen. Afshan et al. [25] concentrated on the efficacy of features of voice quality for
depression detection. The features like F0, F1, F2, MFCCs, F3, H2-H4, H1-H2, A1, A2, A3, H4-H2k and CPP are used for
voice quality in their research. The samples used in was of 10s and its accuracy of 77%, the accuracy also depend on the
length of sample audios. In the [26] researchers explored correlation between voiced metrics and changes in seriousness of
depression with period of time. The hypothesis presented by them shows that quantitative features in voiced prosody can
be used for depression detection classification. The results demonstrated that voiced prosody analysis is worthwhile for
analysis of depression. Sharifa, Goecke, Roland, Wagner, Michael, Parker, Gordon, Breakspear & Michael (2013) Tzirakis
et al. used deep learning methodology to evaluate depression severity of a person using CNN on audios and Deep Residual
System (ResNet) having 50 layers on visual information.
As discussed in above related studies, researchers have used various Machine Learning and deep learning models,
different features of audio, selection of relative features and use of proper recognition pattern to be used in depression
detection. It is found to be difficult having literatures which compares all these recognition patterns.
Actually, as per [27,28] the relevant features for audio analysis task are intensity, shimmer, harmonic, pitch, rates of
speech, formants still all these features are not required for analysis. The required ones are used for depression detection
using speech.
Paper Concept
Natural Language Processing Methods for Acoustic Speech features using proven natural language
and Landmark Event based Features in Speech-based processing methods
Depression Detection [23]
A Review of Depression and Suicide Risk Effect of depression on paralinguistic speech
Assessment Using Speech Analysis [24] characteristics
Effectiveness of voice quality features in detecting Automatic assessment of depression using conventional
depression [25] cepstral features
Detecting depression from facial actions and vocal The feasibility of automatic detection of depression
prosody [26]
Inferring clinical depression from speech and spoken Features from harmonic model improve the performance
utterances [27] of detecting depression from spoken utterances
Vocal-source biomarkers for depression: A link to Vocal biomarkers are introduced for detecting different
psychomotor activity [28] cognitive load conditions
Table 1: Previous works in depression detection using speech

III. CONCLUSION
The speech of a person shows useful characteristics for detecting depression or mental stress in an individual. This paper
shows the study of literature works on this field of depression detection using speech or voice samples of people. Content
of this research paper will provide guidelines to the researchers to work upon more research in area of depression detection
using speech.

REFERENCES
[1]. American Medical Association, “for Treatment of Mental Disor-ders in the World Health Organization,” Jama,
vol. 291, no. 21, pp. 2581–2590, 2004.
[2]. N. Salari, A. Hosseinian-Far, R. Jalali, A. Vaisi-Raygani, S. Ra-soulpoor, M. Mohammadi, S. Rasoulpoor, and B.
Khaledi-Paveh, “Prevalence of stress, anxiety, depression among the general pop-ulation during the covid-19
pandemic: a systematic review and meta-analysis,” Globalization and health, vol. 16, no. 1, pp. 1–11, 2020.
[3]. R. Barzilay, T. M. Moore, D. M. Greenberg, G. E. DiDomenico, L. A. Brown, L. K. White, R. C. Gur, and R. E.
Gur, “Resilience, covid-19-related stress, anxiety and depression during the pandemic in a large population
enriched for healthcare providers,” Translational psychiatry, vol. 10, no. 1, pp. 1–8, 2020.
Copyright to IJARSCT DOI: 10.48175/568 188
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 2, April 2022

Impact Factor: 6.252
[4]. C. D. Mathers and D. Loncar. “Projections of global mortality and burden of disease from 2002 to 2030,” PLoS
medicine, vol. 3, no. 11, pp. e442, 2006.
[5]. W. H. Organization et al., “Preventing suicide: A global impera-tive," 2014.
[6]. K. Kroenke, R. L. Spitzer, and J. B. Williams, “The phq-9: validity of a brief depression severity measure,” Journal
of general internal medicine, vol. 16, no. 9, pp. 606–613, 2001.
[7]. A. T. Beck, C. H. Ward, M. Mendelson, J. Mock, and J. Erbaugh, “An inventory for measuring depression,”
Archives of general psy-chiatry, vol. 4, no. 6, pp. 561–571, 1961.
[8]. S. Montgomery and M. Asberg, “A new depression scale designed to be sensitive to change,” Acad. Department
of Psychiatry, Guy’s Hospital. 1977.
[9]. A. J. Mitchell, A. Vaze, S. Rao, “Clinical diagnosis of depressionin primary care: a meta-analysis” The Lancet,
vol. 374, no. 9690, pp. 609–619, 2009.
[10]. F. Dornaika, B. Raducanu, “Inferring facial expressions fromvideos: Tool and application,” Signal Processing:
Image Com-munication, vol. 22, no. 9, pp. 769–784, 2007.
[11]. France DJ, Shiavi RG, Silverman S, Silverman M, Wilkes DM, “Acoustical properties of speech as indicators of
depression and suicidal risk,” IEEE Trans Biomed Eng, vol. 47, no. 7, pp. 829–37, Jul 2000.
[12]. Mundt JC, Snyder PJ, Cannizzaro MS, Chappie K, Geralts DS, “Voice acoustic measures of depression severity
and treatment response collected via interactive voice response (IVR) technology,” J Neurolinguist, vol. 20, no. 1,
pp. 50–64, Jan 2007.
[13]. Stasak B, Epps J, Goecke R., “Elicitation design for acoustic depression classification: An investigation of
articulation effort, linguistic complexity, and word affect,” Interspeech; 2017 Aug 20–24, Stockholm, pp. 834–8.
[14]. J. C. Mundt, P. J. Snyder, M. S. Cannizzaro, K. Chappie, and D. S. Geralts, “Voice acoustic measures of
depression severity and treatment response collected via interactive voice response (IVR) technology,” Journal of
Neurolinguistics, vol. 20, no. 1, 2007, pp. 50–64.
[15]. J. C. Mundt, A. P. Vogel, D. E. Feltner, and W. R. Lenderking, “Vocal acoustic biomarkers of depression severity
and treatment response,” Biological Psychiatry, vol. 72, no. 7, 2012, pp. 580– 587.
[16]. Tolkmitt, F., Helfrich, H., Standke, R., and Scherer, K., “Vocal indicators of psychiatric treatment effects in
depressives and schizophrenics,” J. Commun. Disord, vol. 15, pp. 209–222, 1982, doi: 10.1016/0021-
9924(82)90034-X.
[17]. N. Cummins, J. Epps, M. Breakspear, R. Goecke, “An investigation of depressed speech detection: Features and
normalization,” In: Twelfth Annual Conference of the International Speech Communication Association, 2011.
[18]. N. Cummins, “Automatic assessment of depression from speech: Paralinguistic analysis, modelling and machine
learning,” Ph.D. dissertation, The University of New South Wales, 2016.
[19]. J. Williamson, T. Quatieri, B. Helfer, G. Ciccarelli, and D. Mehta, “Vocal and facial biomarkers of depression
based on motor incoordination and timing,” in Proceedings of AVEC’14, 2014.
[20]. P. Lopez-Otero, L. Docio-Fernandez, and C. Garcia-Mateo., “A study of acoustic features for depression
detection,” in Proceedings of IWBF, pp. 1–6, 2014.
[21]. P. Lopez-Otero, L. Fernández, C. Garcia-Mateo, “A study of acoustic features for the classification of depressed
speech,” in Proceedings of MIPRO, pp. 1331–1335, 2014.
[22]. A. Pampouchidou, P. Simos, K. Marias, F. Meriaudeau, F. Yang,M. Pediaditis, M. Tsiknakis, “Automatic
assessment of depres-sion based on visual cues: A systematic review,” IEEE Transac-tions on Affective Computing,
2017.
[23]. Z. Huang, J. Epps, D. Joachim, and V. Sethu, “Natural Language Processing Methods for Acoustic and
Landmark Event-based Features in Speech-based Depression Detection,” IEEE Journal on Selected Topics in
Signal Processing, vol. 14, no. 2, pp. 435-448, 2019.
[24]. Cummins, N., Scherer, S., Krajewski, J., Schnieder, S., Epps, J., & Quatieri, T. F., “A Review of Depression and
Suicide Risk Assessment Using Speech Analysis,” Speech Communication, vol. 71, pp. 10–49, 2015.
http://doi.org/10.1016/j.specom.2015.03.004

Copyright to IJARSCT DOI: 10.48175/568 189

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 2, Issue 2, April 2022

Impact Factor: 6.252
[25]. Afshan, Amber, Jinxi Guo, Soo Jin Park, Vijay Ravi, Jonathan Flint, and Abeer Alwan, “Effectiveness of voice
quality features in detecting depression,” In Proc. Interspeech, pp. 1676-1680, 2018.
[26]. J.F. Cohn, T.S. Kruez, I. Matthews, Y. Yang, M.H. Nguyen, M.T. Padilla, F. Zhou, F. De, la Torre, “Detecting
depression from facial actions and vocal prosody,” International Conference on Affective Computing and
Intelligent Interaction and Workshops, pp. 1–7, 2009.
[27]. Asgari, M.; Shafran, I.; Sheeber, L.B., “Inferring clinical depression from speech and spoken utterances,” In Proc.
IEEE International Workshop on Machine Learning for Signal Processing (MLSP), Reims, France, 21–24
September 2014, pp. 1–5.
[28]. Quatieri, T.F., Malyska, N., “Vocal-source biomarkers for depression: A link to psychomotor activity,” In Proc.
Thirteenth Annual Conference of the International Speech Communication Association, Portland, OR, USA,
September 2012, pp. 9–13.
[29]. karol j. piczak, “Environmental sound classification with convolutional neural networks,” 2015 IEEE
International Workshop on Machine Learning for Signal Processing, pp. 17–20.
[30]. Nguyen T., Pernkopf, “Acoustic scene classification using a convolutional neural network ensemble and nearest
neighbour filters,” Detection and Classification of Acoustic Scenes and Events 2018 Workshop (DCASE2018),
Surrey.

Copyright to IJARSCT DOI: 10.48175/568 190

www.ijarsct.co.in