International Journal of E-Health and Medical Communications

Volume 12 • Issue 4 • July-August 2021

Vocal Folds Analysis for Detection and

Classification of Voice Disorder:
Detection and Classification of Vocal Fold Polyps
Vikas Mittal, National Institute of Technology, Kurukshetra, India
https://orcid.org/0000-0003-3808-5057

R. K. Sharma, National Institute of Technology, Kurukshetra, India

ABSTRACT

The detection and description of pathological voice are the most important applications of voice
profiling. Currently, techniques like laryngostroboscopy or surgical microlarynoscopy are popularly
used for the diagnosis of voice pathologies but are invasive in nature. Disorders of vocal folds impact
the quality of voice, and therefore, the accuracy of voice profiling is reduced. This paper presents
a better solution to differentiate normal and pathological voices based on the glottal, physical,
and acoustic and equivalent electrical parameters. These parameters have been correlated using
mathematical equations and models. Results reveal that the glottal flow is strongly influenced by
physical parameters like stiffness and viscosity of vocal folds in case of pathological voice. However,
their direct measurement requires complex invasive medical procedures or costly and complex
electronic hardware arrangements in case of non-invasive methods. Glottal parameters, on the other
hand, facilitate much simpler estimation of vocal folds disorders. In this work, the authors have
presented two non-invasive approaches for better accuracy and least complexity for differentiating
normal and pathological voices: 1) by using correlation of glottal and physical parameters, 2)by using
acoustic and equivalent electrical parameters.

Keywords
Acoustic Circuit, Electrical Circuit, Glottal Parameters, Physical Parameters, Two Mass Model, Vocal Disorders

1. INTRODUCTION

The risk of pathological voice related disorders has increased manifolds. This is due to modern
lifestyle, environmental issues, self medications and even a profession. About 25% of the population
is engaged in activities that are “vocally demanding” (Amami & Smiti, 2017). The examples include
professors, lawyers, auctioneers, aerobics instructors, singers, actors and manufacturing supervisors.
For the diagnosis of voice pathologies, invasive endoscopy procedures are the current state of the
art. But recently non-invasive digital techniques (like voice profiling and image processing) have
evolved and are assisting medical professionals for early detection of voice disorders. In voice
based detection, the most common method for extracting voice features is determination of acoustic
parameters directly from the voice signal. Since most of the voice disorders are due to vocal fold
dynamics, the researchers have started to work with glottal parameters of vocal folds to expedite the
detection of related disorders. The detection of voice pathologies needs further improvement so as

DOI: 10.4018/IJEHMC.20210701.oa6
This article, published as an Open Access article on April 23rd, 2021 in the gold Open Access journal, the International Journal of Informa-
tion and Communication Technology Education (converted to gold Open Access January 1st, 2021), is distributed under the terms of the
Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/) which permits unrestricted use, distribution, and
production in any medium, provided the author of the original work and original publication source are properly credited.

97
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

to increase the accuracy of voice detection as well as their classification. This work aims improved
detection and classification of voice disorders using vocal folds glottal, physical, acoustic, and
equivalent electrical parameters.
Vocal acoustic evaluation is popularly used for the assessment and diagnosis of voice disorders
(Teixeira et al., 2020). Xiao Yao et al. claimed that when the speaker is under stress, certain vocal organs
are affected (Yao et al., 2015).Xiao Yao et al. further discussed the physical parameter, glottal flow and
stress output relationships(Yao et al., 2018). Although the voice parameters like the vibration of the
vocal folds, shape of the glottis and the glottal airflow have been extensively researched in literature,
yet their individual impacts on the voice quality cannot be accurately computed (Ramsay, 2019). It is
a fact that thorough study and evaluation of vocal folds behavior essentially require characterization of
vocal folds and their relationship with vocal tract. This paper focuses on the diagnosis of pathological
voice using physical, glottal parameters as well as acoustic and its equivalent electrical parameters.
The major contributions of this paper are summarized as under:

• It is a fact that speech disorders in the voice is caused fundamentally by the physiological changes
of vocal folds leading to deviation in their natural vibrations and are reflected by glottal flow.
The authors have proposed a novel method to find physical parameters of vocal folds. On the
other hand IAIF method is used to extract glottal flow parameters from the given voice samples.
The physical parameters are then correlated with glottal flow parameters. Any change in glottal
parameters reflects change in physical parameters which are then utilized to classify pathological
voices. Hence the contents of the paper presents a method for detection and classification of
voice disorders based on physical speech production model and characteristics of glottal flow.
• Furthermore, authors have also developed relation between vocal folds length and parameters
of the acoustic model of the vocal folds. Change in vocal folds length, due to voice disorders
reflects change in acoustic model parameters and the same has been used to classify pathological
voices. The results of acoustic parameters variations due to voice disorders are shown in Table
10 at page-15 of the manuscript.
• It has also been experimented that current variation in equivalent electrical model is a function
of change in vocal folds length. This feature has been used to classify pathological voices as
shown in Figure 16.

2. PHYSICAL MODEL OF VOCAL FOLDS

The vocal folds are situated in an anterior-posterior orientation in the middle of the glottis. There is
a “V” shaped form in the right and left folds. The gap in the shape of “V “forms the entrance to the
trachea. Each vocal fold is attached with muscles on both sides of the larynx. The contraction and
relaxation of the muscles lead to glottal flow and is termed as the glottal source. According to body-
cover theory stiffness properties of vocal folds depends on thyroarytenoid (TA) and cricothyroid (CT)
muscles. As a result, the behavior of vocal folds depends on the combined effect of body oscillation
and layers of cover. This paper considers only the symmetrical two-mass vocal folds model, shown
in Figure 1. Each vocal fold is composed of a lower mass (m1) and an upper mass (m2), stiffnesses
(k1, k2), and viscous resistances (r1, r2). Ps is Subglottal pressure which defines the pressure of the
airflow in the trachea below the glottis. The vocal fold vibration is the phonation source, which
further determines the nature of the glottal flow. Since voice disorders result in variations in vocal
fold stiffness (k1), viscous resistance (r1) and varying airflow patterns in the glottis may affect glottal
flow production (Hirano, 1974).

98
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 1. Symmetrical vocal folds two mass model (Riede, 2011)

3. METHODOLOGY

For determining the relation between glottal flow and physical parameters, the procedure shown in
Figure 2 has been used.
Two mass model is used to compute physical parameters (Ishizaka & Flanagan, 1972) meanwhile
the glottal parameters are estimated from voice samples of normal and pathological voices(Mokhkari
et al., 2018), using Aalto Aparat tool. The proposed work makes use of the German database;
‘Saarbrucken Voice Database (SVD)’. This database is freely available online. The authors have
also recorded real voice samples from MMIMSR, Mullana hospital. There are total 16 samples that
include eight (8) healthy and eight (8) pathological subjects, suffering with vocal folds polyps. Each
category includes four (4) male and four (4) female samples all above 18 years. We have considered
recordings of vowels /a/ produced at normal pitch. The durations of samples varies between 1 to 3
seconds. All recordings are sampled at 50 kHz with 16 bit resolution.

4. RELATIONSHIP AMONG PHYSICAL & GLOTAL PARAMETERS

To achieve the objective of this work, the following procedure is used to demonstrate a correlation
between physical and glottal parameters. Furthermore, the relationships among these parameters
have been plotted.

4.1. Glottal Parameters

Aalto Aparat tool is used to extract glottal parameters & these parameters are briefly explained as under:

4.2. Normalized Amplitude Quotient (NAQ)

The parameter NAQ specifies the closing phase of vocal folds, and given as:

99
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 2. Blocks of methodology used

100
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 3. Representation of one glottal pulse (Forero et al., 2014)

AQ
NAQ = (1)
T

where T is the glottal period, AQ (Amplitude Quotient) is defined by the maximum amplitude of the
glottal flow (Mittal & Sharma, 2020).

4.3. Speed Quotient (SQ)

It defines the ratio of opening and closing intervals of vocal folds and expressed as (Mittal & Sharma,
2020):

To1
SQ = (2)
Tc

Some authors consider two parameters for the speed quotient: SQ1 and SQ2, as:

T − Tc
SQ1=  (3)
Tc

T
SQ2 =  o1 (4)
Tc

To1 is a time interval between the beginning instant of opening and the instant when the opening
is maximum & time interval between max. opening to the complete closure is Tc .

101
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

4.4. Opening Quotient (OQ):

It is defined as:

To1 + Tc
OQ = (5)
T

Some authors consider two parameters for the OQ: OQ1 and OQ2. These are defined as(Mittal
& Sharma, 2020):

To1 + Tc
OQ1 = (6)
T

and

To2 + Tc
OQ2 = (7)
T

Where To2 as observed in Figure 3

4.5. Closing Quotient (CIQ)

It is defined as:

Tc
CIQ = (8)
T

4.6. Physical Parameters:

Physical parameters stiffness and viscosity that characterize the vocal folds are discussed as under.

4.6.1. Stiffness
Stiffness is related to muscle tension and fundamental frequency (Cataldo et al., 2006). It, therefore,
affects the closing and opening of vocal folds.

d 2x 1 dx 1
m1 2
+ r1 + s1 (x 1 ) + kc (x 1 − x 2 ) = F1 (9)
dt dt

d 2x 2 dx 2
m2  2
+ r2  + s2 (x 2 ) + kc (x 1 − x 2 ) = F2 (10)
dt dt

Where mi, ri, si and Fi are viscous resistance, elasticity, and airflow respectively. xi is the horizontal
displacement from the balance of the two masses. kc denotes to the stiffness of the coupling between
the two masses. The elasticity can be computed as:

102
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

( )
si (x i ) = ki x i +ηx i 3 i = 1, 2 (11)

where ki is the stiffness parameter.

In the following discussion the relations of the fundamental frequency, physical parameters, and
glottal parameters have been derived.
The fundamental frequency F0 is expressed as under,

1
F0 = (12)
T

T is given as:

AQ
T= (13)
NAQ

As per in Ishizaka and Flanagan model 1972, standard value of m1 can be considered equal to
0.125g. Further the researchers (Ishizaka & Flanagan, 1972) relate m1& m2 and k1 & k2 as expressed
below:

m1 k1
m2 = and k2 = (14)
5 10

where k1 is lower spring stiffness and m1 is lower mass. Similarly, k2 is upper spring stiffness and
m2 is upper mass.
F0, as a function of k and m, can be defined as:

1 k
F0 = (15)
2π m

where k = k1 + k2 and m = m1 + m2.Consequently, F0 reduces to:

1 1.1k1
F0 = (16)
2π 1.2m1

From Equation (16), k1 can be calculated as:

(F 0 * 2π) * (1.2m )
2
1
k1 = (17)
1.1

In above developed formula k1 is proportional to F0 so it is inversely proportional to T. NAQ,

OQ and CIQ glottal parameters are inversely proportional to T and thus stiffness parameter k1

103
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

is proportional to NAQ. SQ parameter is proportional to glottal time period T, so it is inversely

proportional to k1.

4.6.2. Viscosity
Viscosity of vocal fold tissues plays an important role in vocal fold oscillation and it varies during
phonation. Kaneko,et al. (Kaneko et al., 1972) and Isshiki (Isshiki, 1977) have measured the viscous
resistance. The vocal folds viscous resistance reflects the stickiness of surfaces during vocal fold
contraction, which can be calculated as:

r1 = 2ζ1 m1k1 r2  = 2ζ2 m2k2 (18)

where ζ1 , ζ2 refer to damping ratios for the viscous resistances r1 and r2 .
r1 is computed assuming damping ratio ζ1 =0.1 as in Ishizaka and Flanagan model 1972(Ishizaka
& Flanagan, 1972) . Viscous resistance r1 is proportional to k1. So, r1 is also proportional to NAQ
and inversely proportional to SQ glottal parameter.

5. RESULTS AND DISCUSSION

The mathematical relations develop above are used to process voice samples of healthy and pathological
subjects. In this work, Saarbrucken Voice Database (SVD) (Barry & Putzer, 2015) samples are used
to test relationships for the speakers in the database. Using the same set of speech samples, relevant
physical parameters, and glottal parameters are derived and plotted.
The computed values of stiffness (k1) and NAQ of Normal and Pathological Voices are shown in
Table 1 and their graphical representations are shown in Figure 4. Stiffness (k1) represents the tension
in the cricothyroid (CT) muscle. The high value of stiffness (k1) is responsible for the contraction of
the cricothyroid (CT) muscle which causes slow vocal folds closure during vibration. An increase
in k1 raises NAQ. Stiffness (k1) for normal and pathological voices is obtained using Equation17
whereas NAQ for both types of voices is obtained using inverse filtering.

Table 1. Computed values of k1 and NAQ

Sample No. Normal Voice Pathological Voice

Stiffness k1(kdyn/cm) NAQ Stiffness k1(kdyn/cm) NAQ
1 44.5 0.035 44.5 0.021
2 55.9 0.032 52.7 0.027
3 76.1 0.027 77.4 0.042
4 102.4 0.039 149.9 0.039
5 174.2 0.036 194.1 0.105
6 176.1 0.075 226 0.073
7 237.1 0.041 228.2 0.048
8 253.2 0.060 262.6 0.049

The computed values of k1 and SQ of Normal and Pathological Voices are shown in Table 2.

104
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 4. Relationship between NAQ and k1

Figure 5 shows graphical representation of the relation of stiffness (k1) with speed Quotient (SQ).
The value SQ parameter is higher in pathological voice. It is clear from Figure 5 that shrinking in the
structure of vocal folds due to the presence of vocal diseases.
The values of k1 and OQ of Normal and Pathological Voices are shown in Table 3.
Figure 6 shows the relation of stiffness k1 with the Primary Opening Quotient (OQ). It is obvious
from Figure 6 that value of the OQ parameter is higher in pathological voice. The increased value of
OQ is due to the partial closing of vocal folds.
The computed values of k1 and CIQ are shown in Table 4 and Figure 7 shows the relation of
stiffness (k1) with Closing Quotient (CIQ). The value CIQ parameter is higher in pathological voice.
The computed values of r1 and NAQ are shown in Table 5.
Figure 8 demonstrates the relationship between the glottal parameter NAQ and the physical
parameter viscous resistance r1. The value of NAQ is high in the case of pathological voices. An
increase in viscous resistance r1 will lead to an increase in NAQ. Due to that, the vocal folds activity
of adduction and abduction is delayed.
The computed values of r1 and SQ are shown in Table 6.

Table 2. Computed values of k1 and SQ

Sample No. Normal Voice Pathological Voice

Stiffness k1 (kdyn/cm) SQ Stiffness k1 (kdyn/cm) SQ
1 44.5 3.7 44.5 6.9
2 55.9 3.1 52.7 3.4
3 76.1 2.3 77.4 2.8
4 102.4 2.2 149.9 3.2
5 174.2 2.1 194.1 2.9
6 176.1 0.4 226 2.2
7 237.1 1.1 228.2 2
8 253.2 1.7 262.6 2.2

105
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 5. Relationship between SQ and k1

Figure 9 demonstrates the relationship between SQ and the viscous resistance r1. An increase in
viscous resistance r1 will lead to a decrease in SQ. The value of SQ is high in the case of pathological
voices. It implies that a more asymmetric glottal flow is developed.
The computed values of {r1 and OQ} and {r1 and CIQ} are shown in Table 7 and Table 8.
Figure 10 and Figure 11 demonstrates the relationship between the OQ, CIQ and the viscous
resistance r1. The value of OQ and CIQ is high in case of pathological voices. An increase in viscous
resistance r1 will lead to an increase in OQ and CIQ that mean slowing down the vibration speed of
the vocal folds.
Authors have correlated physical parameters (stiffness, viscous resistance) and glottal flow
parameters (NAQ, SQ, OQ1 and CIQ) for normal and pathological voice.
Table 9 shows the average values of computed parameters for normal and pathological voice
data. The increase in k1 for pathological voices is due to the contraction of CT which in turn happens
due to distorted muscle under tension. Increased k1 leads to deceleration of vocal folds and thus
asymmetrical glottal flow. This behavior of glottal flow is also reflected by larger NAQ, SQ, OQ1,

Table 3. Computed values of k1 and OQ

Sample No. Normal Voice Pathological Voice

Stiffness k1 (kdyn/cm) OQ Stiffness k1 (kdyn/cm) OQ
1 44.5 0.305 44.5 0.243
2 55.9 0.201 52.7 0.224
3 76.1 0.230 77.4 0.233
4 102.4 0.268 149.9 0.330
5 174.2 0.124 194.1 0.521
6 176.1 0.247 226 0.446
7 237.1 0.304 228.2 0.299
8 253.2 0.349 262.6 0.338

106
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 6. Relationship between OQ and k1

Table 4. Computed values of k1 and CIQ

Sample No. Normal Voice Pathological Voice

Stiffness k1 (kdyn/cm) CIQ Stiffness k1 (kdyn/cm) CIQ
1 44.5 0.037 44.5 0.167
2 55.9 0.051 52.7 0.073
3 76.1 0.057 77.4 0.069
4 102.4 0.076 149.9 0.065
5 174.2 0.17 194.1 0.016
6 176.1 0.244 226 0.35
7 237.1 0.10 228.2 0.118
8 253.2 0.106 262.6 0.110

Figure 7. Relationship between CIQ and k1

107
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Table 5. Values of r1 and NAQ

Sample No. Normal Voice Pathological Voice

Viscous resistance (r1) NAQ Viscous resistance (r1) NAQ
1 14.9 0.035 14.9 0.021
2 16.7 0.032 16.2 0.027
3 19.5 0.027 19.6 0.042
4 22.6 0.039 27.3 0.039
5 29.5 0.036 31.1 0.105
6 29.6 0.075 33.6 0.073
7 34.5 0.041 33.7 0.048
8 35.5 0.060 36.2 0.049

Figure 8. Relationship between NAQ and r1

Table 6. Computed values of r1 and SQ

Sample No. Normal Voice Pathological Voice

Viscous resistance (r1) SQ Viscous resistance (r1) SQ
1 14.9 3.7 14.9 2.8
2 16.7 3.1 16.2 3.1
3 19.5 2.3 19.6 3.8
4 22.6 2.2 27.3 3.2
5 29.5 2.1 31.1 2.9
6 29.6 0.4 33.6 2.2
7 34.5 1.1 33.7 2
8 35.5 1.7 36.2 6.9

108
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 9. Relationship between SQ and r1

Table 7. Computed values of r1 and OQ

Sample No. Normal Voice Pathological Voice

Viscous resistance (r1) OQ Viscous resistance (r1) OQ
1 14.9 0.305 14.9 0.243
2 16.7 0.201 16.2 0.224
3 19.5 0.230 19.6 0.233
4 22.6 0.268 27.3 0.330
5 29.5 0.124 31.1 0.521
6 29.6 0.247 33.6 0.446
7 34.5 0.304 33.7 0.299
8 35.5 0.349 36.2 0.338

and CIQ. The Viscous resistance r1 for pathological voice is substantially greater than normal voice.
This leads to a stickier surface of vocal folds.
The physical parameters reflect the physical characteristics in the physiological system but their
computation is complex. In the detection of voice pathology, the glottal parameters often perform well
and the estimation method is simple. In this work, the glottal parameters are estimated for normal
and pathological voice samples. Further, these parameters have been related to physical parameters.
As seen from results, these parameters are effective in detecting vocal folds disorder as presented
above in Table 9.
In subsequent sections, a method based on acoustic and electrical circuits is proposed to
differentiate between normal and pathological voices.
Flanagan and Landgraf in 1968(Flanagan & Landgraf, 1968) were the first researchers who
modeled the acoustic behavior of vocal folds using one mass model and assumed lungs as a constant-
pressure source should it be denoted as Ps . Further Van den Berg (Van den Berg, 1958), experimented
on one mass model and studied the impact of constant and variable pressure on air volume velocity

109
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 10. Relationship between OQ and r1

Table 8. Computed values of r1 and CIQ

Sample No. Normal Voice Pathological Voice

Viscous resistance (r1) CIQ Viscous resistance (r1) CIQ
1 14.9 0.037 14.9 0.167
2 16.7 0.051 16.2 0.073
3 19.5 0.057 19.6 0.069
4 22.6 0.076 27.3 0.065
5 29.5 0.170 31.1 0.016
6 29.6 0.244 33.6 0.35
7 34.5 0.100 33.7 0.118
8 35.5 0.106 36.2 0.110

(Ug) and express the time-varying glottal impedance as function of viscous non-flow dependent
resistance (Rv), a kinetic flow dependent resistance (Rk) and inertance (Lg).
The acoustic circuit of vocal folds based on two- mass model (Ishizaka & Flanagan, 1972) is
shown in Figure 12. The elements of the acoustic circuit are defined as,

µL2d1
Rv1 = 12 (19)
Ag31

µL2d2
Rv2 = 12 (20)
Ag32

110
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 11. Relationship between CIQ and r1

Table 9. Average value of physical and glottal parameters

Physical and glottal Normal Voice Pathological Voice (%) increase in pathological
Parameters parameters

NAQ 0.03 0.05 66

SQ 2.07 3.36 62

OQ1 0.25 0.32 28

CIQ 0.08 0.10 25

k1(kdyn/cm) 139.9 154.4 10

r1 25.3 26.5 04

ρd1
Lg1= (21)
Ag 1

ρd2
Lg2= (22)
Ag 2

0.19ρ
Rk1= (23)
Ag21

111
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 12. The acoustic circuit representation for the vocal folds (Ishizaka & Flanagan, 1972)

  A    A 
   
ρ  0.5 −  g 2  1 −  g 2 
  A1    A1 
Rk2=   (24)
Ag22

Where,

d1 and d2: thickness of mass m1& m2,respectively,

Ag1and Ag2: cross-sectional areas at the two masses (m1& m2),
Lg: effective length of the vocal folds,
Rv1and Rv2: viscous non-flow dependent resistances,
Lg1 and Lg2: inertances due to the two masses (m1& m2),
Rk1 and Rk2: kinetic flow dependent resistances
Viscosity of air (μ) = 1.86*10-4 dyne-sec/cm2, at 300C
Air density (ρ) = 1.14*10-3 g/cm3, 300C

In the acoustic circuit given in Figure 12, if there is any change in subglottal pressure or in the
values of components, the volume velocity (Ug) changes. Change in Ug reflects the change in the
fundamental frequency of normal or pathological voices. From the above circuit, it can be shown that:

dU g
(R k1
+ Rk 2 ) U g U g + (Rv 1 + Rv 2 )U g + (Lg 1 + Lg 2 ) = Ps (25)
dt

Assuming constant lung pressure Ps of the acoustic circuit, if the length of the vocal folds (L)
changes, the values of components of acoustic circuit also changes {eqns. 19 to 24}. The changed
values of the components imply a change in volume velocity Ug and consequently, change in the
fundamental frequency. Table 10 shows values of Rv1, Rv2, Rk1, Rk2, Lg1, and Lg2 w.r.t different vocal
folds length (L) and a corresponding change in fundamental frequencies for normal and pathological
voices.

112
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Table 10. Computed values of Rv1, Rv2, Rk1, Rk2, Lg1 and Lg2 for varying vocal lengths (L)

Normal Voices
L (cm) Rv1 Rv2 Rk1 Rk2 Lg1 Lg2 F0
0.94 0.0039 0.118 0.00095 0.063 0.60 0.60 180
0.8 0.0055 0.13 0.0013 0.088 0.70 0.70 210
1.6 0.0027 0.069 0.00035 0.021 0.35 0.35 102
1.23 0.0036 0.09 0.00056 0.037 0.46 0.46 138
0.78 0.00057 0.14 0.0014 0.09 0.73 0.73 217
1.7 0.0025 0.064 0.00033 0.017 0.32 0.32 98
1.42 0.0032 0.077 0.00042 0.027 0.40 0.40 119
0.93 0.0039 0.11 0.00095 0.063 0.60 0.60 181
Pathological Voices
L (cm) Rv1 Rv2 Rk1 Rk2 Lg1 Lg2 F0
0.82 0.0055 0.13 0.0013 0.088 0.71 0.71 205
0.76 0.0057 0.14 0.0014 0.09 0.73 0.73 221
1.4 0.0032 0.077 0.00042 0.027 0.40 0.40 120
1.0 0.0044 0.108 0.00084 0.055 0.57 0.57 167
0.89 0.0057 0.14 0.0013 0.088 0.71 0.71 190
1.7 0.0027 0.067 0.00037 0.023 0.38 0.38 99
1.26 0.0040 0.108 0.00084 0.055 0.57 0.57 134
0.82 0.0057 0.14 0.0014 0.09 0.73 0.73 206

Equation 26 relates fundamental frequency (F0) and vocal folds length (L). Decrease in L results
into increase in F0.

 
 8.67   A2  4.61L 0 
L

F0 =   1 + 5.69 e (26)
 L   L2 

Where,

L0 is abducted reference length,

A is vibration amplitude

For the human vocal folds, amplitude to length (A/L) ratio is of the order of 0.1 and 0.5<L/
L0<1.0(Titze, 1989a).
Another simpler relationship connecting fundamental frequency (F0) and vocal folds membranous
length (Lm) is (Titze, 1989b) shown in Equation 27.

1700
F0 = (27)
Lm

113
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Table 11 and Figure 13 show that if vocal folds length reduces, the fundamental frequency
increases.

Table 11. Computed values of fundamental frequency and vocal folds length

Normal Voices Pathological Voices

*fundamental frequency(F0) ** vocal folds length *fundamental frequency(F0) ** vocal folds length
Hz L(cm) Hz L(cm)
98 1.7 99 1.7
102 1.6 120 1.4
119 1.42 134 1.26
138 1.23 167 1
180 0.94 190 0.89
181 0.93 205 0.82
210 0.8 206 0.82
217 0.78 221 0.76
*Fundamental frequencies are obtained using Aalto Aparat tool.
** Length of vocal folds is obtained using equation 27.

Figure 13. Fundamental frequency vs Vocal folds Length

Now, if the lung pressure Ps in the acoustic circuit changes then volume velocity (Ug) also changes
and leading to change in fundamental frequency. These observations are depicted in Table 12 and
Figure 14. Here it is assumed that the values of circuit components are constant.
The value of lung pressure Ps is calculated with help of following relationships as given by Hocine
Teffahi (Baken & Orlikoff, 2000),

F0 (Hz) = 2.3Ps + 48Q +1.98 (28)

114
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Table 12. Computed values of fundamental frequency and subglottal pressure using Equation 28.

Normal Voices Pathological Voices

*fundamental **subglottal pressure (Ps) in *fundamental frequency(F0) **subglottal pressure
frequency(F0)Hz cmH2O Hz (Ps) in cmH2O
180 14.7 205 25.6
210 27.8 221 32.6
102 4.9 120 5.2
138 6.2 167 9.1
217 30.8 190 19.1
98 4.5 99 4.5
119 5.2 134 6.1
181 14.8 206 26
*Fundamental frequencies are obtained using Aalto Aparat tool
**subglottal pressure Ps is obtained using Equation 28

Figure 14. Fundamental frequency vs subglottal pressure

Assume, Q (=3) is vocal cord tension (Teffahi, 2009), which is proportional to Fundamental
frequency (F0).
The equivalent electrical circuit derived from the acoustic circuit (Figure 12) is shown in Figure
15. Circuit components values are a function of the length of the vocal fold. The voltage (V) of the
circuit which is equivalent to lung pressure (Ps) is calculated by using the following relation (Baken
& Orlikoff, 2000),

1cmH2O Pressure = 1.27 * Palv + 5.94 (29)

where Palv is the alveolar pressure approximated as dc voltage equivalent.

The mathematical model of the circuit is given by the following relation:

115
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 15. Equivalent electrical circuit for the vocal folds

dI
(R k1
+ Rk 2 ) I I + (R v1 + R v2 ) I + (L g1 + L g2 ) =V (30)
dt

To analyze the above circuit two scenarios are considered.

a) When voltage is constant and components values are variable: The current in the circuit is
different for normal and pathological voices. Current variations are inversely related to variations
in fundamental frequencies of normal or pathological voices.

Figure 16 shows the variations of output current (I) of the equivalent electrical circuit as a
function of the vocal folds length for normal and pathological voices. It is clear from Figure 16 that
as the vocal folds length is decreasing the output current (I) is also decreasing.

b) When voltage is variable and components values are constant: The current in the circuit is
different for normal and pathological voices. This is inversely related to variations in fundamental
frequencies of normal or pathological voices

6. CONCLUSION AND FUTURE SCOPE

Improvement of health and non-invasive diagnosis and treatment of chronic diseases are the major
requirements in Biomedical Field. Voice disorders can have a significant negative impact on the
social and professional life. Although such disorders are often underestimated, their early detection
and accurate diagnosis are necessary to reduce serious consequences. The health of persons may be
severely affected by their individual pathological voice conditions. This may financially burden such
patients and even the society at large. One of the most frequently utilized tools to diagnose these
vocal disorders is a laryngoscope. Laryngoscopy, an invasive and painful technique, is an expensive
time-consuming process that requires trained personnel to perform the test. To address these issues,
researchers have been experimenting with non-invasive techniques for detecting vocal disorders.
This paper has successfully presented an alternate non invasive diagnostic method to accurately and
quickly classify voice disorders. The proposed method provides an opportunity to further improve
the existing medical techniques that are necessary to diagnose voice disorders. The results obtained
in this paper reveal the essence and mechanism for the pathological voice and establish a theoretical

116
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Figure 16. Trend of output current of electrical circuit for normal and pathological voices

and experimental foundation for pathological voice detection. Physical & glottal parameters are
defined, related and their impact on normal & pathological voices has been presented in the paper.
Stiffness (k1) and Viscous Resistance (r1) are higher for pathological voices. The Dependence of glottal
parameters on physical parameters is also studied and verified. It is also concluded that if k1, r1, or
both change then F0, and NAQ, SQ, OQ and CIQ change. All these variations enable us to distinguish
between normal and pathological voices. Compared to normal voices, the pathological voices have
shown increased values of both physical & glottal parameters. Further acoustic & electrical models
have been synthesized that also detect pathological voices. As future work, authors plan to develop
database and try other classification methods to reach a definitive diagnosis for Vocal folds disorders.

ACKNOWLEDGMENT

The authors are thankful to the Special Manpower Development Program, Chip-to-System Design
(SMDP-C2SD), initiated by the Ministry of Electronics & Information Technology (MeitY), Govt.
of India, for providing lab facilities in the School of VLSI Design and Embedded Systems, NIT,
Kurukshetra.

117
International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

REFERENCES

Amami, R., & Smiti, A. (2017). An incremental method combining density clustering and support vector
machines for voice pathology detection. Computers & Electrical Engineering, 57, 257–265. doi:10.1016/j.
compeleceng.2016.08.021
Baken & Orlikoff. (2000). Clinical measurement of speech & voice (2nd ed.). Cengage Learning.
Barry, W. J., & Putzer, M. (2015). Saarbrucken Voice Database. Institute of Phonetics, University of Saarland.
http://www.stimmdatenbank.coli.uni-saarland.de/
Cataldo, Leta, Lucero, & Nicalato. (2006). Synthesis of voiced sounds using low-dimensional models of the
vocal cords and tim-varying subglottal pressure. Mechanical Research Communications, 33, 250-260. doi:
10.1016/j.mechrescom.2005.05.007
Flanagan, J., & Landgraf, L. (1968). Self Oscillating source for vocal-tract synthesizers. IEEE Transactions on
Audio and Electroacoustics, 16(1), 57–64. doi:10.1109/TAU.1968.1161949
Forero, Kohler, Vellasco, & Cataldo. (2014). Classification of Vocal Aging Using parameters extracted from
glottal signal. Journal of Voice, 28, 532-537. doi: 10.1016/j.jvoice.2014.02.001
Hirano. (1974). Morphological structure of the vocal cord as a vibrator and its variations. Folia Phoniatr., 26(2),
89–94. doi: 10.1159/000263771
Ishizaka, K., & Flanagan, J. L. (1972). Synthesis of voiced sounds from a two-mass model of the vocal cords.
The Bell System Technical Journal, 51(6), 1233–1268. doi:10.1002/j.1538-7305.1972.tb02651.x
Isshiki, N. (1977). Functional Surgery of the Larynx. Kyoto University.
Kaneko, T., Asano, H., Naito, J., Kobayashi, N., Hayashi, K., & Kitamura, T. (1972). Biomechanics of the vocal
cords-on damping ratio. J. Jpn. Bronchoesophagol. Soc., 25(3), 133–138. doi:10.2468/jbes.25.133
Mittal, V., & Sharma, R. K. (2020). Voice Signal Analysis with the Application in Biomedicine. Sensor Letters,
18(2), 122–127. doi:10.1166/sl.2020.4187
Mokhkari, Story, Alku, & Ando. (2018). Estimation of the glottal flow from speech pressure signals: Evaluation
of three variants of iterative adaptive inverse filtering using computational physical modeling of voice production.
Speech Communication, 104, 24-38. doi: 10.1016/j.specom.2018.09.005
Ramsay. (2019). Mechanical speech synthesis in early talking automata. Acoustics Today, 15, 11-19. doi:
10.1121/AT.2019.15.2.11
Riede. (2011). Subglottal pressure and fundamental frequency control in contacts calls of juvenile alligator
mississippiensis. doi:10.1242/jeb.051110
Teffahi, H. (2009). A Two-Mass model of the vocal cords: determination of control parameters. Proceeding
International Conference on Multimedia Computing and Systems. doi:10.1109/MMCS.2009.5256726
Teixeira, J. P., Alves, N., & Fernandes, P. O. (2020). Vocal Acoustic Analysis:ANN Versos SVM in Classification
of Dysphonic Voices and Vocal Cords Paralysis. International Journal of E-Health and Medical Communications,
11(1), 37–51. doi:10.4018/IJEHMC.2020010103
Titze, I. R. (1989a). On the relation between subglottal pressure and fundamental frequency in phonation. The
Journal of the Acoustical Society of America, 85(2), 901–906. doi:10.1121/1.397562 PMID:2926005
Titze, I. R. (1989b). Physiologic and acoustic diffrences between male and female voices. The Journal of
Acoustical Society of America, 85, 1699-1707. doi: 10.1121/1.397959
Van den Berg, J. (1958). Myoelastic-aerodynamic theory of voice production. Journal of Speech Hearing
Research, 1, 227-244. doi: 10.1044/jshr.0103.227
Yao, X., Jitsuhiro, T., Miyajima, C., Kitaoka, N., & Takeda, K. (2015). Modeling of Physical Characteristics
of Speech under stress. IEEE Signal Processing Letters, 22(10), 1801–1805. doi:10.1109/LSP.2015.2434732

118
 International Journal of E-Health and Medical Communications
Volume 12 • Issue 4 • July-August 2021

Yao, X., Xu, N., Liu, X., Jaing, A., & Zhang, X. (2018). Research on speech under stress based on glottal source
using a physical model. IEEE Access : Practical Innovations, Open Solutions, 6, 44473–44482. doi:10.1109/
ACCESS.2018.2860130

Vikas Mittal received his M.Tech in Electronics and Communication Engineering from Kurukshetra University
Kurukshetra. Currently, he is pursuing Ph.D. at the National Institute of Technology (NIT) in the School of VLSI
Design and Embedded Systems. His research interests include Biomedical Signal Processing and Circuits Design.

R. K. Sharma received his M.Tech in Electronics and Communication Engineering and PhD degree in electronics and
communication from Kurukshetra University, Kurukshetra (through National Institute of Technology Kurukshetra),
India in 1993 and 2007, respectively. Currently, he is a Professor with the Department of Electronics and
Communication Engineering, NIT Kurukshetra, India. His main research interests are in the field of embedded
applications, low power, digital design, and disease/ stress detections using voice profiling of human beings.

119