Human Voice Recognition System to Predict the

Gender using Random Forest Algorithm

Veeramreddy Sunil Kumar Reddy Saravanan M.S Surendran R
Department of Computer Science Department of Computer Science Department of Computer Science
and Engineering, and Engineering, and Engineering,
Saveetha School of Engineering, Saveetha School of Engineering, Saveetha School of Engineering,
Saveetha Institute of Medical and Saveetha Institute of Medical and Saveetha Institute of Medical and
Technical Sciences, Technical Sciences, Technical Sciences,
Chennai, India. Chennai, India. Chennai, India.
[email protected] [email protected] [email protected]

.
A human voice contains different sound waves
Abstract- There is an effective strategy in this research and different frequencies, Gender identification is
work to predict the human voice recognition regardless based on voice that deploys various applications like
of gender involving Random Forest in correlation with advertising, marketing strategies [3]. Contrasted with
Novel Decision Tree calculation with further developed male and female the recurrence is lower in male and
precision. Random Forest algorithm is an AI based higher in female. The voice recurrence in male is in
calculation which is a type of Recursive Element the reach 80Hz to 180Hz, with a typical worth. For a
Disposal. The grouping precision of the Novel Decision
Female voice recurrence it goes from 165Hz to 255
Tree was further developed in the wake of applying the
Random Forest to the dataset. Experimental results
Hz [4]. To order the orientation by ascertaining the
show that Novel Decision Tree outflanked. Almost all mean frequency of the speech samples [5]. This
the near calculations in characterization exactness should be possible by utilizing AI via preparing
enabled orientation voice recognition to be a reasonable models with Random Forest and Novel Decision tree
possibility. Resources and Procedure: The unique algorithms Associated the whole thing of machine
calculations with G-power levels of 80% with 10 learning and AI have been functional with respect to
research cases. In order to categorize the voice dataset the conjecture of human voice affirmation whether
into male or female, various properties are taken into male or female [6].
account. Nearly 112 voice tests are included in the
training dataset. Male or female voice recognition may An examination article by them has utilized
be predicted with 97.79% accuracy according to different classification, like, Irregular Backwoods,
research by Random Forest. Result: This examination Original Choice Tree, Backing Vector Machine, and
concentrate on saw as 96.74% of exactness for an Convolutional Neural Network [7]. The neutral is to
expectation of human voice acknowledgment utilizing break down the estimations and catch which is fit to
the Novel Decision Tree algorithm with a measurably forecast the voice affirmation whether male or female
huge contrast among the double gatherings (p = 0.090; [8]. Orientation Distinguishing proof procedure that
p < 0.05) with a 95% assurance span. Conclude: These actions the speaker's discourse signal force to go with
study explanations that the Random Forest algorithm a choice. Voice power is determined utilizing
scheduled the prediction of human voice response is Simpson's standard to quantify region under
fundamentally well associated to the novel decision tree standardized bend [9]. An adjusted threshold has been
algorithm. utilized to pursue a novel decision of the kind of
Keywords- random forest, novel decision tree, machine orientation. Machine learning is an innovation that
learning, voice recognition, accuracy, gender comprises an extensive variety of utilizations utilized
identification.
in various fields to foresee values in money, banking
I. INTRODUCTION and promoting, and so on by utilizing calculations and
information to prepare the PCs [10]. The target of
A Gender Identification proof is a significant these trials is to give more precise and effective
capability which can expand the effectiveness of calculations for the arrangement.
different applications like showcasing, voice
acknowledgment and online notices including voice In any case, orientation acknowledgment by voice
[1]. Human speech is a good communication tool is as yet viewed as a troublesome and testing task for
which consists of various same features such as age, an exact expectation model [11]. Gender
language, voice and gender identification [2]. Identification depends on the area of vocal sound,
assuming it is over the limit the orientation type is
female in any case male. Our fundamental point is to
foresee the orientation with various calculations in AI
utilizing a voice informational collection [12]. The On the off chance that N comprises of occasions of
various voice collection were analyzed using modern exclusively one session,
tools, this creates a new research gap for this study. Return
Aim of this research study is to separate of male or else
female voice recognition prediction processes with Haphazardly pick Y % of the suitable dividing
machine learning algorithms. components now M
II. MATERIALS AND METHODS Choice the trademark F with the very best records
harvest to cut up on
In this investigation, focus on making use of two Make f little child hubs of M, M1,…., Mf ,where F
classes of characterization algorithms. Calculating the has f reasonable qualities (G1,...., Gg)
RF and the DT independently makes up Groups 1 and For A = 1 to f do
2, respectively. 56 sample sizes with 80% G power
Set the items in Mi to Ei, the spot Ei is all
values, a critical value of 0.05, and 95% confidence
circumstances in M that fit
intervals were obtained using the G power tool,
version 3.1.10, to forecast the size of each sample Gi
[13]. Call Build Tree (Mi)
End for
In this proposal, used kaggle.com, an online End if
community of data scientists and AI experts. A search
function allows clients to locate updated datasets. It Novel Decision Tree computation is one of the
also offers a free Representations Handling Unit and a directed AI estimations which can be used for both
flexible individual Jupiter scratch cushion portrayal and ID of human voice acknowledgment. It
environment. The planning dataset contains nearly is basically used for portrayal-type ID issues. A novel
112 voice tests with various boundaries and the
Decision Tree resembles a non- parametric mastering
dataset utilized here comprises 21 attributes and
computation due to the estimation now not awaiting
contains 12 highlights that can be utilized to foresee
human voice acknowledgment [14]. whatever about the necessary data.

Unlike regular Timberland, Irregular Timberland Pseudo code of Decision Tree

is categorized by the type of computations you tell it EFFORT: T1 where T = set of ordered cases
to be. Both clustering and relapse can benefit from the HARVEST: Decision Tree
use of RF. The preference variable is null in the RF Require: SØ, no.of_attributes > 0
order. The Gini Index, a rate functionality for Route BUILD TREE
analyzing pairs of components, must be computed recap
prior to implementing a Random Forest [15]. After a marGain + 0
scheme the Gini index, partitioning the data-
separate B -invalid
collection, and calculating all of its constituent parts, a
e Entrap (Properties)
root node is created to bring together the resulting
fragmented introduction. The tree is constructed via For entirely Credits an in T do
recursive splitting, with the terminal hub serving as acquire Data Increase (a, e) on the off chance that
the basis for the final assumption. procure > marGain,
max Gain obtain cut up Aa
The speech recognition pictures dataset is End if
analyzed using the Random Forest method, the tree End for
model, and the supplementary pseudocode. The Parcel (S. separate A)
datasets will be used as input by pseudocode, and the until all segments handled
output will be communicated via accuracy boundaries
and a collecting mechanism. B. Statistical Analysis
A. Random Forest pseudo code to produce c The Statistical investigation was completed
classification using the IBM SPSS tool. In order to investigate the
information, Random Forest and Novel Decision Tree
For A = 1 to b do
models are utilized. The SPSS tool was used to
Haphazardly design the instruction data E with choice
analyze the projected accuracy value for both
to create Ei
projected algorithms and existing algorithms. Among
Make a source hub, Mi containing Ei
the dependents variables stay precision, the picture
Call Build Tree (Mi)
measurement and the breadth of the number on the
End for
variety plate are objective factors.
Build Tree (N)
 10 10 90.11 87.25

Table 1. Comparison between RF and DT algorithm Table 2. Statistical results of RF and DT algorithms.
with N=10 samples of the dataset with the highest Mean accuracy value, standard deviation and standard
performance of 97.79 and 96.74 in the sample (when error mean for RF and DT algorithms are obtained for
N=1) using the dataset size = 100 and the 70% of 10 iterations. It is observed that the RF (95.98)
training and 30% of testing data. algorithm performed better than the DT (94.87)
Sample Dataset RF algorithm DT
algorithm.
(N) size / Rows Accuracy in % algorithm
in % Accuracy
in % Algorith Sample Me Standard Standard
ms (N) an Deviation Error Mean
1 100 97.79 96.74 (Accurac
y)
2 90 97.12 96.25

RF
3 80 95.28 95.45
algorithm 10 95. 1.39877 .44233
984
4 70 95.27 95.10
0
5 60 94.96 94.25

DT
6 50 93.88 93.85
algorithm 10 94. 1.46066 .46190
871
7 40 92.451 92.58
0
8 30 92.11 90.25

9 20 91.78 88.15

Table 3. The Independent sample t-test of the significance level RF and DT algorithms results with a statistically
significant difference between the two groups (p=0.090; p<0.05). Therefore both the RF and the DT algorithms have
a significance level less than 0.05 with a 95 % confidence interval.

Levene's Test T-test of Equality of Means 95% of the confidence

for Equality interval of the
of Variances Difference
Accuracy
Sig (2- Mean Std Error
t df
F Sig. tailed) Difference Difference Lower Upper

Equal
Variance .000 .090 1.740 18 .049 .099 1.11300 -.23062 2.45662
Assumed

Equal
Variance Not - - 1.740 17.966 .049 .099 1.11300 -.23080 2.45680
Assumed

III.
IV. RESULT total workload, respectively. The results are shown in
With the assistance of 10 instances and the Table 1. The RF calculation is thought to be
utilization of unique data, the RF machine learning significantly higher than the DT estimate.
algorithm and DT machine learning algorithm are Algorithms that consume RF , DT technologies to
contrasted. The dataset comprises 100 columns where anticipate human voice response and orientation
the accuracy of both the RF and DT computation are identification. The red variety bend addresses the base
obtained for 10 emphases. The dataset preparation and recurrence of the voice, dark variety bend addresses
testing processes account for 70% and 30% of the the greatest recurrence of the voice and green variety
bend addresses orientation distinguishing proof as Fig. 3. Shown the distance recurrence of human
displayed in Fig. 1. voice acknowledgment and its contrasting both
Blunder rate and n Worth.
Table. 1 below displays the findings of a statistical
study of 10 samples: The RF calculation generated
1.39 standard deviations with 0.44 standard error, in
contrast to be a DT calculation 1.46 standard
deviations and 0.46 standard error. Considering that
the significant result (p=0.090) is less than 0.05, our
hypothesis appears to have some validity. Changes in
the information values (the independent variable) are
shown to have an impact on the related yield values
(the dependent variables) in Table 3.
Tables 2 and 3 represent the data that was
gathered. The results of an independent samples t-test,
which showed a statistically significant difference
between the two groups (p=0.090; p0.05), confirm our
working hypothesis. As a result, it has been
demonstrated that there are significant differences
between RF and DT algorithms. A statistically
significant difference was found at the 0.05 level
when the accuracy of the two computations was
evaluated using an independent samples t-test. Able to
get an accuracy of 97.79% using the RF model. In an
A/B test, the proposed RF classifier performed better
Fig. 1. Graph represents that flow of voice recognition using RF than the alternate computation.
and DT algorithms. The red colour curve represents the minimum
frequency of the voice, black color curve represents the maximum As shown in Fig. 4, the mean accuracy of the RF
frequency of the voice and green color curve represents gender 95.98 and the DT 94.87 algorithms are compared, and
identification.
the results show that the RF algorithm is much
superior to the DT method.
In Fig. 2, human voice acknowledgment using RF
and Novel DT algorithms represents values like
density, mean, median, mode, and maximum. As well

Fig. 2. Graph represents the human voice recognition whether male or female using RF and DT algorithms like Density, mean, median, mode,
maximum
 Fig. 3. Graph represents distance frequency of human voice
recognition and its comparing both Error rate and n Value.
Fig. 4. Comparison of RF algorithm and DT in terms of mean
V. DISCUSSION accuracy. The mean accuracy of RF is better than DT and the
There is a general consensus that Random Forest standard deviation of RF is slightly better than the DT algorithm.
X-axis: (GROUPS) RF vs DT algorithm and Y axis: Mean
is superior to Novel Decision Tree due to its increased accuracy of prediction ±2 SD
precision [16]. When compared to the Novel Decision
Tree classifier, the Random Forest classifier has a VI. CONCLUSION
dramatic effect on the accuracy, speed, and efficiency The speech recognition dataset is employed in this
of the prediction process [17]. In this study, the experiment with the hope that Random Forest and
effectiveness of the distinctive approach DT algorithm Novel Decision Tree will recognize human voice. The
is combined [18]. First, the test data is displayed in Novel Decision Tree classifier has an accuracy value
DT format. At the forefront of cutting-edge, evidence- of 96.74, compared to the Random Forest classifier's
based research, and have successfully implemented accuracy value of 97.79. When determining whether a
speech recognition technology for the next phase, human voice is being detected as male or female, RF
wherein they believe DT is vastly superior to RF [19]. appears to be more reliable than DT.
To aid with this endeavour, a different classifier
REFERENCE
for the Human Voice Acknowledgment expectation
was implemented in a number of techniques, [1] Chaparro, Valentina, Alejandro Gomez, Alejandro Salgado,
O. Lucia Quintero, Natalia Lopez, and Luisa F. Villa. 2018.
including RF and DT algorithms [20]. Additionally, “Emotion Recognition from EEG and Facial Expressions: A
deep learning is getting closer, although Singh 2020 Multimodal Approach.” Conference Proceedings: Annual
used AI computations with 10 cases as data. Thus, International Conference of the IEEE Engineering in
when compared to the presentation of several Medicine and Biology Society. IEEE Engineering in
Medicine and Biology Society. Conference 2018 (July): 530–
methods, the proposed RF technique surpassed DT 33.Cook, Darren. 2016.
calculations. In contrast to this investigation, other [2] Practical Machine Learning with H2O: Powerful, Scalable
exploratory findings examined the whole picture of Techniques for Deep Learning and AI. “O’Reilly Media,
voice recurrence and only performed a small number Inc.”Dhakal, Parashar. 2018.
of tests on administrators while using the face voice [3] Novel Architectures for Human Voice and Environmental
area to predict voice recognition [21]. Sound Recognition Using Machine Learning
Algorithms.Fuente Garcia, Sofia de la, Fasih Haider, and
The Random Forest method, which was used to Saturnino Luz. 2021.
analyses this speech recognition dataset with notable [4] R. S. Sai Dinesh, R. Surendran, D. Kathirvelan and V.
highlights, is the only method included in this analysis Logesh, "Artificial Intelligence based Vision and Voice
study. Assuming that in order to achieve better results, Assistant," 2022 International Conference on Electronics and
Renewable Systems (ICEARS), Tuticorin, India, 2022, pp.
it investigates the behavior of these two meetings. In 1478-1483, doi: 10.1109/ICEARS53579.2022.9751819.
subsequent testing, the new characteristics will [5] “How Can Familiar Voice Recognition Be Intact If
actually want to predict whether male or female Unfamiliar Voice Discrimination Is Impaired? An
human voice recognition will be used with Random Introduction to This Special Section on Familiar Voice
Forest and Novel Decision Tree algorithms [22]. Recognition.” Neuropsychologia.
https://doi.org/10.1016/j.neuropsychologia.2018.04.003.
[6] Gantner, S., P. Caffier, P. Hulin, M. Fuchs, P. Kummer, and [20] Rajagopal, Sureshkumar, Tamilvizhi Thanarajan, Youseef
A. Lorenz. 2021. “[Singing Voice Handicap Index-12 : Alotaibi, and Saleh Alghamdi. "Brain Tumor: Hybrid Feature
Development and validation of a German version].” HNO, Extraction Based on UNet and 3DCNN." Computer Systems
August. https://doi.org/10.1007/s00106-021-01092-3. Science and Engineering, 2023, 45(2), 2093-
[7] García-Guerrero, M. Acebo, Javier Peña, Leire Zubiaurre- 2109. https://doi.org/10.32604/csse.2023.032488
Elorza, Danele Benítez, Anais M. Hernández, Ainara Oribe, [21] Taşkin, Ayşenur, Fatma Esen Aydinli, Koray Başar, Esra
and Natalia Ojeda. 2022. Özcebe, Sinem Şimşek, and Taner Yilmaz. 2020.
[8] “Voice, Body Cues and Facial Expression in Emotion “Investigating Validity and Reliability of the Turkish Version
Recognition of Spanish Children and Adolescents: The of Transsexual Voice Questionnaire (Male to Female).”
Validation of Bell-Lysaker Emotion Recognition Test.” The Journal of Voice.
Spanish Journal of Psychology 25 (February): e7.Jian, Liang, https://doi.org/10.1016/j.jvoice.2020.06.020.
and Rui Shu. 2017. [22] Wilkins, Clara L., Joseph D. Wellman, and Katherine D.
[9] Surendran, R., and B. Parvatha Varthini. "Inject an elastic Schad. 2017. “Gender Identification Measure.” PsycTESTS
Grid Computing Techniques to Optimal Resource Dataset. https://doi.org/10.1037/t60656-000. Young,
Management Technique Operations." J. Comput. Sci. 9.8 Miriama. 2016. Singing the Body Electric: The Human Voice
(2013): 1051-1060. and Sound Technology. Routledge.
[10] “Action Recognition and Human Interaction.” Human
Activity Recognition and Prediction.
https://doi.org/10.1007/978-3-319-27004-3_2. Mohapatra,
Smaranika. 2020.
[11] “A Novel Approach to Analyze and Predict the Crop Yield
Productivity Using Machine Learning Algorithms.” Journal
of Advanced Research in Dynamical and Control Systems.
https://doi.org/10.5373/jardcs/v12sp3/20201234.Nelson,
Rachel S., Elizabeth V. Lonsdorf, Karen A. Terio, Kaitlin R.
Wellens, Sean M. Lee, and Carson M. Murray. 2022.
[12] S. R, K. R and J. B, "Implementation of Dynamic Scanner to
Protect the Documents from Ransomware using Machine
Learning Algorithms," 2021 International Conference on
Computing, Electronics & Communications Engineering
(iCCECE), Southend, United Kingdom, 2021, pp. 65-70, doi:
10.1109/iCCECE52344.2021.9534855.
[13] Orlandi, Silvia, Carlos Alberto Reyes Garcia, Andrea
Bandini, Gianpaolo Donzelli, and Claudia Manfredi. 2016.
“Application of Pattern Recognition Techniques to the
Classification of Full-Term and Preterm Infant Cry.” Journal
of Voice: Official Journal of the Voice Foundation 30 (6):
656–63.
[14] Pisanski, Katarzyna, Agata Groyecka-Bernard, and Piotr
Sorokowski. 2021. “Human Voice Pitch Measures Are
Robust across a Variety of Speech Recordings:
Methodological and Theoretical Implications.” Biology
Letters 17 (9): 20210356.
[15] Roldan-Vasco, Sebastian, Andres Orozco-Duque, Juan
Camilo Suarez-Escudero, and Juan Rafael Orozco-Arroyave.
2021. “Machine Learning Based Analysis of Speech
Dimensions in Functional Oropharyngeal Dysphagia.”
Computer Methods and Programs in Biomedicine 208
(September): 106248.
[16] Singh, Jitendra. 2020. “Struggles for a Dignified Life: A
Documentation of Invisible Resistance of Kuchbandhiya
Community in Bundelkhand Region of Uttar Pradesh.”
Contemporary Voice of Dalit.
https://doi.org/10.1177/2455328x20922697.
[17] Smith, Gary A., Thitphalak Chounthirath, and Mark
Splaingard. 2020. “Comparison of the Effectiveness of
Female Voice, Male Voice, and Hybrid Voice-Tone Smoke
Alarms for Sleeping Children.” Pediatric Research.
https://doi.org/10.1038/s41390-020-0838-1.
[18] Riya, K. S., R. Surendran, Carlos Andrés Tavera Romero, and
M. Sadish Sendil. "Encryption with User Authentication
Model for Internet of Medical Things
Environment." Intelligent Automation & Soft Computing 35,
no. 1 (2023).
[19] Su, Yue, Kainan Ma, Xu Zhang, and Ming Liu. 2022. “Neural
Network-Enabled Flexible Pressure and Temperature Sensor
with Honeycomb-like Architecture for Voice Recognition.”
Sensors 22 (3). https://doi.org/10.3390/s22030759.