Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Voice Gender Recognition

Sakshi Malhotra1 Kushagra Singh2
Department of Information Technology Galgotias Department of Information Technology Galgotias
College of Engineering and Technology College of Engineering and Technology
Greater Noida, India Greater Noida, India

Mukul Nag3 Kuldeep Nirmal4

Department of Information Technology Galgotias Department of Information Technology Galgotias
College of Engineering and Technology College of Engineering and Technology
Greater Noida, India Greater Noida, India

Abstract:- Due to its widespread application in a variety II. LITRATURE REVIEW

of circumstances, the gender classification system has
become more and more relevant including social media After studying the existing publications that are
platforms and criminal investigations. Prior studies in relevant to the idea of our proposed system, we found out
this area have mostly focused on discrimination against that a huge amount of voice gender recognition models are
both men and women. Nevertheless, since transgender built on Support Vector Machine (SVM), CART, Random
persons have just received legal recognition, it has been Forest, and deep learning techniques like Multilayer
vital to create techniques for accurately diagnosing Perceptron (MLP) and GBM [1][2][3][7][8][9]. Several
gender from a specific voice, which can be a challenging programs have helped with gender classification in the past,
undertaking. To extract pertinent characteristics from a but only for men and females, not transgender people One
training set that may be utilised to create a model for of the findings was that it uses gender and language to
gender categorization, researchers have employed a identify the language of spoken utterances and identify the
number of techniques. Following then, a vocal signal’s speaker’s gender based on their voice. Some of them use
gender may be ascertained using this model. The study orthographic transcription accomplished by recording and
makes three significant contributions: first, it provides a analyzing the speaker’s speech using the Gaussian Mixture
thorough analysis of well-known voice signal features Model and MFCC feature ex- traction called automatic
using a well-known dataset; second, it investigates a speech recognition (ASR) and semi- supervised learning
variety of machine learning models from a variety of [4][5][13] focus male and female only. The transformer input
theoretical families to classify voice gender; and third, layer is used to implement ASR and Trans- former Keras
it uses three well-known feature selection algorithms to voice recognition, Librosa, Mel Spectrogram which
select the features that have the greatest potential to interprets audio waves to identify transgenders from male
improve classification models. and female voice datasets. In order to create classifiers that
are more accurate, current research focuses on fusing
Keywords:- Python, Machine Learning, Transformer, Ten- ensemble learning strategies with semi-supervised learning
Sorflow, Spectrogram, Matplotlib, Pandas, HTML, CSS, frameworks. Making emotional speech understandable in
Django. order to determine the speaker’s gender makes the challenge
even more intriguing [6][10]. Recently, a number of cutting-
I. INTRODUCTION edge gender recognition methods based on several
biometrics, in- cluding the face, body form, and voice, have
Gender can be determined by using speech and voice been presented. The worst one of them is relying solely on
recognition technology. Based on the frequency and voice. Voice verification, gender categorization from voice,
volume of a person’s voice and speech, the ear has an and native (mother tongue) linguistic context were all
excellent system for determining their gender. A explored. Four classifiers are used in a stacked ensemble for
method known as machine learning voice recognition gender voice identification as the basis classifiers: LR,
employs machine learning algorithms to assist computers in KNN, SVM, SGD, and LDA[11][12][14].
comprehending and interpreting spoken language. The
algorithm is fed a lot of labelled data during the III. PROPOSED SYSTEM
training process, allowing it to learn from examples and
get better over time. When the AI calculation has been The VOICE GENDER RECOGNITION proposes to
prepared, it tends to be utilized to perceive and decipher solve the classification of genders through voice not only in
human discourse continuously. Virtual assistants, speech- males and females but also to include transgenders. System
to-text transcription, language translation, and voice- design of multiple technologies are included in voice gender
controlled devices like smart speakers and home automation recognition. The system uses Python and machine learning
systems are just a few of the many uses for this technology. for the backend and HTML, CSS, Django, and JavaScript for
the front-end development. Here, algorithms will help in the

IJISRT23MAY1247 www.ijisrt.com 1451

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
classification of the genders by matching the extracted data C. CSV Dataset
from the provided input with the stored dataset. In order to Wav file is a sound format that can be played in
host the programme on the website, a user interface is also Windows Media Player, QuickTime and even iTunes. The
created. The Voice Gender Recognition GUI is designed in .wav exten- sion stands for Waveform Audio File and it was
the first place, on a website. Users can offer speech data to created by Microsoft.
be recognised as input on the website thanks to a voice fetch
mechanism. The model is created using machine learning  Implementation
methods in the way that follows: This project’s primary goal is to address the issue of
trans- gender voice gender recognition. The implementation
 Extraction of data Machine learning is a process in makes use of the following programmes, devices, or
whichthe data is extracted from an original source, and frameworks:
then it is processed to obtain information. In order to
extract the data from the .wav audio file, we use D. Web Design
Spectrogram. Any system includes a large portion called the
 Training the data Fitting a model to fresh data is the Graphical User Interface (GUI). An Audio file entry Page
process of machine learning. The main aim of the has been made with the aid of HTML, CSS, Django, and
training is to identify the model’s ideal parameters. To JavaScript utilising the UI of VOICE GENDER
discover these parameters in practice, we often employ RECOGNITION. The website is utilised to get user voice
an opti- misation approach (such as gradient descent). data in.wav file format. The audio file is further processed
Machine learning training is the process of developing a as input speech to the machine learning model, which in the
model from unstructured data. We use Python, backend performs pattern-matching algorithms to give the
Spectrogram, Transformer, TensorFlow, matplotlib, most accurate result possible which is then displayed to the
PyCharm, NumPy, Pandas and Jupyter Notebook. user.
 All the acoustic features are extracted with the use of
librosa library, it is then stored in .csv file for further
processing. Pattern matching is the collection of tasks
performed by various Machine Learning tools like pan-
das, matplotlib etc., for identifying trends.

A. Supervised Learning and Classification of Rules

One can classify rules based on the information they
contain with supervised learning. It can be very useful for
finding patterns in our data that will help us decide
which rule to apply to a new situation or case more
effectively.

B. Voice Gender Recognition Fig 2 GUI Implementation

The ML-based voice identification tool takes the .wav
audio file as input from the user, and will first extract the E. Workflow Graph
features using R Programming, it is then stored in .csv file Figure 4, explains how the whole workflow is done in
for further processing. Pattern matching is performed to the whole project. The graph completes each step of the
identify the trends by using various ML tools like project very accurately. The training phase describes how
the model
matplotlib, and pandas etc., Based on the pattern matchings
and various trends the decision logic algorithms like SVM,
XGBoost etc., come into play for identifying gender and
put the final decision.

Fig 3 Web UI for input

Is being trained for male, female, and transgender

voices and which is then tested with an untouched dataset in
Fig 1 System Flow Diagram the testing phase.

IJISRT23MAY1247 www.ijisrt.com 1452

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
Text summarization, sentiment analysis, and language
trans- lation. The self-attention feature enables the model to
weigh different parts of the input in the representation of the
input sequence. To do this, attention weights are calculated
for each point in the input sequence depending on how
similar the current position is to all other positions in the
series. There are two types of transformers: an encoder and
a decoder. The former processes the input sequence,
generating hidden representations for each input position.
The latter generates the output sequence, based on the
encoder’s hidden rep- resentations. Unlike RNNs that
process inputs sequentially, transformers can parallelly
process input sequences, making it one of their key
advantages. Many computer machine learning systems,
including those that comprehend human speech, do picture
analysis, and employ voice recognition, frequently use a
component of the apparatus called the spectrograph.

H. Spectrogram
Machine learning methods can be used to analyse
spectral data and extract relevant features from the data.
These algo- rithms can find patterns in spectral data that
are challenging to notice with the naked eye. By examining
the patterns and correlations in the spectrum data, machine
Fig 4 Web UI for Input
learning algorithms are able to find individual
spectroscopic fingerprints that are unique to particular
F. Workflow & Process Definition for Final Model
materials or compounds. Spectrum data analysis and
The process of analyzing a speech signal to extract
insight extraction are now made possible by machine
relevant information in a form that is smaller than the
learning, which was previously impractical or impossible to
speech signal it- self is known as speech analysis. Multiple
do. As spectroscopic research develops, machine learning is
application domains make extensive use of AI and machine
anticipated to play a bigger role in the analysis and
learning. Vectors of features An ordered list of numerical
interpretation of spectrum data.
properties of observed phenomena is called a feature vector.
A prediction-making ma- chine learning model uses it as
I. Dataset
input features. Decisions can be made by humans by
Here we are Using CSV Files for our model. The
analyzing qualitative data. A conceptual framework that
fields meanfreq, sd, median, Q25, Q75, IQR, skew, Kurt,
standardizes communication between diverse networks is
sp. ent,sfm, mode, centroid, meanfun, minfun, maxfun,
provided by reference models.
meandom, mind,maxdom, range, modding, label correspond
to the sam- ple’s gender. Acoustic qualities are specified in
G. Transformer Keras Voice recognition
the remaining fields. Along with the pre-processed dataset,
A transformer is a machine-learning technique that
the training data also includes the raw voice samples.WAV
utilizes self-attention mechanisms for sequential input data
files kept in a different location.
processing. It has become more widely used as a method for
completing numerous natural language processing (NLP)
tasks, including

Fig 6 Architecture Diagram

Fig 5 Workflow & Process Definition for Final Model

IJISRT23MAY1247 www.ijisrt.com 1453

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
J. Pearson Correlation of Features

Fig 9 GUI Representing Result

V. CONCLUSION & FUTURE WORK

VOICE GENDER RECOGNITION is the tool we are

building to identify the gender of a person based on vocal
data and information. In this paper we are presenting the
design, the proposed system in this paper and the
implementation. Speech and sound specialists have found it
challenging to deduce a person’s gender based on their
voice, even when employing a number of technologies,
Fig 7 Pearson Correlation of Features
such as CRM systems’ Effective advertising and marketing
strategies. Investigating the voice of the culprit in crime
The Pearson correlation measures the degree to which
scenes. improving conversation systems and other human-
two variables have a linear relationship. A value of -1
computer interaction (HCI) technologies. In the medical
indicates a total negative linear correlation, a number of 0
sector, diagnosing people with voice issues may be highly
means there is no association, while a value of +1 means
helpful. By taking over duties that don’t require humans, it
there is a completelypositive correlation.
increases efficiency. The purpose of acousticcharacteristics is
to react to sound waves. Gender recognition is a method for
IV. RESULT
figuring out a speaker’s gender category. The length,
strength, frequency, and filtering of an acoustic signal may
After giving the model a defined input under accurate
be learnt from the signals of a recorded voice. Adaptation of
bounds, the output is generated in terms of Male, Female
music for waiting rooms so that different kinds of music can
and Transgender. The précised output is displayed to the
be played depending on age and gender. It can help in
user through GUI:
various other fields like Robotics, finance and banking,
computer vision, etc.

ACKNOWLEDGMENT

We appreciate Ms Shakshi Malhotra, our project

mentor, for supporting us and assisting us in resolving some
of the most difficult difficulties in creating our project.

REFERENCES

[1] P. Gupta, S. Goel and A. Purwar, ”A Stacked

Technique for Gender Recognition Through Voice,”
2018 Eleventh International Conference on
Contemporary Computing (IC3), Noida, India, 2018,
pp. 1-3, doi: 10.1109/IC3.2018.8530520.
[2] G. Sharma and S. Mala, ”Framework for gender
recognition using voice,” 2020 10th International
Conference on Cloud Computing, Data Science &
Engineering (Confluence), Noida, India, 2020, pp. 32-
Fig 8 Result

IJISRT23MAY1247 www.ijisrt.com 1454

Volume 8, Issue 5, May – 2023 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
37, doi: 10.1109/Confluence47617.2020.9058146. [13] Al khammash, Eman & Hadjouni, Myriam &
[3] K. Zvarevashe and O. O. Olugbara, ”Gender Voice Elshewey, Ahmed. (2022). A Hybrid Ensemble
Recognition Using Random Forest Recursive Feature Stacking Model for Gender Voice Recognition
Elimination with Gradient Boost- ing Machines,” Approach
2018 International Conference on Advances in Big Electronics.11.1750.10.3390/electronics11111750.
Data, Computing and Data Communication Systems [14] Fahmeeda, Sayyada Ayan, Mohamed & Shamsuddin,
(icABCD), Durban, South Africa, 2018, pp. 1-6, doi: Mohamed & Amreen, Aliya. (2022). Voice-Based
10.1109/ICABCD.2018.8465466. Gender Recognition Using Deep Learning.
[4] W. Li, D. -J. Kim, C. -H. Kim and K. -S. Hong, International Journal of Innovative Research &
”Voice-Based Recognition System for Non- Growth. 3.649- 654.
Semantics Information by Language and Gender,”
2010 Third International Symposium on Electronic
Commerce and Security, Nanchang, China, 2010, pp.
84-88, doi: 10.1109/ISECS 2010.27.
[5] N. M and A. S. Ponraj, ”Speech Recognition with
Gender Identification and Speaker Diarization,”2020
IEEE International Conference for Inno- vation in
Technology (INOCON), Bangluru, India, 2020, pp.
1-4, doi: 10.1109/INOCON50539.2020.9298241.
[6] Livieris, Ioannis & Pintelas, Emmanuel & Pintelas,
P.. (2019). Gen- der Recognition by Voice using an
Improved Self-Labeled Algo- rithm. Machine
Learning and Knowledge Extraction. 1. 492-503.
10.3390/make1010030.
[7] Büyükyılmaz, Mücahit & Ç ıbıkdiken, Ali. (2016).
Voice Gender Recog- nition Using Deep Learning.
10.2991/msota-16.2016.90.
[8] S. Chaudhary and D. K. Sharma, ”Gender
Identification based on Voice Signal
Characteristics,” 2018 International Conference on
Ad- vances in Computing, Communication Control
and Networking (ICAC- CCN), Greater Noida, India,
2018, pp. 869-874, doi: 10.1109/ICAC-
CCN.2018.8748676.
[9] L. Jasuja, A. Rasool and G. Hajela, ”Voice
Gender Recog- nizer Recognition of Gender from
Voice using Deep Neural Net- works,” 2020
International Conference on Smart Electronics and
Communication (ICOSEC), Trichy, India, 2020, pp.
319-324, doi: 10.1109/ICOSEC49089.2020.9215254.
[10] M. Kotti and C. Kotropoulos, ”Gender classification
in two Emotional Speech databases,”2008 19th
International Conference on Pattern Recognition,
Tampa, FL, USA, 2008, pp.1-4, doi:
10.1109/ICPR.2008.4761624.
[11] M. Wang, Y. Chen, Z. Tang and E. Zhang, ”I-vector
based speaker gen- der recognition,”2015 IEEE
Advanced Information Technology, Elec- tronic and
Automation Control Conference (IAEAC),
Chongqing, China, 2015, pp. 729-732,
doi:10.1109/IAEAC.2015.7428651.
[12] O. Iloanusi et al., ”Voice Recognition and Gender
Classification in the Context of Native Languages
and Lingua Franca,” 2019 6th In- ternational
Conference on Soft Computing & Machine
Intelligence (IS- CMI), Johannesburg, South Africa,
2019, pp.175-179, doi: 10.1109/IS-
CMI47871.2019.9004306.

IJISRT23MAY1247 www.ijisrt.com 1455