IRJET- A Raspberry Pi based Speaker Recognition System for Access Control

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 06 Issue: 03 | Mar 2019 www.irjet.net p-ISSN: 2395-0072
© 2019, IRJET | Impact Factor value: 7.211 | ISO 9001:2008 Certified Journal | Page 7412
A Raspberry pi Based Speaker Recognition System for Access Control
Nkolika O. Nwazor
Lecturer, Department of Electrical/Electronic Engineering, University of Port Harcourt, Rivers State, Nigeria
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This work presents the design of an automatic
speaker recognition(biometric) system using the raspberry
pi as the control unit. Speaker recognition is the process of
automatically recognizing who is speaking on the basis of
individual information included in speech waves. This
technique makes it possible to use the speaker's voice to
verify their identity and control access to services such as
voice dialing, banking by telephone, telephone shopping,
database access services, information services, voice mail,
security control for confidential information areas, and
remote access to computers. Such a speaker recognition
system has potential in many security applications. For
example, users have to say a PIN (Personal Identification
Number) in order to gain access to their office doors. By
checking the voice characteristics of the input utterance,
using an automatic speaker recognition system similar to
the one that we will describe the system is able to add an
extra level of security. The Bob spear algorithm is used in a
python library to achieve speaker recognition on the
raspberry pi.
Key Words: Raspberry pi, Automatic, Speaker
Recognition, Access Control, python, Bob Spear Algorithm
1. INTRODUCTION
In recent time, the issue of security and privacy has
become of utmost importance. Government agencies,
security agencies, and even private organizations seek to
make their assets safer and to restrict access to private
areas.
Biometric authentication is the verification of a user's
identity by means of a physical trait or behavioral
characteristic that can't easily be changed, such as a
person’s face” (Russel, 2005). This is the strength that the
biometric systems have. In fact, the value of most part of
the biometric properties of a person is assumed not to
change for a long period, so after their storage, it is
possible recognize someone through corporal and
behavioral characteristics.
In forensic applications, the use of fingerprint
identification can prove to be futile as it is quite easy these
days to put on hand gloves when committing a crime or to
forge another person’s fingerprint in the bid to frame the
individual.
Furthermore, it is quite difficult or almost impossible to
distinguish between twins when using the face recognition
approach for access control. This could lead to granting of
access to wrong persons and compromise the security of
the asset being guarded.
The use of voice biometrics for identification and
recognition is more secure as it is capable of filtering out
background noise and is not affected by temporary
changes because up to 50 different characteristics are
measured to create a voice print.
2. LITERATURE REVIEW
In the work (Sanchez, 2010), a speaker recognition tool for
handheld computers was developed. The tool is written in
C++ and it requires a single word to be used for the
training phase and the same word for the recognition
phase. It used the Euclidean distance to find the best
match. (CMUSphinx, 2014) is a solution originally for
speech recognition, but its modular design allows for
extension in order to perform voice biometrics. Other
solutions in Python such as voiceid (Google, 2011) and
Speaker Recognition (SPEAR) (Python, 2012) also exist.
Given the fact that the task of voice biometrics consists of
computationally expensive operations, such as feature
extraction and creation of models, systems that utilize a
server-based approach have been proposed. (Chowdhury,
2010) describes a distributed system for speaker
recognition is which uses Gaussian Mixture Model
Universal Background Model (GMM-UBM) models.
{Brunet et al, 2013 }proposed a method that performs the
entire speaker recognition process on an Android mobile
device by extracting the MFCC features, and storing them
as a distance vector. For the pattern-matching step, during
the testing phase, they compare the test samples and
extract their Euclidean distance. They used samples both
from a publicly accessible and a private database,
achieving promising accuracy results.
3. METHODOLOGY
The design of the Voice biometrics Module for Access
Control, using the Raspberry Pi Single board computer,
SBC is mostly software based. All the hardware
components are ready-made peripherals, which are
directly attached to the Raspberry Pi and configured for
the specific purposes.
Basically, the project can be classified as an Internet of
Things (IoT) project. It requires internet connectivity for it

to function. This internet connectivity can be provided by
a wireless dongle attached to the raspberry pi zero or a
LAN cable or a USB 3G dongle. In this project, a USB Wi-Fi
adapter is used to provide the raspberry pi zero with
internet connectivity.
This project is made up of fundamentally two parts:
i. The Hardware Part
ii. The Software Part
The fig.1 shows the different blocks that make up the
module.
3.1 The Hardware Module
Figure 1 shows the block diagram of the hardware module
of the system.
The different components that make up the module as
follows:
1. Raspberry pi zero: The Raspberry Pi Zero is a full-
fledged member of the Raspberry Pi family,
featuring the following:
a) A Broadcom BCM2835 application
processor
b) 1GHz ARM11 core (40% faster than
Raspberry Pi 1)
c) 512MB of LPDDR2 SDRAM
d) A micro-SD card slot
e) A mini-HDMI socket for 1080p60 video
output
f) Micro-USB sockets for data and power
g) An unpopulated 40-pin GPIO header
h) Identical pinout to Model A+/B+/2B
i) An unpopulated composite video header
j) Our smallest ever form factor, at 65mm x
30mm x 5mm
Raspberry Pi Zero runs Raspbian OS, a Debian
distribution. It also runs some Linux applications,
including Scratch, Minecraft and Sonic Pi.
2. Wi-Fi Dongle: The USB Wi-Fi dongle is use to
provide internet connectivity the raspberry pi
zero. It is a low power device and it can be
powered by the RPi Zero (Raspberry Pi Zero).
3. USB Mic: The USB Mic comes with an inbuilt
sound card and is used as an input device for the
voice recognition. It is low power and can run
comfortably off the power supplied by the
raspberry pi USB port.
4. USB Hub: The USB Hub is used to provide multiple
USB inputs to a single port on the raspberry pi.
The USB Hub used for this project is not at active
type, i.e. is not a powered USB hub. The Wi-Fi
dongle and the USB mic. are attached to the USB
hub.
5. OTG Cable: The On-The- Go (OTG) cable is used as
a converter for the micro USB port present on the
raspberry pi to a normal port. The cable is also
used as an extension for the peripheral devices
attached to the RPi (Raspberry Pi).
6. 5VDC Power Adapter: A standard 5V 2A Power
Adapter is used to power the raspberry pi zero.
7. Servo: The 5C servo motor is attached to the GPIO
pins of the raspberry pi. This is used to simulate a
door access when the speaker is recognized.
8. Memory card: This is the hardware which
contains the OS (Operating System) running on
the raspberry Pi. It is advisable to use at least an
8G Class10 Micro SD Card as the OS takes up to 4
gigabytes on installation. The Class 10 stands for
the read and write speed of the memory card.
As the power adapter supplies the raspberry pi zero with
power. The servo in turn takes its power from the
raspberry pi’s 5V Voltage output supply pin. As a pre-
registered command is spoken and the system identifies
the speaker, the program is written in such a way that it
send a pulse width modulated signal through the GPIO pin
in which the servo’s signal pin is connected to. This causes
the servo to rotate.
Fig. 2 is the circuit schematic of all the peripheral devices
used with the raspberry pi zero:
Fig-1: Block Diagram of Voice biometric system
for Access Control

Figure 2: Circuit Schematic of the Speaker Authentication
System
3.2 Software Module
The software used to accomplish the objective of this
project are as follows:
i. The Raspbian Jessie Pixel: This is a Linux distribution
which is more of a combination of Debian and
Raspberry. It is built specifically to run on the Raspberry
Pi. It has a desktop view and also a command line
interface.
ii. The SD Card Formatter: This is a Windows-based
software which is used to format the microSD card in
which the Raspbian image is going to be mounted on.
iii. Etcher: This is the software used to write the Raspbian
image to the Raspberry Pi microSD card.
iv. Putty: Putty is a tool that enables one to connect to the
command line interface (CLI) of the raspberry pi
through a protocol known as SSH (Secure Shell). The
raspberry pi and the computer that request access to the
raspberry are required to be connected on the same
Local Area Network (LAN). Also, the username and
password of the raspberry pi to be accessed is needed
when trying to login to the raspberry through SSH, using
Putty.
v. VNC Viewer: This is a Windows-based application that
provides a server for viewing the X windows of the
raspberry pi
vi. Google Assistant SDK: The google assistant SDK is a
tool kit which contains functions and parameters which
were used to build the speech to text and text to speech
functionality. After the Speaker has been identified by
the module, what the module uses to understand exactly
what is being said is the Google assistant SDK. It is a
cloud based tool and therefore requires internet
connectivity to be actively ON in order for it to function.
vii. The Piwho Library: This is an open source Python
library which has been employed in this project to
achieve Voice biometrics.
viii. GPIO library: This is an open source Python based
library that is used to address the GPIO (General
Purpose Input Output) pins of the raspberry pi zero
3.3 Design Steps
The process of achieving the voice biometrics for access
control using the raspberry pi has been broken down into
steps as follows:
i. The Raspbian OS image is downloaded
ii. The SD Card formatter ids used to format the microSD
Card in order to prepare it for the mounting of the
Raspbian OS image.
iii. The Etcher Software is used to burn the image file to
the microSD Card
iv. After the image has been burnt into the microSD Card,
the it is then mounted into the Raspberry Pi Zero and
the Pi Zero is powered up.
v. All the peripherals are then attached to the USB hub
vi. After the raspberry pi zero is connected to a wireless
network, the firmware is updated and upgraded.
vii. The Piwho library is downloaded from gitHub
viii. All the source files are built using the Cmake tool
ix. Thereafter, the training phase is launched. Here a user
records are many short audio samples of his/her voice
and saves it to a folder.
x. The folder is then used to train the piwho software and
various parameters are assigned to the .xml file which
is generated. This .xml file contains all the features of
the voice which has just been trained
xi. The enrolment phase is complete
xii. An account is created with google in order to gain
access to the google assistant SDK
xiii. The google assistant SDK is then downloaded and a
python script is run to install it on the raspberry pi zero
xiv. Finally, a python script is written to integrate the whole
system and trigger the servo when a registered user
issues a command it recognizes.
Fig.3 is the flow chart of the python script written for this
purpose

Fig. 3 - Flowchart for Voice biometrics for Access Control
System
4. RESULTS
A prototype of the proposed system was developed.
Experiments were performed to test the proposed system
and to measure the accuracy of the system. Different audio
samples (in .wav format) were taken and used to train the
speaker recognizer. Thereafter, the voice of the same
person which was enrolled was used in different
conditions to test the accuracy of the system.
The various conditions by which the speaker recognition
system was tested include:
i. Recognition in a crowded place with a lot of
background noise
ii. Recognition in a silent place with little or no
background noise
iii. Recognition when the speaker’s voice is not loud
iv. Recognition when the speaker’s voice is loud
v. Recognition when the speaker’s voice has slight
variations due to cold
The results obtained are as shown in fig. 4.
A = Recognition in a crowded place with a lot of
background noise
B = Recognition in a silent place with little or no
background noise
C = Recognition when the speaker’s voice is not loud
D = Recognition when the speaker’s voice is loud
E = Recognition when the speaker’s voice has slight
Same Speaker under Different ConditionsFig. 4 -
Result of Tests Carried out on Same Speaker under

Figure 4 represents the results of how well the module
could perform on 150 test samples of the speaker’s voice,
after the speaker’s voice has been pre-enrolled into the
database of the module. Table1 shows these statistics:
Table.1: Success Rate of Speaker Recognition System
S/No. Test Condition
Number of
Accuracy
Percentage
(%)
1.
Recognition in a
crowded place
with a lot of
background
noise
90/60 60.00
2.
Recognition in a
silent place with
little or no
background
noise
120/30 80.00
3.
Recognition
when the
speaker’s voice
is not loud
130/20 86.67
4.
Recognition
when the
speaker’s voice
is loud
100/50 66.67
5.
Recognition
when the
speaker’s voice
has slight
variations due to
cold
80/70 53.33
The success rate in percentage is shown in fig.5:
5. CONCLUSION
Speaker recognition for access control has been
successfully implemented on the Raspberry Pi Single
Board Computer. This system can be further industrialized
and made to work in companies as their means of
entrance to only authorized places in the firm.
Furthermore, it can be used to securely log the data of the
persons that have had an access to a particular place. From
the tests and the results carried out, it can be deduced that
the speaker recognition module performs best when the
speakers voice is loud and when there is silence in the
place where the module is performing computation. The
response time of this module is relatively fast.
Fig.5: Success Rate of Speaker Recognition Module (%)

Figure 3.6: Flowchart for Voice biometrics for
Access Control System
CHAPTER 4 IMPLEMENTATION AND TESTING

4.1 Tests
Experiments have been performed to test the proposed
system and to measure the accuracy of the system.
Different audio samples (in .wav format) were taken and
used to train the speaker recognizer. Thereafter, the voice
of the same person which was enrolled was used in
different conditions to test the accuracy of the system.
The various conditions by which the speaker recognition
system was tested include:
vi. Recognition in a crowded place with a lot of
background noise
vii. Recognition in a silent place with little or no
background noise
viii. Recognition when the speaker’s voice is not loud
ix. Recognition when the speaker’s voice is loud
x. Recognition when the speaker’s voice has slight
4.2 Results
The results obtained are as follows:
A = Recognition in a crowded place with a lot of
background noise
B = Recognition in a silent place with little or no
background noise
C = Recognition when the speaker’s voice is not loud
D = Recognition when the speaker’s voice is loud
E = Recognition when the speaker’s voice has slight
The
figure 4.1 represents the results of how well the module
could perform on 150 test samples of the speaker’s voice,
after the speaker’s voice has been pre-enrolled into the
database of the module.
Table 4.1 Shows these statistics:
Table 4.1: Success Rate of Speaker Recognition System
S/No. Test Condition
Number of
Accuracy
Percentage
(%)
1.
Recognition in a
crowded place
with a lot of
background
noise
90/60 60.00
2.
Recognition in a
silent place with
little or no
background
noise
120/30 80.00
3.
Recognition
when the
speaker’s voice
is not loud
130/20 86.67
4.
Recognition
when the
speaker’s voice
is loud
100/50 66.67
5.
Recognition
when the
speaker’s voice
has slight
variations due to
cold
80/70 53.33
Representing the success rate in percentage, we have the
figure 4.2:
Figure 4.1: Result of Tests Carried out on
Figure 4.2: Success Rate of Speaker Recognition
Module (%)

REFERENCES
D. Kornack and P. Rakic, “Cell Proliferation without
Neurogenesis in Adult Primate Neocortex,” Science,
vol. 294, Dec. 2001, pp. 2127-2130,
doi:10.1126/science.1065467.
M. Young, The Technical Writer’s Handbook. Mill Valley,
CA: University Science, 1989.
R. Nicole, “Title of paper with only first word capitalized,”
J. Name Stand. Abbrev., in press.
K. Elissa, “Title of paper if known,” unpublished.

IRJET- A Raspberry Pi based Speaker Recognition System for Access Control

More Related Content

What's hot

Similar to IRJET- A Raspberry Pi based Speaker Recognition System for Access Control

More from IRJET Journal

Recently uploaded

IRJET- A Raspberry Pi based Speaker Recognition System for Access Control