AVoiceControlledE-CommerceWebApplication.pdf

See discussions, stats, and author profiles for this publication at: https://www.researchgate.net/publication/330474122
A Voice Controlled E-Commerce Web Application
Conference Paper · November 2018
DOI: 10.1109/IEMCON.2018.8614771
CITATIONS
18
READS
1,132
3 authors:
Mandeep Kandhari
Queen's University
2 PUBLICATIONS 18 CITATIONS
SEE PROFILE
Farhana H. Zulkernine
Queen's University
123 PUBLICATIONS 2,230 CITATIONS
SEE PROFILE
Haruna Isah
University of New Brunswick
28 PUBLICATIONS 963 CITATIONS
SEE PROFILE
All content following this page was uploaded by Haruna Isah on 17 June 2021.
The user has requested enhancement of the downloaded file.

A Voice Controlled E-Commerce Web Application
Mandeep Singh Kandhari
School of Computing
Queen’s University
Kingston, Canada
mandeep.kandhari@queensu.ca
Farhana Zulkernine
School of Computing
Queen’s University,
Kingston, Canada
farhana@cs.queensu.ca
Haruna Isah
School of Computing
Queen’s University,
Kingston, Canada
isah@cs.queensu.ca
Abstract— Automatic voice-controlled systems have changed
the way humans interact with a computer. Voice or speech
recognition systems allow a user to make a hands-free request to
the computer, which in turn processes the request and serves the
user with appropriate responses. After years of research and
developments in machine learning and artificial intelligence, today
voice-controlled technologies have become more efficient and are
widely applied in many domains to enable and improve human-to-
human and human-to-computer interactions. The state-of-the-art
e-commerce applications with the help of web technologies offer
interactive and user-friendly interfaces. However, there are some
instances where people, especially with visual disabilities, are not
able to fully experience the serviceability of such applications. A
voice-controlled system embedded in a web application can
enhance user experience and can provide voice as a means to
control the functionality of e-commerce websites. In this paper, we
propose a taxonomy of speech recognition systems (SRS) and
present a voice-controlled commodity purchase e-commerce
application using IBM Watson speech-to-text to demonstrate its
usability. The prototype can be extended to other application
scenarios such as government service kiosks and enable analytics
of the converted text data for scenarios such as medical diagnosis
at the clinics.
Keywords— Speech Recognition Systems, Voice Recognition
Systems, Speech-to-Text, Text-to-Speech, Word Error Rate,
Recognition Rate, IBM Watson, Google API, Amazon Alexa
I. INTRODUCTION
Voice recognition is used interchangeably with speech
recognition, however, voice recognition is primarily the task of
determining the identity of a speaker rather than the content of
the speaker’s speech [1]. Speech recognition is a process of
converting the sound of words or phrases spoken by humans into
electrical signals to which a meaning is assigned [2] by
comparing the signals with sets of phonemic representations for
a close match. The phonemic representations are matched
against words that are predefined in a word vocabulary. The goal
of speech recognition is to enable people to communicate more
naturally and effectively. This often requires deep integration
with many natural language processing (NLP) components.
Speech recognition can be applied to many domains and
applications. It can remove barriers to human-human
interactions by aiding people who speak different languages to
be able to talk to each other without a human interpreter. It can
be used in a messaging system to transcribe voice messages left
by a caller into text that can be easily sent to the recipient
through emails or instant messaging [3]. Speech recognition
technology has made it possible to develop computer-based
reading coaches that listen to students, assess the performances,
and provide immediate customized feedbacks [1].
The traditional methods of data entry (keyboard and mouse)
fail the accessibility requirements to support all types of users.
Therefore, it is necessary to develop systems and applications
with enhanced usability for all users. Our research focuses on a
cloud-based Speech Recognition System (SRS) for e-commerce
applications as a use case scenario. It is crucial for an
organization or company to design and develop a web
application that is informative, interactive and easily accessible
to the users. Companies can use SRS to attract web audiences,
advertise their products and services, educate customers about
the features and usability of the products, and provide assistance
in troubleshooting as are given through online chat programs
today. Most of the existing applications lack the accessibility
characteristics [4, 5], which poses a barrier especially to the
visually-impaired users in terms of efficient access, and use of
the information and web-services provided by the organizations.
The web-accessibility standards established by the World Wide
Web Consortium (W3C) Web Content Accessibility Guidelines
(WCAG 2.1) [6] are not sufficient to address all accessibility
problems [7]. SRS can enable people with physical and learning
disabilities to interact with web applications at their own pace.
Over the years, various web assistive technologies like
screen readers, special browsers, and screen magnification
techniques have been developed to address the problems faced
by visually impaired web users. These systems have helped the
users either by reading, enabling voice commands, or providing
ways for screen magnification to comprehend the contents on a
web page. However, most of these solutions have failed in terms
of accuracy as the frequency of misinterpretation is often high.
Web applications equipped with voice-enabled systems can
not only provide flexibility in terms of users’ choice of web
interaction but can also increase the usability of the applications
for the general users when they are unable to use the traditional
human-computer interaction mechanisms. By allowing users to
control the functionality of the applications with their voice,
SRS can enhance users’ browsing experience, and allow users
to effectively convey their instructions and requests using
natural languages.
In this paper, we present a study of the state-of-the-art speech
recognition systems and propose a taxonomy of SRS. We also
present a voice-controlled e-commerce application using IBM
Watson speech-to-text service as a part of a comparative study
with other speech-to-text systems such as Google and Amazon.

IBM Watson speech-to-text service uses advanced NLP and
machine learning technologies for voice synthesization and text
conversion. Our web-application takes a voice command,
converts it to text, extracts meaning from the text, and then
performs a wide variety of tasks including searching, browsing,
reading and writing text.
The rest of the paper is organized as follows. Section II
presents a background of the concepts, components, and
applications of SRSs. The taxonomy of speech recognition
systems is presented in Section III. Section IV presents a review
of the related work including a comparison of some of the
cutting edge SRSs. Section V describes an overview of the
proposed framework while the implementation is described in
section VI. Section VII concludes the paper with a discussion of
the future work.
II. BACKGROUND
The concept of interacting with a computer using voice has
led to a new research area of SRS over the years which has
significantly improved the human-to-computer interactions. It
allows people to control their surroundings with their voice.
Speech recognition technology can be used to dictate short
messages, and thereby, reduce the effort needed to send a short
text or email message, which otherwise needs to be typed. It can
also be used to recognize and index speeches and lectures so that
users can easily find the information that is interesting to them
[1]. SRSs can be used to address the communication and
interaction problems faced by people with disabilities.
Modern SRSs are built on statistical principles. The
architecture of a typical SRSs, as shown in Fig. 1, consists of a
voice input source or speech signal, feature extraction module,
search engine, language model, acoustic model, adaptation
model, and output or transcription components. The input data
is the speaker’s voice or speech, which is transformed into a
speech waveform or signal data, and passed on to the feature
extraction module for noise removal and transformation to the
required representation format. The extracted signal is then
moved to a search engine that is connected to language and
acoustic models. The acoustic model includes knowledge about
acoustics, phonetics, microphone and environment variability,
and gender and dialect differences among speakers. The
language model includes semantic knowledge or primarily
meanings of words. It also describes what constitutes a possible
word, what words are likely to co-occur, and in what sequence.
It also houses the semantics and functions related to an operation
a user may wish to perform. A modern SRS is able to handle a
lot of uncertainties associated with speaker characteristics, rate,
and style of speech; recognize basic speech segments, possible
words, likely words, unknown words, and grammatical
variations; process noise interference and non-native accents,
and compute confidence scores of the results. This is the main
function of the search engine component. The adaptation unit is
used to modify the outputs from the acoustic or language models
coming via the search engine to improve the overall
performance [2].
There are many existing SRSs, and yet more are being
developed. The next section presents the taxonomy for
categorizing SRSs.
Fig. 1. The architecture of a speech recognition system.
III. TAXONOMY OF SRS
Recent studies have shown that there are three major
approaches to SRSs. These include acoustic phonetic, pattern
recognition, and artificial intelligence approaches. In terms of
design, SRSs can be divided into various categories on the basis
of the type of input speech, vocabulary, style of speech, and
speaker model [8]. Fig. 2 presents the taxonomy of SRSs. The
criteria for the categorization of the SRS are described below.
A. Type of Speech
1) Isolated Words: These models are only capable of
detecting a single utterance at a time i.e. the words are isolated
from the quiet phases at both the ends of the sample window.
The system accepts a single word or an utterance and while the
user is in the wait state, the system processes it and generates
the appropriate result of the input speech. This approach
computes changes in the energy of the input signal to leverage
speech detection.
2) Connected Words: These systems allow users to give a
set of words or utterances as inputs. The system identifies
words with minimum pauses as units, that are trained in an
isolated word mode. When the user enters the wait state, the
system processes the input and generates the output
accordingly.
3) Continuous Speech: In these systems, the user speaks
continuously and it is the responsibility of the system to
determine the word units and its segmentation. The
implementation of such a system is quite difficult as special
methods are required to recognize the text and to determine the
utterance boundaries simultaneously.
4) Spontaneous Speech: These systems use advanced
machine learning and NLP techniques to efficiently recognize
spontaneous voice inputs, process and understand the meaning
of the text, and generate appropriate responses. These types of
systems are basically used in the query-based applications
where the system returns answers to user queries.
B. Vocabulary
The performance of an SRS depends on the size of the
vocabulary it supports. The vocabulary size affects the accuracy,
complexity, and the processing rate of a system. A system with
a larger vocabulary size will have much better performance as
compared to its peers since it allows the model to select from a
wide range of words but will also require greater resources to
Search
Engine
Feature
Extraction
Adaptation
Speech
Signal
Language
Model
Acoustic
Model
Transcription

maintain efficiency in terms of response time. SRSs can be
further categorized based on the following vocabulary sizes that
they support:
• Small - 1 to 100 words or sentences
• Medium - 101 to 1000 words or sentences
• Large - 1001 to 10,000 words or sentences
• Very-Large – More than 10,000 words or sentences
C. Style of Speech
These SRSs apply the knowledge of accent and style of
speech for word recognition. Although it requires knowledge
about speech styles, style independent recognition is generally
more challenging since, for matching phonemes, the internal
representation of the speech must somehow be global.
D. Speaker Model
Speaker independent recognition is more difficult because
the internal representation of the speech must somehow be
global enough to cover all types of voices and all possible ways
of pronouncing words, and yet specific enough to discriminate
between the various words of the vocabulary. For a speaker
dependent system, the training is usually carried out by the user,
but for applications such as large vocabulary dictation systems,
this is too time-consuming for an individual user. In such cases,
an intermediate technique known as speaker adaptation is used.
Here, the system is bootstrapped with speaker-independent
models which gradually adapts to the specific aspects of the
user.
Fig. 2. Taxonomy of SRSs.
IV. RELATED WORK
Many studies have been conducted where the researchers
have integrated the SRS functionality with applications to make
them usable by visually-impaired users. In an early study, Kevin
Christian et al. [9] compared the performances of voice
controlled and mouse controlled web browsing. Based on the
subjective ratings of the 18 test subjects who used both
controlling methods on hypertext forms, textual and numbering
links concluded that the textual links are preferable to numbered
links. Bajpei et al. [10] proposed a system where a SRS was
incorporated in a browser. The system would accept users’ voice
as input and based on a predefined set of operations the browser
would perform an operation.
a.
http://blog-archive.griddynamics.com/2016/01/automatic-speech-recognition-services.html
The idea of the voice-operated browser was extended by Han
et al. [11] such that it could be used in a smart TV environment.
The research focused on navigating and controlling a
dynamically generated hierarchical menu on a webpage with
voice keywords. Sagar et al. [12] proposed an application where
email services were combined with the text-to-speech and
speech-to-text services thus enabling users to write, send, and
read their e-mails. The main idea was to implement an add-on
feature for the e-mail services.
A. Comparison of Cutting Edge SRSs
In this section, we present a summary of comparative studies
of different SRS from the literature. A number of offline and
cloud-based SRSs are available in the market which differ in
terms of the tradeoff between privacy and accuracy. The offline
systems provide a high degree of privacy, which is preferred
when working in a specific domain where security of the data is
a big concern. However, they have lower accuracy compared to
the less private systems due to the lack of access to knowledge
resources imposed partially by the constraints on sharing
information. Based on accessibility, SRS can be basically of two
types, cloud-based and open source.
TABLE I. EXAMPLES OF VOICE RECOGNITION SYSTEMS (VRS)
Cloud-based Proprietary SRS Open source SRS
IBM Watson Kaldi
Google Speech API CMU Sphinx
Microsoft Speech API HTK
For comparing the quality of the SRS, a comparison1
was
conducted for cloud-based SRS, where a set of 3,000 different
voice search phrases for e-Commerce were used. A recorded
audio file with search phrases was given as an input to a number
of SRSs and then the output text generated by these services
were analyzed and compared with the actual search phrases. The
quality of the results was determined by the phrase recognition
rate (PRR) defined as the ratio of the number of recognized
phrases (correct and incorrect) to all phrases as shown in Eq. 1,
and the word error rate (WER) defined as the ratio of the number
of correctly identified words to all phrases as shown in Eq. 2.
Results of the experiments are given in Table II. In terms of
quality, Google proved to be superior to the other systems as it
was able to identify 73.3% of the text with only 15.8% WER and
73.3% PRR. Nuance’s Dragon system achieved around 39.7%
WER with a PRR of 44.1%. IBM Watson recognized 46.3% of
the text with a WER of 42.3% and 46.3% PRR. The WERs for
AT&T and WIT were similar, about 63.3%. However, the PRR

for AT&T was better than that of WIT (32.8% as compared to
29.5%).
TABLE II. CLOUD-BASED SRS WER AND PRR
Cloud-based SRS WER (%) PRR (%)
Google 15.8 73.3
Nuance 39.7 44.1
IBM 42.3 46.3
AT&T 63.3 32.8
WIT 63.3 29.5
Këpuska and Bohouta [13] built a tool to compare the
performances of the cloud-based Google and Microsoft Speech
Recognition (SR) APIs with the open source CMU Sphinx
system. A recorded file was given as an input to the systems and
the system generated output was used for calculating the PRR
and WER. The results concluded that the Google Speech API
had a better WER performance of 9% followed by the Microsoft
Speech API (18%) and Sphinx-4 (37%).
The performances of open-source speech recognition
toolkits are evaluated by Gaida et al. [14] on two datasets namely
the vermobil1 corpus and the wall street journal1 corpus. They
compared the Hidden Markov Model Toolkit (HTK) with
HDecode and Julius, CMU Sphinx, and the Kaldi toolkit. The
authors reported that the Kaldi toolkit provided the most
promising results followed by the Sphinx and HTK. The results
generated by the Kaldi toolkit were better since a significant
amount of knowledge and effort had been spent on training the
model.
TABLE III. PERFORMANCE OF SENTENCE RECOGNITION [15]
SRS Kaldi Watson
Sentences with
(total 141)
No. of
Sentences
% error No. of
Sentences
% error
Errors 135 95.7% 28 95.7%
Substitutions 39 27.7% 23 67.3%
Deletions 15 10.6% 21 10.6%
Insertions 90 63.8% 10 27.7%
TABLE IV. PERFORMANCE OF WORD RECOGNITION [15]
Elma Gazetić [15] compared the performances of Kaldi and
IBM Watson, using an audiobook ‘Emma from Jane Austin’ as
an input. The performances were measured in terms of WA and
WER. It was found that the cloud-based applications had far
superior performances over the open source tools such as Kaldi
because the former systems are trained for thousands of hours as
compared to the latter which was trained only for a few hundred
hours.
V. OVERVIEW OF THE FRAMEWORK
Majed Alshamari [16] evaluated the accessibility of the three
well known Saudi e-commerce websites with five testing tools:
Achecker, TAW, Eval Access, Mauve, and FAE. The study
concluded that the websites were not competent enough to
address the basic accessibility problems such as navigation,
readability, timing, and input assistance.
According to the McNair’s report [17], the worldwide retail
e-commerce business in the year 2018 was expected to reach
$2.774 trillion dollars which accounts for 11.6% of the total
retail sales. The sales numbers are expected to touch $4 trillion
dollars’ mark by the end of the year 2021, which constitute
15.5% of the total retail sales. The numbers are evident enough
to depict that e-commerce has emerged as an alternative to the
traditional retail business, giving the users a convenient option
of ordering the products online. However, the accessibility
issues [14] in the websites discourage a large number of visually
impaired users who remain deprived of the facilities provided by
the websites and the e-commerce business loses some of its
potential customers. This motivated us to design a SRS based e-
commerce system to address the accessibility problems faced by
the disabled users. This section presents an overview of the
design of the framework. Implementation details of the
prototype are described in the next section.
A. Watson SRS for E-COMMERCE
Poor accessibility of websites causes negative impacts on
web users and causes hit scores for those websites to decrease
with time. For e-commerce businesses, this can be disastrous as
most of the transactions are done online. Researchers have
discussed assistive technologies like screen readers, voice
browsers and a variety of adaptive strategies that users apply to
surf the internet, but these are not sufficient and often fail to
address the accessibilities issues. Borodin et al. [18] describe the
usability problems faced by the screen-reader users. Even
though the technology allows access to web pages, it can often
become unbearable especially if a large number of links are
present on the page. The screen-reader would make
announcements every time it encounters a link, thus making it
impractical and annoying for the users. An e-commerce web
application is a perfect example as it often contains a large
number of links to different product pages. The incorporation of
speech services can provide an alternative mechanism to the
users where users can directly interact with the applications.
SRS will further enhance the accessibility of a webpage and will
empower the users with the ability to efficiently access,
navigate, skim through a large amount of information, get input
assistance, and even control the functionality of the webpage.
Based on the comparative study of the existing SRS in
Section 3, we can conclude that the performances of cloud-based
systems are better than the open source systems in terms of WER
and PRR. Higher accuracy ensures that the users do not have to
give the same voice commands repeatedly to the application.
The open source tools can also support high accuracy rates, but
a significant amount of time is needed to train and customize the
system for a particular application. The main motivation for
using Watson speech services is that access to the research
platform and the software are provided by our collaborators
from IBM. Also according to IBM tech report [19], the use of
SRS Kaldi Watson
Total Words: 2060 Number of
words
% error Number of
words
% error
% total error 205 10% 63 3.1%
% correct 1855 90% 1997 96.9%
% substitution 51 2.5% 22 1.1%
% deletions 48 2.3% 24 1.2%
% insertions 106 5.1% 17 0.8%
% Word accuracy 84.9% 96.1%

deep learning technologies has allowed Watson speech
recognition to achieve 5.5% WER, which gives a good
motivation to use Watson speech services in our research.
Building a highly accurate SR engine requires access to
knowledge about grammar and language structure along with
mechanisms to compose audio signals and detect keywords.
SRS should allow real-time transcription of the voice which is
updated continuously with the incoming voice data. The custom
model support can provide the freedom and flexibility to train
the service according to the nature of the application. All these
features are supported by Watson and make it a great tool to use
in our web application.
The architecture of our prototype e-commerce web
application using Watson SRS is depicted in Fig. 3. Users, with
the help of the microphones in their computer systems, will give
voice commands to the application. The input signal is then
streamed through the Watson Speech-To-Text (STT) service
which converts the audio signal to the appropriate text. One of
the key features of the Watson STT service is the capability of
efficiently identifying keywords in the text. Since every web
application has to perform a predefined set of operations, the
application can look for these keywords. When a matching
keyword is found, the application can trigger a predefined
functionality. In the case an application cannot find any
matching keyword, it can use Watson Text-To-Speech (TTS)
service to generate a voice response to inform the user about it.
Fig. 3. The architecture of the Web Application.
The performance of a SRS depends on the size of the
vocabulary it supports for a certain level of accuracy,
complexity, and information processing rate. The vocabulary
size of the cloud-based Watson system is very large and thus the
STT and TTS generate good results. We implemented
continuous speech type so that the user can make descriptive
requests to the application without stopping. The other aspect of
the SRS is that it is speaker independent i.e., the application can
support voices and the pronunciation of words on a global level
thus making it useful for a greater audience. Lastly, the SRS uses
advance deep neural network techniques to generate accurate
and precise voice transcriptions.
In our SRS application for e-commerce, we demonstrate a
voice-controlled web-based commodity purchase scenario,
which incorporates all the above aspects. The system accepts a
user-query as a voice command, looks for particular keywords
in it, and based on the closest match found, executes a job like
b.
https://stateofjs.com/
searching for a particular item and traversing through different
web pages. The supplementary information provided by the
Watson services along with the transcribed words are used to
build an intelligent query and matching system.
VI. IMPLEMENTATION DETAILS
The architecture of our proposed e-commerce web
application integrated with the Watson SRS is depicted in Fig.
3. In this section, we describe the implementation details of our
prototype application.
A. Software Components
We used Watson STT and TTS to build our SRS prototype
e-commerce application. The application combines a) client-
side scripts (JavaScript and HTML) used to present information
to the users, b) a back-end application powered by server-side
scripts (JavaScript and Python), and c) a database system, where
the latter two allows storing and retrieving information and
generating responses to send back to the client. For our
application, the components were picked on the basis of the
survey2
where 20,000 developers were asked for their opinion
about the tools and frameworks currently used by them. The
results showed that the React JavaScript library and the Express
are currently the most preferred tools. For the database, we chose
MongoDB, which is a non-relational database that can store
information in the BSON (Binary JavaScript Object Notation)
format. It provides an efficient way of building a highly scalable
application supporting diverse data types. Data about the
merchandise which is used to build our e-commerce website was
obtained from the edX “front-end capstone project” 3
course.
The course teaches how to develop an e-commerce website and
uses a dataset containing a list of products on their cloud in
JSON format. This data is retrieved and fed into our application
for rendering the products on the screen.
The different components and the information flow of our
SRS e-commerce web application are shown in Fig. 4.
Fig. 4. Information flow in our SRS web application.
c.
https://courses.edx.org/courses/course-v1:Microsoft+DEV238x+2T2018/course/

1) Information Flow: When the application loads for the
first time, the client script makes a request to the server for the
IBM Watson STT and TTS services. The backend, after
verifying the application’s credentials, sends a success response
back to the frontend enabling it to use the services as shown in
Fig. 5.
Fig. 5. Server verifying the credentials of the application.
As the application calls the STT service with the voice input
data, it sends a JSON response back to the front end. At the same
time, the client-side script renders the sign-in page on the screen
asking for the existing customer information to prevent
unauthorized access. The new users can use the register page to
register to the applications database. After the initial rendering
is done and the components are mounted on the screen (sign-in
form), users can use the STT service to enter their information
to the sign-in or register form. The frontend will then send this
information to the backend which in turn will be sent to the
database for verification or registration purposes. After
advancing through the initial check, the frontend will make a
fetch request based on users’ requests as shown in Fig. 6, which
on success will return a promise that resolves to the response of
that request as shown in Fig. 7 indicating keywords. These types
of systems are basically used in the query-based applications
where the systems return answers to user queries.
Fig. 6. Fetch request.
In our use case scenario based on the keywords, the backend
application returns a response containing a JSON dataset as
shown in Fig. 8. The response contains all the product or
merchandise related information like image links, names, costs,
and descriptions of the products. It is then parsed and processed
by the frontend application (see data flow in Fig. 4) to render the
products on the screen. Users can then use their voice to activate
the TTS service and give commands to search, add, or remove a
particular item from the cart. In the case the searched item is not
available, the TTS service is used to generate a voice response
to inform the customers about it. So, by incorporating both
Watson STT and TTS we can develop an intelligent application
that can actually interact with the users and help users with
accessibility problems.
Fig. 7. Screenshot of Response from Watson STT service.
Fig. 8. The response about the products in JSON format.
Fig. 9. Products page as shown on the website.
B. Ongoing Work
The basic functionality of the system has been developed
where users can use their voice commands with the application
instead of typing to give inputs and perform a search to see
available goods as shown in Fig. 9. The next step would be to
incorporate the TTS service in the application so that in the case
the application is not able to service users’ commands, it can
inform the same to the users using a generated voice, complete
purchase for the user, and confirm or cancel transactions. This
application will address the accessibility problems of the e-

commerce websites to some extent. In the future, we will focus
on incorporating a virtual assistant to provide recommendations.
In this way, we will be able to implement a system that will
enhance users’ comfort and experience.
VII. CONCLUSIONS
We present a study of some of the cutting edge SRSs in this
research, a summary of performance comparisons of some of the
popular SRSs from the literature, a taxonomy for categorizing
SRSs based on their functionality, and a preliminary prototype
of a voice-controlled e-commerce web application using IBM
Watson STT and TTS services. Enhancing accessibility to
commercial websites is crucial for today’s e-commerce
dependent economy. We implement a SRS-based e-commerce
web application with a view to leverage accessibility to web
applications for the visually-impaired users such that they can
use their voice as a means to operate the application. SRS
enabled applications can enhance usability for all users by
promoting ease of interaction and multi-tasking, and support a
lean environment where users can make requests using natural
language.
For the future work, we plan to compare the performances
of other cloud-based services such as Google and Amazon in the
context of the same e-commerce application. The application
can be extended to identify a specific speaker based on the voice.
Furthermore, the application can also be equipped with the
language translation APIs that will allow users to give
commands to the application in their own native language. This
can prove to be beneficial for the users who are restrained from
using the websites because of the language barrier. The
application can be designed in such a way, that it can render
user-specific interfaces based on their interests. This means that
the interface can be rendered distinctively for every user thus
making the application more appealing and user-friendly. The
application can also be extended to other domains such as for
online education, providing services at government service
kiosks, and for emergency online assistance.
ACKNOWLEDGMENT
We like to acknowledge the support from IBM and the
Centre for Advanced Computing at Queen’s University to
provide us access to the Watson services, and the online course
website for the product data used in this research.
REFERENCES
[1] National Research Council. (2002). Technology and Assessment:
Thinking Ahead--Proceedings from a Workshop. National Academies
Press.
[2] N. Indurkhya, F. J. Damerau, Handbook of natural language processing
Vol. 2, CRC Press, 2010.
[3] D. Li and Y. Dong, “Deep learning: methods and applications,”
Foundations and Trends{textregistered} in Signal Processing, vol. 7, pp.
197-387, 2014.
[4] B. Hashemian, "Analyzing web accessibility in Finnish higher
education", ACM SIGACCESS Accessibility and Computing, no. 101, pp.
8-16, 2011.
[5] K. Nahon, I. Benbasat and C. Grange, "The Missing Link: Intention to
Produce Online Content Accessible to People with Disabilities by Non-
professionals," 2012 45th Hawaii International Conference on System
Sciences, Maui, HI, 2012, pp. 1747-1757.
[6] W3C, “Introduction to web accessibility”, available from:
https://www.w3.org/TR/2018/REC-WCAG21-20180605/
[7] M. Akram and R. Bt Sulaiman, "A Systematic Literature Review to
Determine the Web Accessibility Issues in Saudi Arabian University and
Government Websites for Disable People", International Journal of
Advanced Computer Science and Applications, vol. 8, no. 6, 2017.
[8] M. A. Anusuya and S. K. Katti, "Speech Recognition by Machine,"
International Journal of Computer Science and Information Security,
IJCSIS, vol. 6, no. 3, pp. 181-205, 2010.
[9] K. Christian, B. Kules, B. Shneiderman and A. Youssef, “A Comparison
of Voice Controlled and Mouse Controlled Web Browsing,” Proceedings
of the Fourth International ACM Conference on Assistive Technologies,
no. ACM, pp. 72-79, 2000.
[10] A. B. BAJPEI, M. S. SHAIKH and N. S. RATATE, “VOICE
OPERATED WEB BROWSER,” International Journal of Soft
Computing and Artificial Intelligence, vol. 3, no. 1, pp. 30-32, May-2015.
[11] S. Han, G. Jung, B.U. C. Minsoo Ryu and J. Cha, “A Voice-controlled
Web Browser to Navigate Hierarchical Hidden Menus of Web Pages in a
Smart-tv Environment,” Proceedings of the 23rd International
Conference on World Wide Web, pp. 587-590, 2014.
[12] S. Sagar, V. Awasthi, S. Rastogi, T. Garg, S. Kuzhalvaimozhi, “Web
application for voice operated e-mail exchange”,
[13] V. K{"e}puska and B. Gamal, “Comparing speech recognition systems
(Microsoft API, Google API and CMU Sphinx,” Journal of Engineering
Research and Application, vol. 7, no. 3, pp. 20-24, 2017.
[14] C. Gaida, P. Lange, R. Petrick, P. Proba, A. Malatawy and D.
Suendermann-Oeft, “Comparing open-source speech recognition
toolkits,” Tech. Rep., DHBW Stuttgart, 2014.
[15] E. Gazetić, “Comparison Between Cloud-based and Offline Speech
Recognition Systems”. Master’s thesis. Technical University of Munich,
Munich, Germany, 2017.
[16] M. Alshamari, “Accessibility evaluation of Arabic e-commerce web sites
using automated tools,” Journal of Software Engineering and
Applications, vol. 9, no. 09, p. 439, 2016.
[17] C. McNair, “Worldwide retail and ecommerce sales: emarketer’s updated
forecast and new mcommerce estimates for 2016–2021,” Industry Report,
eMarketing , 2018.
[18] Y. Borodin, J. P. Bigham, G. Dausch and I. Ramakrishnan, “More than
meets the eye: a survey of screen-reader browsing strategies,”
Proceedings of the 2010 International Cross Disciplinary Conference on
Web Accessibility (W4A), p. 13, 2010.
[19] Alison DeNisco Rayome, 2017, “Why IBM's speech recognition
breakthrough matters for AI and IoT”, white paper, online at:
https://www.techrepublic.com/article/why-ibms-speech-recognition-
breakthrough-matters-for-ai-and-iot/
View publication stats

AVoiceControlledE-CommerceWebApplication.pdf

More Related Content

Similar to AVoiceControlledE-CommerceWebApplication.pdf (20)

Recently uploaded (20)

AVoiceControlledE-CommerceWebApplication.pdf