Ney Yibeogo - Hello World: A Voice Service Development Platform

to Bridge the Web’s Digital Divide

André Baart

, Anna Bon

, Victor de Boer

, Wendelien Tuijp

and Hans Akkermans

Amsterdam Business School, Universiteit van Amsterdam, Plantage Muidergracht 12,

1018 TV, Amsterdam, The Netherlands

Network Institute, Vrije Universiteit Amsterdam, De Boelelaan 1081, 1081 HV, Amsterdam, The Netherlands

Computer Science Department, Vrije Universiteit Amsterdam, De Boelelaan 1081,

1081 HV, Amsterdam, The Netherlands

Centre for International Cooperation, Vrije Universiteit Amsterdam, De Boelelaan 1081,

1081 HV, Amsterdam, The Netherlands

Network Institute, Vrije Universiteit Amsterdam, De Boelelaan 1081, 1081 HV, Amsterdam, The Netherlands

Keywords:

Digital Divide, Low Literacy, Sub-Saharan Africa, Voice-based Services, Low-resource Hardware, Services

Development Software Kit.

Abstract:

The World Wide Web is a crucial open public space for knowledge sharing, content creation and application

service provisioning for billions on this planet. Although it has a global reach, still more than three billion

people do not have access to the Web, the majority of whom live in the Global South, often in rural regions,

under low-resource conditions and with poor infrastructure. However, the need for knowledge sharing, content

creation and application service provisioning is no less on the other side of this Digital Divide. In this paper

we describe the Kasadaka platform that supports easy creation of local-content and voice-based information

services, targeted at currently ‘unconnected’ populations and matching the associated resource and infrastruc-

tural requirements. The Kasadaka platform and especially its Voice Service Development Kit supports the

formation of an ecosystem of decentralized voice-based information services that serve local populations and

communities. This is, in fact, very much analogous to the services and functionalities offered by the Web, but

in regions where Internet and Web are absent and will continue to be for the foreseeable future.

1 INTRODUCTION

The World Wide Web is a unique public space for

knowledge sharing, content creation and application

service provisioning for billions on this planet. Al-

though it has a global reach, still more than three bil-

lion people do not have access to the Web: the ‘Digital

Divide’ (Fuchs and Horak, 2008). The majority lives

in the Global South, often in remote rural regions, un-

der low-resource conditions and with poor or even ab-

sent infrastructures.

However, needs for knowledge sharing, locally

relevant content and application service provisioning

are certainly no less beyond the current borders of the

Web.

To overcome the Digital Divide, various policies

are promoted to improve global access to Internet,

Web and its vast arsenal of resources. A prominent

one, for which quite large funds have been made

available by donors such as the World Bank, is the

attempt to roll out forms of “affordable Internet” to

currently unconnected regions.

Basically, the un-

derlying idea is a form of relatively straightforward

technology transfer from advanced countries to devel-

oping and emerging regions (USAID, 2017; Schmida

et al., 2017; The World Bank Group, 2016).

Our research focuses on information exchange

and knowledge sharing support for small holder and

family farmers in the African Sahel (including e.g.

Mali, Burkina Faso, northern Ghana). In a country

such as Mali, around 80% of the population depend

for their livelihood on work in small subsistence agri-

culture in remote rural regions where there is no In-

ternet, very limited electricity, and high levels of low-

literacy (around 50% on average, for women even sig-

niﬁcantly higher). Under these conditions it is highly

unlikely that a technology transfer policy of Inter-

See: https://webfoundation.org/our-work/projects/

alliance-for-affordable-internet/

Baart, A., Bon, A., Boer, V., Tuijp, W. and Akkermans, H.

Ney Yibeogo - Hello World: A Voice Service Development Platform to Bridge the Web’s Digital Divide.

DOI: 10.5220/0006893600230034

In Proceedings of the 14th International Conference on Web Information Systems and Technologies (WEBIST 2018), pages 23-34

ISBN: 978-989-758-324-7

Figure 1: ICT4D Field research methodology, from (Bon et al., 2016).

net roll-out to bridge the Digital Divide will come to

fruition in some foreseeable future.

This does not imply that nothing can be done. The

contribution of this paper is that one can, and that it is

possible to develop and deliver web-reminescent ser-

vices for information and knowledge exchange, but

not in a one-size-ﬁts-all technology transfer approach.

It requires a thorough investigation in the ﬁeld of

conditions, requirements and local speciﬁcities. This

leads to insights and technical directions that cannot

be derived from advanced but far-away technology

considerations alone.

Accordingly, in this paper we present the

Kasadaka platform intended to support easy cre-

ation of local-content and voice-based information

services, targeted at currently ‘unconnected’ popula-

tions and meeting the harsh conditions at the other

side of the Digital Divide.

The Kasadaka platform and especially its Voice

Service Development Kit aims to facilitate the for-

mation of an ecosystem of many decentralized voice-

based information services that serve local popula-

tions and communities. This is, in fact, very much

analogous to the services and functionalities offered

by the Web, but in regions where Internet and Web

are and will continue to be absent for the foreseeable

future.

2 APPROACH AND

METHODOLOGY

The design of a service development platform for

low-tech and low-resource environments as sketched

above, cannot be based on generic technological con-

siderations alone, but requires in-depth study of local

contexts and conditions, in addition to and beyond the

usual ones that hold for ICT development in general.

In our approach, we cater for this situatedness of

information systems and services by carrying out ex-

tensive ﬁeld research. The underlying — iterative

and collaborative — ﬁeld research methodology is de-

picted in Figure 1 in the form of an intention-strategy

map (Rolland, 2007), and is discussed in more detail

in (Bon et al., 2016).

Subsequently, this ﬁeld research is utilized to de-

rive requirements for the platform (discussed in Sec-

tion 3). Apart from a good understanding of the

‘Digital Divide’ context in which services are to be

deployed, platform requirements originate from the

spectrum of service use cases the platform is sup-

posed to support and execute. These use cases, de-

veloped with end users and stakeholders in the ﬁeld,

similarly are the result of ﬁeld research, often in the

form of collaborative workshops.

Thus, the methodology employed to come to the

KasaDaka service platform services for low-tech low-

resource environments is in brief:

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

• Field-based context analysis: i.e. ﬁnding out the

local conditions with respect to technical infras-

tructure, environmental conditions, availability of

technical support etc;

• Platform requirements analysis, as also derived

from the service use cases, elicited from local

users during ﬁeld research;

• Technical design and implementation of the plat-

form;

• Testing and evaluation, also with local users in the

local context.

Below, Section 3 presents an analysis of the plat-

form requirements, Section 4 describes the architec-

ture and technical implementation issues, and Sec-

tion 5 describes various evaluations of the platform.

Sec. 6 discusses related work and Section 7 summa-

rizes our main conclusions.

3 PLATFORM REQUIREMENTS

ANALYSIS

Requirements in a technology development project

are derived from the needs that the context of the

project brings to the table. The development of a sys-

tem that is intended for those on the other side of the

Digital Divide has to deal with several combinations

of circumstances and issues that are rarely encoun-

tered in technology development projects in the de-

veloped world.

3.1 Societal Challenges

Voice-based access to information is an essential

requirement for bridging the digital divide, and

reaching the world’s rural poor. In these populations,

literacy rates are low, which disqualiﬁes any service

that is text-based. In several sub-Saharan African

countries (such as Niger, Mali and Burkina Faso)

the literacy rates are below 40%, which makes the

vast amounts of textual information on the Internet

out of reach for a major part of the population in

these regions (UNESCO, 2011). Furthermore, many

indigenous cultures have a strong oral tradition in

communication, so that voice-based services have a

natural ﬁt with the locally already existing means of

communication.

Voice services for the world’s rural poor have

to support under-resourced languages, which implies

that they can not use advanced speech technologies.

While many developing countries have a technolog-

ically well-supported ofﬁcial language (often a rem-

nant of colonial times), this language is not necessar-

ily spoken by the entire population. Rather, the lo-

cal population speaks their own indigenous language

which is tied to their local region. Africa for instance

has around 2000 local languages, which each often

have local dialects (Heine and Nurse, 2000). The ma-

jority of these languages are spoken languages, mean-

ing that there exists little to no literature in these lan-

guages. Furthermore, due to the populations speaking

these languages being poor and relatively small, these

populations do not provide a proﬁtable market for the

development of Text To Speech, Automatic Speech

Recognition and Natural Language Processing tech-

nologies in these languages. Most of the recently

developed voice platforms that offer complex infor-

mation services (e.g. Apple’s Siri, Amazon Alexa,

etc.) rely on the use of these technologies. While

these technologies are in widespread use around the

world, they require signiﬁcant research and a lot of

work (and thus signiﬁcant ﬁnancial investment) in or-

der to support a language at a level that is sufﬁcient for

usage in voice services. (Bagshaw et al., 2011; Black

and Lenzo, 2000; Farrugia, 2005; McTear et al., 2016;

Besacier et al., 2014; de Vries et al., 2014) The num-

ber of languages that has well-developed speech tech-

nologies is rising, but these do (almost) not include

any of the indigenous languages found in the devel-

oping world. This situation is not likely to change, as

there is little (ﬁnancial) incentive to develop technolo-

gies for these languages. Taking into account these

restrictions, these languages are referred to as under-

resourced languages (Berment, 2004).

3.2 Resource and Infrastructure

Constraints

Information Services for the World’s poor should be

affordable and accessible through locally adopted

technologies, i.e., especially mobile (dumb-)phones.

Developing countries are some of the poorest in the

world, where large parts of the population live on less

than e 2 per day.

In order for a voice service to be

of use to the general population, the cost of accessing

and using it thus have to be very low. This implies

that the users should be able to access the service

without having to purchase a new device or service,

but rather using a device they already own or have

access to. The initial and running costs of a voice

service should also be low enough to be affordable

(and to provide sufﬁcient return on investment) for

Sub-Saharan Africa: https://data.worldbank.org/indicator/

NY.GDP.PCAP.CD?locations=ZF

Ney Yibeogo - Hello World: A Voice Service Development Platform to Bridge the Web’s Digital Divide

the rural poor.

The voice-service platform should function with

limited infrastructure. The (digital) infrastructure in

these countries is unreliable and expensive, especially

in the rural areas. While some villages have access to

electricity, it is often unreliable and black-outs often

happen multiple times per day (also in cities). The

majority of the population does not have (direct) ac-

cess to electricity

. The hardware that hosts access to

Internet is slowly becoming more common (Poushter,

2016), but is very expensive and unreliable

, due to a

lack of local hosting and limited international (back-

bone) connections. While Internet adoption is low,

mobile phones have become a successful means of

communication in much of the developed world, hav-

ing become the main means of telecommunication in

sub-Saharan Africa (GSMA, 2016). The coverage of

mobile telephony networks is often quite good, cov-

ering a large part of the population (also rural areas).

3.3 Financial Sustainability of Services

The platform should be able to provide ﬁnancially

sustainable voice services. This is achieved by re-

ducing the cost of voice services — which consist of

hardware costs, development costs and maintenance

costs — as far as possible. This has consequences

for all elements in the architecture of the platform,

which have to be chosen and designed in such a way

that costs are minimized. Financial sustainability as-

sures that the services are targeted at the needs of lo-

cal communities and thus provide sufﬁcient value to

offset their cost.

The platform should facilitate development by

local developers with limited programming skills.

The development process of voice services should

thus be simple, ﬂexible, not require advanced pro-

gramming skills and should take place in a graphical

interface. A very small amount of the population

owns (or has access to) a computer, let alone a

connection to the Internet. Also, there are few local

software developers and technicians available for the

development and maintenance of local infrastructure,

systems and applications. From this small pool the

amount of software developers that have experience

with voice services will thus likely be extremely low.

Hiring foreign developers is not an option, as the

cost of foreign labor is extremely high, conﬂicting

with the above requirement of ﬁnancial sustainability.

Ensuring local development is thus essential for the

https://data.worldbank.org/indicator/EG.ELC.ACCS.ZS?

end=2014&locations=ML&start=2014&view=map

http://100mega.ml/

formation of a voice service ecosystem targeted at

the unconnected. In order to increase the size of the

potential pool of voice service developers, the process

of development should be accessible to users that

do not have programming skills. This simpliﬁcation

should allow for people with a basic understanding of

using computers to be trained in the development of

voice services. Besides the ﬁnancial aspect of local

voice service development, an additional beneﬁt is

that local developers have a smaller distance to the

end-user of the voice service, not only in a spatial

sense but also in social and cultural sense. This

improves the local relevancy of the services as well

as the understanding of the end-user’s needs. The

platform should facilitate the formation of small

businesses and entrepreneurs that are specialized

in developing and hosting voice services, enabling

them to make a living from selling customized voice

services to local companies and communities.

The platform should run on low-resource hard-

ware and be based on Free/Libre and Open Source

software. In order to keep the costs of running voice

services as low as possible — and thus contribute

to the ﬁnancial sustainability — the hardware used

in the platform should be cheap, robust, and con-

sume little energy. Another aspect that inﬂuences

the cost of the platform is the cost of software li-

censes. The prices of commercial telephony products

and other software are at a level that is acceptable in

the Global North, which is not affordable in the de-

veloping world. Furthermore the liberating nature of

open-source software allows for the practice of brico-

lage: tinkering with existing technologies in new and

innovative ways which allows for the formation of

successful innovations (Ali and Bailur, 2007). Ac-

cepting that the usage of technology cannot be tightly

controlled and that successful innovations often come

from unexpected directions, can be a determining fac-

tor of success. By explicitly granting the general pop-

ulation the freedom to use the technology in any way

they see ﬁt, practicing bricolage is facilitated and the

available technology is more likely to be (eventually)

applied in a way that is most relevant and innovative

to the local context.

3.4 Example Use Cases of Voice Services

Below we outline two examples of the types of voice

services that the platform should facilitate. These use

cases have been elicited and analyzed during our var-

ious ﬁeld visits to Mali and Burkina Faso.

Foroba Blon, a system for village reporting. We

brieﬂy describe here the case of Radio Sikidolo, a

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

small radio station in Konobougou, a village in the

south of Mali several hours from the Malian capital

Bamako. It reaches up to 80,000 listeners in the re-

gion. According to its director, Adama Tessougué,

this radio works with free-lance village reporters who

collect news and announcements in the surrounding

villages for broadcasting. Example topics are wed-

ding announcements, funerals, lost animals, inter-

views and interesting stories. In the absence of Inter-

net in these remote areas, village reporters use simple

GSM mobile phones to send news to the radio. For

this, the program maker at the radio station had to be

available in person on the phone, and then write down

the incoming information on paper for broadcasting.

Evidently, this task is time consuming and inefﬁcient.

Foroba Blon is a voice-based system allowing village

reporters to phone in and to submit spoken news items

that are off line stored in the system (Gyan et al.,

2013). Messages can then be accessed and managed

by the radio journalist through a web interface on his

laptop, without the need for Internet. The radio sta-

tion uses the messages for interactive programming,

or receives (ﬁnancial) compensation for the spreading

of advertisements and announcements. The Bambara

name Foroba Blon refers to the Malian village square

where everyone is allowed to speak out, though re-

spectfully.

The Foroba Blon use case has been used during

the evaluation of the platform, which is covered in

Section 5.2

Weather information crowd-sourcing. Many farm-

ers and families in Burkina Faso depend on rain-fed

agriculture. The rainy season is short (three months)

and so pertinent information on actual and forecast

rainfall is extremely important, for example, to better

plan cropping calendars and improve harvests. Dur-

ing recent collaborative use-case and requirements

workshops in Gourcy, Burkina Faso, organized by

local NGO Réseau MARP, regional radio stations,

the association of innovative farmers in the Yatenga

province, and the W4RA team of authors, it became

abundantly clear that important weather information

never reaches local farmers in Burkina Faso. Global

weather information is in principle available through

the Web, but it is not accessible to farmers that face

the familiar issues of lack of electricity, of digital

infrastructures, and issues of language and literacy.

Furthermore this information is often inaccurate, due

to a lack of measurement infrastructure and accurate

weather models. The Burkina Faso weather voice ser-

vice allows farmers to receive data on the amount of

rainfall, as measured by fellow farmers that have a

measurement bucket on their land. These farmers call

in their measurements periodically. Besides providing

other farmers with essential information, the informa-

tion is also used to accurately track historical rainfall

in the region.

4 KASADAKA TECHNICAL

IMPLEMENTATION

The platform that we propose is called Kasadaka

(talking box in a number of northern Ghanaian lan-

guages). The platform consists of a combination of

hardware and accompanying software. Figure 2 is a

visual representation of the architecture of the system

and highlights the interactions between the compo-

nents.

4.1 Hardware

The hardware forming the foundation of the

KasaDaka platform is the Raspberry Pi, which is a

low-resource computer based on an ARM processor

(like found in many smart phones). The main advan-

tages of the Raspberry Pi are it’s low power consump-

tion (and subsequently no need for cooling), good on-

board connectivity and the low price

(and thus also

a low replacement cost). As the Raspberry Pi does

not include a Real Time Clock (RTC), it cannot ac-

curately keep time when the power is lost. To solve

this problem, a small and cheap battery powered RTC

is connected to the Pi’s general connector. The Rasp-

berry Pi is a very popular product for experimentation

and many projects, and is thus widely available, mak-

ing it easy to replace should hardware problems arise.

To provide the Raspberry Pi with connectivity to

the local mobile phone network, a USB 3G modem is

used. The exact make and model of this modem can

differ, as long as it is on the supported hardware list

of the chan_dongle Asterisk extension.

4.2 Software

On top of the Raspbian Operating System several ap-

plications run that work together to provide the voice-

service functionality. Almost all applications used are

open-source and thus free to use and adapt.

Telephone exchange software: Asterisk. Aster-

isk is a very popular open-source Private Branch

Exchange (PBX) telephony application. It is used

A Raspberry Pi 3 (including case, power supply and SD

card) costs around e 60 at the time of writing.

https://github.com/bg111/asterisk-chan-dongle/wiki/

Requirements-and-Limitations

Ney Yibeogo - Hello World: A Voice Service Development Platform to Bridge the Web’s Digital Divide

Figure 2: Overview of the Kasadaka system architecture.

for the routing of incoming calls to its destination

using Voice-over-IP technologies. In the use cases

of the KasaDaka platform, Asterisk provides the

connection between the phone network (3G dongle)

and the VoiceXML interpreter. To enable Asterisk

to interface with the 3G dongle an extension is

required. Kasadaka uses chan_dongle

, which is

an open-source Asterisk extension that provides con-

nectivity between GSM/3G modems and Asterisk. It

enables Asterisk to receive and place calls using the

connected modem, as well as send and receive SMS

messages.

Voice application document standard: VoiceXML.

VoiceXML

is a document standard for voice applica-

tions, based on XML. It is a standard designed by the

World Wide Web Consortium and is used for creating

documents that describe voice-based interactions.

It supports interactive voice dialogues between the

computer and the user and usually contains text (in

written form) that is later processed by a TTS engine.

Responses by the user can happen through pressing

a number on the phones keypad of by speaking

(for this ASR needs to be available). As the voice

applications that use the Kasadaka framework mainly

focus on under-resourced languages, TTS and ASR

are not used (nor available). Fortunately VoiceXML

https://github.com/bg111/asterisk-chan-dongle

https://www.w3.org/TR/voicexml21/

also supports the playback of audio ﬁles, much alike

embedding images in an HTML page. This allows

the use of pre-recorded fragments to build up the

voice services, but restricts the way of interaction to

using the phone’s keypad. A VoiceXML document is

‘rendered’ for the user in a way that is comparable to

the rendering of a HTML ﬁle in a web-browser, but

in this case is done by a voice browser.

VoiceXML interpreter: VXI. The software com-

ponent that is used for ‘rendering’ VoiceXML ﬁles is

VXI

, a closed-source VoiceXML interpreter built by

the company I6NET

. VXI connects with Asterisk

as an end-point for incoming calls. When a call is

redirected to VoiceXML a pre-conﬁgured URL is

passed on to VXI, which it loads and ‘displays’ to

the user as initial voice interaction. Normally this is

the principal document belonging to a voice service.

VXI currently is the only closed-source component

used in the KasaDaka platform. While the goal is to

use only open-source software, there is no currently

maintained open-source alternative.

HTTP server: Apache. VXI loads the VoiceXML

ﬁles it interprets over a HTTP connection, just like

loading a HTML page on the web, but locally. In

order to serve these ﬁles (and the audio ﬁles that are

http://www.i6net.com/technology/voicexml-ivr/

http://www.i6net.com/

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

referenced in the VoiceXML ﬁles), a web server is

required. There are many open-source web-servers,

one of the most used is Apache 2.

VSDK development framework: Django (Python).

In order to make the VSDK easy to extend by

developers, Python is a programming language of

choice as it is a popular language that is well sup-

ported and has several popular web-frameworks. As

VoiceXML documents are comparable to HTML doc-

uments, most web-frameworks can also be used to

generate VoiceXML ﬁles. Django

was chosen as

the Python web-framework, as it has very good and

extensive documentation, is well-supported and fol-

lows a Model-View-Controller (MVC) methodology

(Krasner et al., 1988). Django is open-source and

has a rich collection of projects and libraries that can

be used to extend it’s functionality. Django has a

good implementation of internationalization function-

alities, which enable the interface of the administrator

interface to be translated to different languages.

4.3 Voice Service Development Kit

The main goal of the Voice Service Development

Kit

(VSDK) is to support the development of voice-

services in the context of the developing world. As the

voice services are hosted on a Raspberry Pi and Inter-

net connectivity is not to be expected, the develop-

ment of voice services happens off line. Using a web-

based interface is preferable to running a development

environment on a computer because it solves prob-

lems with compatibility (different devices, operating

systems) and reduces complexity (does not require in-

stallation of software). Another advantage of this ap-

proach is that the development and hosting of voice

services are integrated, allowing for instantaneous re-

sults (and testing) of changes made to the application.

The VSDK is hosted on the Raspberry Pi, which also

hosts a local wireless network, through which it is ac-

cessible. Local entrepreneurs can use the VSDK to

develop voice applications on the Kasadaka platform,

without requiring them to have programming skills.

The structure of the voice-application is stored

in the database, using Django’s model functionality.

When an element in the voice-application is requested

by the user in a phone call, the VoiceXML interpreter

(VXI) requests the element through an HTTP call.

Django then retrieves the information about this el-

ement from the database, and uses a view to ‘render’

the element in VoiceXML. The VoiceXML interpreter

https://www.djangoproject.com/

The VSDK’s code can be found on GitHub. See:

https://github.com/abaart/KasaDaka-VSDK

then interprets this VoiceXML ﬁle and ‘displays’ it to

the user.

While the interactions in voice services are always

different, most of them can be generalized to a small

set of interaction types, such as making a choice, play-

ing back an audio message, or recording (voice) in-

put of the user. The VSDK provides a set of these

building-blocks, which consist of a VoiceXML tem-

plate, view and an administrator interface to use and

customize them. The current set (which will be ex-

panded in the future) consists of a menu-based inter-

action, recording of user voice input and the playback

of messages. While this set is limited, it offers sufﬁ-

cient functionality for many voice services and serves

as a demonstration of the method of voice service de-

velopment.

Voice-services in the developing context have to

support under-resourced languages, for which there

are no speech technologies available. The VSDK sup-

ports different languages in voice services by utilizing

pre-recorded audio fragments that are relevant for the

use-case domain. During the development of the ser-

vice, all the necessary voice-fragments are recorded

in the different languages in which the service has to

be accessible. These voice-fragments are stored in the

ﬁle system and referenced in a “voice label” element

that is stored in the database. This voice label refers

to voice-fragments that represent a fragment of text,

spoken in different languages(de Boer et al., 2015).

5 EVALUATIONS

The hardware conﬁguration of the Kasadaka platform

has already been tested previously in several pilot de-

ployments. These deployments have shown that the

hardware runs well in the conditions of sub-Saharan

Africa and allows voice services to be hosted indepen-

dently on all encountered mobile telephony networks.

The evaluation of the VSDK is structured in two

steps: the ﬁrst was an evaluation in the Netherlands

with inexperienced users, the second validation was a

case study in Mali with a user from the intended user

group of the VSDK.

5.1 Evaluation by Development of

Several Use Case Prototype Services

The VSDK was evaluated with 10 student groups dur-

ing the ICT for Development (ICT4D) course at the

Vrije Universiteit Amsterdam. The groups each de-

veloped a voice service for several distinct use cases,

which were co-created with rural communities and

relevant in the context of the developing world. The

Ney Yibeogo - Hello World: A Voice Service Development Platform to Bridge the Web’s Digital Divide

Figure 3: An example screen shot of the voice-service development interface of the VSDK. Shown is an example of a choice

interaction element. A voice service developer uses this GUI to develop voice-based applications on the Kasadaka platform.

Figure 4: Adama Tessougué of Radio Sikidolo shows the

Kasadaka on which the Foroba Blon voice service now runs

with its Bambara language interface.

choice to evaluate with students was made because of

the ease of communication with the students, which is

signiﬁcantly less complex and expensive than travel-

ing to a developing country. While the level of com-

puter literacy of the students is higher than that of

the intended voice service developers in developing

countries and the evaluation took place in the Nether-

lands, feedback of the students is still very relevant

Figure 5: André Baart and Adama Tessougué evaluating the

VSDK running on the Kasadaka platform, at Radio Sikidolo

in Mali.

for verifying the concept of the VSDK. The VSDK

proved to be successful providing the required func-

tionality for the creation of basic voice-service ser-

vices that can be used for rapid-prototyping purposes.

Using a graphical interface, voice-services consisting

of simple choices with associated options and mes-

sages can be designed without having to write any

code. These prototypes can be made in a quickly

and without extensive knowledge of the underlying

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

technologies, which is useful for rapid prototype de-

velopment and evaluation; After set-up, a simple ser-

vice can be developed and tested in under 30 minutes,

however the development of complex use-cases takes

more time. During the course, 80% of the student

groups had successfully built a working voice service

using the VSDK. These 8 applications were devel-

oped for 5 distinct use cases. The included interac-

tion templates allowed the students to quickly build

demonstration prototypes of their voice services. In

order to provide more complex functionality in their

voice services, 78% of the student groups had ex-

tended the functionality of the VSDK with data mod-

els speciﬁc to their use case and 67% of the groups

extended the VSDK with additional interaction tem-

plates.

At the end of the course the students were asked

to ﬁll in a short survey on their experience with creat-

ing a voice service and using the VSDK. The goal of

this survey is to learn about the process that the stu-

dents went through as they developed their ﬁrst voice

service. The survey consisted of statements about the

usefulness of the VSDK, which had to be answered

in a Likert scale. There were also qualitative ques-

tions about VSDK features, improvements and sug-

gestions, as well as questions about their perceptions

during the development process.

This evaluation has shown that the methodology

of building-blocks that is used in the VSDK allows

for the development of simple voice services by inex-

perienced users, which was the goal. It also provided

insight in the limitations and problems of the VSDK.

The main limitation lies in the are of user generated

data management. The VSDK does not yet allow the

creation of custom data models from the development

interface. Other limitations were the limited set of

user interactions provided and the impossibility of the

integration of external data sources. These limitations

prevent the VSDK of being suitable for more complex

voice-services, as ‘traditional’ voice-service develop-

ment skills are still required. In the case of a custom

extension to the VSDK, the functionality of this ex-

tension can be reused throughout the application and

shared with the rest of the development community

(through GitHub). Furthermore the administrator in-

terface can easily utilized by these custom extensions,

which allow voice-service maintainers (without pro-

gramming knowledge) to change settings and other

elements of the extension’s functionality. Thus af-

ter the development of the extension is completed,

maintenance can still be performed by others with-

out knowledge of it’s inner workings, maintaining the

advantage of ease of use offered by the VSDK.

5.2 Case Study: Radio Sikidolo

The results and knowledge of the ﬁrst evaluation

have been used in the second iteration of the VSDK,

which has been evaluated in collaboration with

Adama Tessougué, the director of Radio Sikidolo in

Konobougou, Mali. For more information see Sec-

tion 3.4 While the evaluation with the students was

sufﬁcient for a general validation of the methodology

of the VSDK, it did not evaluate the VSDK while run-

ning on the hardware of the Kasadaka platform, and

was not in the intended context of a developing coun-

try. This evaluation addresses these limitations: it

evaluates the VSDK and the Kasadaka hardware and

software as a whole, in the intended developing world

context. This validation session was done at Radio

Sikidolo, which has electricity and a relatively sta-

ble Internet connection, the latter of which is however

not used in the Foroba Blon use case. While Adama

is comfortable in the usage of a computer, he does

not have any advanced technical skills, such as pro-

gramming. However as he runs the radio station, he

is familiar with processing audio fragments (using the

open-source application Audacity).

The second iteration of the VSDK (which was

used in this evaluation) included various bug-ﬁxes,

and added several features such as service elements

that record user voice input, automatic conﬁguration

of Asterisk and audio ﬁle conversion. This evaluation

has shown that it is possible for a local agent to de-

velop and change elements in a voice service on the

Kasadaka platform, achieving the goal of enabling lo-

cally owned and developed voice services.

During the session Adama has been instructed by

the authors in the usage of the VSDK’s development

interface. Together we walked through the process of

changing properties in the interface, adding new ele-

ments (such as new languages), recording and adding

new voice fragments to the system and various other

aspects. After this short training of about an hour,

we asked Adama to go through the process again

by himself, in order to verify that he was able to

now use the VSDK on his own to change the proper-

ties of the voice service. This went successfully and

Adama found the methodology and functionality of

the VSDK to be well set up, and was satisﬁed with

the way in which he was able to develop and maintain

voice services through the development interface.

During the evaluation, Adama has successfully

used the VSDK to apply and adapt the included voice

service interaction templates to the Foroba Blon use

case. Support for the Malian language Bambara (for

which no TTS and ASR exists) was added to the sys-

tem by recording voice fragments and adding them

Ney Yibeogo - Hello World: A Voice Service Development Platform to Bridge the Web’s Digital Divide

through the development interface. The resulting ser-

vice was tested using the local phone network and

will be evaluated further during the programs of Ra-

dio Sikidolo. The radio will use the service in their in-

teractive radio programming, in which the local pop-

ulation (which is has a low literacy rate) can partic-

ipate by calling to the radio station (with their sim-

ple mobile phones) to ask questions or to state their

opinions. The combination of the Kasadaka’s hard-

ware and the VSDK allow for the off line develop-

ment and maintenance of voice services by Adama,

who falls in the intended user group for the VSDK

and the Kasadaka platform and thus does not have

any programming knowledge or advanced computer

skills (see Figure 4). During this case study the com-

bination of hardware and software in the Kasadaka

platform was successful in enabling the hosting and

development of voice based information services in

the context of the developing world. While it is still

a case study and the outcomes are not guaranteed to

be generalize-able, the outcomes show signiﬁcant po-

tential in bringing the advantages of the Internet to the

world’s disconnected populations. At the time of writ-

ing the Kasadaka platform is being reﬁned iteratively

and is regularly tested by Adama, who describes his

experiences with the platform through phone calls

with the authors.

Adama’s level of computer literacy is around that

of the targeted user group for the development of

voice services on the Kasadaka platform. These voice

service developers do not need to have programming

skills, but some knowledge of using computers is re-

quired such as being able to use more complex web-

based interfaces (such as a web-based e-mail client).

In the future these users could then be trained (over

several days) in the process of designing and develop-

ing voice services for local use cases.

6 RELATED WORK

This section covers existing efforts in the develop-

ment of Web-extensions in the developing context, as

well as tools and applications that facilitate the devel-

opment of information services in low-resource envi-

ronments.

Large-scale Voice Services. Voice based informa-

tion systems that use the local (2G) mobile telephony

network have already proven to be effective in reach-

ing the rural population of the developing world. To

support development of voice-based, mobile micro-

services Orange Labs developed the Emerginov

platform in 2012, targeting users in low resource envi-

ronments such as e.g. rural Africa. It includes support

for generation of voice-content in local languages,

such as Wolof, a local language spoken in Senegal.

Emerginov is normally hosted in the cloud, i.e. in

a data center, connected to the Internet and the local

phone network. Its hardware allows for 32 concur-

rent (in- or outbound) calls. Emerginov was techni-

cally promising, but the service has been discontinued

by the operator after a successful pilot. (Gyan et al.,

2013)

The company Viamo

runs several voice-based

information services in many African countries. Vi-

amo develops voice services for companies and

NGOs. The company has contracts with several

African telecommunication companies, allowing the

local population to call these services without cost,

using a toll-free number. The services Viamo de-

velops are mainly aimed at large populations, with a

very large number of concurrent calls. Although these

services are able to reach a large amount of people,

the large scale of the organization and the infrastruc-

ture that is required to run these large scale services,

causes services services targeted at the rural poor to

be ﬁnancially unsustainable.

Twilio Studio

is a web-based application that

allows graphical voice-service development by drag-

ging and dropping interaction elements into a call

ﬂow, which are the components in a voice service.

However the deployment of voice-services created in

Twilio Studio is limited to the Twilio platform, which

does not offer local phone numbers in many of the

developing countries where voice-services could be

relevant. This severely restricts the availability of the

voice services on the Twilio platform. Twilio Studio

seems to not be usable without an Internet connection,

which can not be assumed to be available. Further-

more, just like the previous examples Twilio makes

intensive usage of TTS and ASR technologies, which

are not available in the languages spoken by the local

population.

SMS-based Data Gathering Tools. In contexts

where a connection to the Internet is not available,

SMS can be used as a medium to exchange informa-

tion with an automated system.

RapidSMS

is a tool set that allows for the de-

velopment of SMS-based services for data collec-

tion and other work ﬂows. RapidSMS is devel-

See: https://emerginov.ow2.org/

See https://viamo.io

https://www.twilio.com/docs/api/studio

https://www.rapidsms.org/

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies

oped by UNICEF and has been used for various use

cases, including remote health diagnostics and nutri-

tion surveillance. RapidSMS is open-source and very

scale able to suit large deployments, but can also run

on a low-end server with a GSM modem. (Ngabo

et al., 2012)

DataWinners

is a data collection platform that is

developed by Human Network International

(HNI).

DataWinners enables the development of SMS and

smart phone based data surveys. These surveys are

primarily aimed at the context of NGOs that need to

retrieve data from their extension workers. By using

SMS data can be collected without a need for an Inter-

net connection, while the data can be still be entered

through a user-friendly graphical interface on a smart

phone. In the DataWinners web-based environment,

new data surveys can be developed in a graphical in-

terface.

Discussion. There exist several platforms for the

development and hosting of large-scale voice ser-

vices. These platforms allow for services that handle

many concurrent calls and are thus well suited to ser-

vices that aim to reach the general population. The

drawback is that the infrastructure and development

processes required for these services, are very expen-

sive and thus out of reach of the local population.

While SMS-based services provide data exchange

in contexts with limited Internet connectivity, it is

only usable by the literate that have knowledge about

the usage of SMS. Large populations in the develop-

ing world are illiterate or do not know how to use

SMS. Thus while SMS-based services work well for

data exchange without the Internet, these services are

not accessible for the general population in the devel-

oping world.

In conclusion, the existing solutions for the host-

ing and development of voice services and SMS based

information services are not capable of providing ben-

eﬁts that are comparable to those of the Internet, at a

cost that allows for ﬁnancially sustainable voice ser-

vices in the developing context. Besides the issue of

cost, other problems for the application of these so-

lutions in the context described in this article lie in

the area of support of under-resourced languages, the

centralized nature of these solutions, and the require-

ment of a reliable connection to the Internet.

https://www.datawinners.com/

http://hni.org/

7 CONCLUSION

The wider aim of the presented Kasadaka platform

and its Voice Service Development Kit is to allow the

populations on the other side of the Digital Divide

to share knowledge and create content, analogous to

the advantages that the Web provides. The platform

is lightweight and is tailored to the harsh circum-

stances that are found in the Global South and takes

into account the information needs of the local popu-

lation. By enabling local voice service development

and making custom voice services affordable for the

world’s rural poor, Kasadaka enables the formation of

a network of decentralized voice services. Such a net-

work has the potential to provide the beneﬁts of the

Internet to the rural poor, reducing the gap of the Dig-

ital Divide and helping to improve the quality of life

and well-being in the developing world.

Future work on the platform will focus on further

expanding the voice service development functional-

ity as well as more sophisticated data management,

to allow for the development of more complex voice

services. The VSDK is currently still too limited for

applications that enable complex data exchange, and

currently still requires writing code in order to support

these more demanding use cases. Furthermore, the

hardware of the platform is to be made more robust

to better withstand the conditions in the developing

context. Other ideas on further expansion include the

implementation of a TTS system that is suitable for

under-resourced languages, solving the dependency

on the closed source VoiceXML browser and allow-

ing for the inclusion of external data sources that are

available on the Internet.

Although there is still a long way to go, we believe

we have made plausible that for Web-like information

and knowledge exchange, alternatives to simple tech-

nology transfer are possible that do much more justice

to the local realities and needs at the other side of the

Digital Divide.

ACKNOWLEDGEMENTS

The authors would like to thank Adama Tessougué,

Amadou Tangara, Francis Dittoh, Gossa Lô, Julien

Ouedraogo, Matthieu Ouedraogo, Tjitske de Groot

and Gamariel Mboya for their contributions.

REFERENCES

Ali, M. and Bailur, S. (2007). The challenge of sustainabil-

ity in ICT4D-Is bricolage the answer. In Proceedings

Ney Yibeogo - Hello World: A Voice Service Development Platform to Bridge the Web’s Digital Divide

of the 9th international conference on social implica-

tions of computers in developing countries. Citeseer.

Bagshaw, P., Barnard, E., and Rosec, O. (2011). VOICES

Deliverable D3.1: Report on state of the art and devel-

opment methodology. Technical report.

Berment, V. (2004). Méthodes pour informatiser les

langues et les groupes de langues «peu dotées». PhD

thesis, Université Joseph-Fourier-Grenoble I.

Besacier, L., Barnard, E., Karpov, A., and Schultz, T.

(2014). Automatic speech recognition for under-

resourced languages: A survey. Speech Communica-

tion, 56(Supplement C):85 – 100.

Black, A. W. and Lenzo, K. A. (2000). Limited domain syn-

thesis. Technical report, Carnegie-Mellon University.

Bon, A., Akkermans, H., and Gordijn, J. (2016). De-

veloping ict services in a low–resource development

context. Complex Systems Informatics and Modeling

Quarterly, (9):84–109.

de Boer, V., Gyan, N. B., Bon, A., Tuyp, W., Van Aart, C.,

and Akkermans, H. (2015). A dialogue with linked

data: Voice-based access to market data in the sahel.

Semantic Web, 6(1):23–33.

de Vries, N. J., Davel, M. H., Badenhorst, J., Basson,

W. D., de Wet, F., Barnard, E., and de Waal, A.

(2014). A smartphone-based asr data collection tool

for under-resourced languages. Speech Communica-

tion, 56(Supplement C):119 – 131.

Farrugia, P.-J. (2005). Text to speech technologies for mo-

bile telephony services. Pace and Cordina [PC03].

Fuchs, C. and Horak, E. (2008). Africa and the digital di-

vide. Telematics and Informatics, 25(2):99–116.

GSMA (2016). GSMA report: The Mobile Economy:

Africa.

Gyan, N. B., de Boer, V., Bon, A., van Aart, C., Akker-

mans, H., Boyera, S., Froumentin, M., Grewal, A.,

and Allen, M. (2013). Voice-based web access in ru-

ral Africa. Proceedings of the 5th Annual ACM Web

Science Conference on - WebSci ’13, pages 122–131.

Heine, B. and Nurse, D. (2000). African languages: An

introduction. Cambridge University Press.

Krasner, G. E., Pope, S. T., et al. (1988). A description of

the model-view-controller user interface paradigm in

the smalltalk-80 system. Journal of object oriented

programming, 1(3):26–49.

McTear, M., Callejas, Z., and Griol, D. (2016). Creat-

ing a Conversational Interface Using Chatbot Tech-

nology. In The Conversational Interface, pages 125–

159. Springer.

Ngabo, F., Nguimfack, J., Nwaigwe, F., Mugeni, C.,

Muhoza, D., Wilson, D. R., Kalach, J., Gakuba, R.,

Karema, C., and Binagwaho, A. (2012). Designing

and implementing an innovative sms-based alert sys-

tem (rapidsms-mch) to monitor pregnancy and reduce

maternal and child deaths in rwanda. The Pan African

Medical Journal, 13.

Poushter, J. (2016). Smartphone ownership and internet us-

age continues to climb in emerging economies. Pew

Research Center, 22.

Rolland, C. (2007). Capturing system intentionality with

maps. In Conceptual modelling in Information Sys-

tems engineering, pages 141–158. Springer.

Schmida, S., Bernard, J., Zakaras, T., Lovegrove, C., and

Swingle, C. (2017). Connecting the Next Four Billion:

Strengthening the Global Response for Universal In-

ternet Access. USAID, Dial, SSG Advisors.

The World Bank Group (2016). Digital dividends, world

bank development report. Technical report, The World

Bank, Washington, US. DOI: 10.1596/978-1-4648-

0728-2.

UNESCO (2011). UNESCO report: Regional overview:

sub-Saharan Africa.

USAID (2017). Closing the access gap: Innovation to ac-

celerate universal internet adoption. Technical report.

WEBIST 2018 - 14th International Conference on Web Information Systems and Technologies