1. Efficient Voice Information
Services for Developing
Countries
• Student: Pernilla Näsfors
Uppsala University, Sweden
• Supervisors:
– Dr. Roger Tucker
The Local Language Speech
Technology Initiative, England
– Dr. Mucemi Gakuru
University of Nairobi, Kenya
• Reviewer: Anders Berglund
Uppsala University, Sweden
Uppsala University, 29 May 2007

2. The Local Language Speech
Technology Initiative
(www.llsti.org)
• Information access by voice for people who
can’t access computers and the Internet
• Especially good for people who are illiterate or
visually impaired
• Using existing technology/devices
– Mobile usage and coverage in rural areas are
increasing rapidly and prices will decrease with
increased usage

3. Motives to use mobile phone
• Easy to use
– Easier to manipulate a phone than e.g. a computer
and many people are already using mobile phones
• Low investment and easy to maintain
– Doesn’t require Internet-connectivity and constant
power supply as computers do
• Provides information on demand
– Compared to e.g. Radio, Television, Newspapers

4. Text-to-Speech (TTS)
• Input any written text into a computer
program and the computer converts the text
into spoken output
• Dynamic system – no need to record new
messages
• Focus so far on technological development of
TTS in Kiswahili, isiZulu and Hindi and some
specific local languages in Nigeria and South
Africa

5. Kiswahili TTS
• TTS in standard dialect Kiunguja,
suitable for Kenya, Tanzania and other
East African countries
• First language of almost 800,000 people
• Second language of 30 million rural
people

6. Linguistic Digital Divide
• Information and Communication Technologies
for Development (ICT4D) was put into
spotlight at WSIS 2003 and 2005.
• WSIS in Tunis 2005 recognised a linguistic
digital divide and were committed to support
development of software in local languages
• Affirmed their “commitment to providing
equitable access to information and knowledge
for all, recognising the role of ICTs for
economic growth and development”

7. Purpose of project
• Demonstrate the technology by creating real
voice service
• Wanted real and useful information –
contacted the National Agriculture and
Livestock Extension Programme (NALEP)
• Farming provide livelihood to 75% of the
Kenyan population
• Difficult to get information today as the
Agricultural Extension Officers are few
(~5,500 trying to reach 800,000)

8. Why a Banana Information Service?
• Complement to extension workers
• Educational tool and reference system –
can confirm details
• Why we started with bananas:
– Grown in many areas
– Important crop in terms of income and food
security perspective

9. Sources of Information
• “Fruits and Vegetables Technical
Handbook” from The Ministry of
Agriculture and Rural Development
• David O. Nyantika, NALEP
• Prof. Esther M. Kahangi, JKUAT
• Kenya Agricultural Commodity
Exchange, KACE

11. Festival and Festvox
• A general multi-lingual speech synthesis
system developed at Centre for Speech
Technology Research (CSTR) at the University
of Edinburgh
(http://www.cstr.ed.ac.uk/projects/festival/)
• Free software written in C++ with a Scheme-
based command interpreter for general control
• The Festvox project aims to make the building
of new synthetic voices more systemic and
better documented, making it possible for
anyone to build a new voice.

12. GNU Bayonne
• A free telecommunications application server
• Supports a wide range of computer telephony
interface hardware and voice over IP
implementations
• Designed to be an application development
platform that easily allows developers to script
telephony applications and add support for
new hardware devices

13. XSL Transformations
• XSLT (Extensible Stylesheet Language
Transformations) is an XML-based language
used for the transformation of XML documents
• The original document is not changed; rather,
a new document is created based on the
content of an existing one
• The new document may be serialized (output)
by the processor in standard XML syntax or in
another format, such as HTML or plain text

15. Usability testing
• Testing with 10 farmers in Kibirigwi, Kirinyaga
• Full compliance with
strict ethical guidelines
and industry standards
relating to undertaking
tests with human
participants

16. Tests
• The tests were compiled by Pernilla Näsfors
and Yvonne Spittles, a PhD student at Thames
Valley University, England, and conducted by
Pernilla Näsfors and Dr. Mucemi Gakuru.
• Each participant was given 3 sheets containing
the following:
– a consent form giving their personal details
– a task sheet with questions to answer from using the
system and
– a user satisfaction questionnaire to give their
impressions of the system.

17. Participant Profile - Gender
Gender
F
30%
M
70%
70% of the participants were male
30% of the participants were female

18. Participant Profile - Age
Age
18-30
10%
31-40
30% 41+
60%
60% of the participants were aged 41+
30% of the participants were aged between 31-40
10% of the participants were aged between 18-30

19. Telephone usage
Telephone usage
Rarer than
weekly
20%
Weekly
50%
Daily
30%
30% of participants used a telephone daily
50% of participants used a telephone weekly
20% of participants used a telephone more rarely

20. Computer usage
Computers usage
Often
10%
A little
10%
Never
80%
80% of participants have never used a computer
10% of participants have only had little use of a
computer before
10% of participants used a computer often

21. Test Results - Ease of use
Ease of use of Banana Information
System
Not Sure
10%
Yes
90%
90% of participants stated that they found the
system easy to use
10% of participants stated that they were not sure
whether or not the system was easy to use

22. Ease of selecting options from menu
Clarity of Menu Options
Not Sure
20%
Yes
80%
80% of participants stated that they could easily tell
which option to select from the menu
20% of participants stated that they were not sure
whether they could easily tell which option to select
from the menu

23. Clarity of speech
Clarity of Speech
Not Sure
30%
Yes
70%
70% of participants stated that the speech from the system
was very clear to understand
30% of the participants stated that they were not sure
whether the speech from the system was very clear to
understand

24. Preference for gender of speech
Preference for Speech in Same Gender
No
30%
Yes
70%
70% of participants indicated that they would prefer the
speech to be in the same gender as themselves
30% of participants indicated that they would not prefer
the speech to be in the same gender as themselves

25. How usability is measured
• Usability is not one single element of a system but
moreover, is considered to be made up of five main
components which are integrated and can be classified
as follows:
– Learnability
– Efficiency
– Memorability
– Errors
– Satisfaction
• The tests undertaken on the Banana Information
System assessed the usability of all of the above
sections apart from memorability and efficiency which
could not be measured in a single test.

26. Learnability
• The Learnability component of usability refers to how easy it
is to use the system.
• Even though the participants had a low level of experience in
using Information and Communication Technology, 80% of
participants stated that they could easily tell which options to
select from the menu and 90% of the participants stated that,
overall, they found the system easy to use.
• 10% of the participants had low levels of literacy skills but the
results suggest that this did not affect their ability to learn this
system.

27. Satisfaction
• The Satisfaction component of usability refers to the
pleasantness of use of the system and the likelihood of
users being subjectively satisfied after using it.
• 100% of participants stated that they would use this
system instead of using other resources such as books
or the Internet.
• This is interesting as 60% of the participants can be
categorised as older users aged 41+ who, traditionally,
have been found to be more resistant to change in ICT
developments.

28. Errors
• The Errors component of usability refers to
the accuracy of using the system. The system
should have a low error rate (percentage of
wrong responses) and if errors are made they
should be quickly recovered from.
• The mean average error rate of 36% was
higher than expected and suggests that, as in
common practice with most usability tests,
some elements of the interface may need
modifications to meet the needs of the end
users.

29. Difference between languages
• All of the participants who listened to the
speech in Kiswahili stated that the speech
from the system was very clear to understand
but of the participants who listened to the
speech in English, 43% stated that they were
not sure on the clarity of the speech.
• A research version of a Kenyan English voice
were played to the participants and they all
said it sounded even clearer than the Kiswahili
voice. Further development of this voice is in
progress.

30. Other findings
• 100% of participants stated that the system
would be useful for banana farmers. They also
expressed a wish to extend the system to
other crops, and especially – to livestock.
• 90% of participants stated that if they had an
issue about growing bananas, they would
usually seek help from Agricultural Extension
Officers; 10% stated that they usually sought
help from other farmers. They commented
that there are not enough officers available.

31. Reactions from NALEP
• The system would fill an information
gap and increase the interest for
extension services
• The system was well received by 30
extension officers
• British English accent not clear – need
to use Kenyan English TTS
• Difficult Kiswahili grammar – need to
make translation informal

32. Reactions from KSB
• The Kenya Society for the Blind (KSB)
acknowledged the need for voice-based
services among people who are visually
impaired
• Their alternative is to have to rely on
other people to provide information to
them

33. Future system:
Other crops + livestock
• Easy to extend to other crops and to
livestock
• Support enterprise selection
• Can have separate phone numbers for
different crops
• Add information about e.g. value
addition

34. Future: Local information
• Local market information
• Suitable crops to grow in a specific area
• Contact information to the local
agricultural extension officers

35. Future: Up-to-date info
• Commodity prices and other market
information
• Weather
• Urgent announcements
– Disease outbreaks, e.g. bird flu

36. Future: Personal information
• Enter acres of land and get calculations of
possible revenue, required investments etc.
• Enter time of planting and get instructions
suitable for the stage your plants are in
• Can use caller-id and pin-code for
identification and keep database with all info
so it doesn’t have to be re-entered.

37. Future: Integration with SMS
• Can be given the option in the
telephone system to get key
information sent by SMS or e-mail
• Can subscribe to e.g. market
information
• Can get alerts via SMS of e.g. urgent
announcements, awaiting e-mails etc.

38. Future: Simple update of info
• Can create website where anyone easily
could update or enter new information
that immediately becomes available in
the telephone system
• Updates can be done at e.g. district or
provincial level by extension workers
• Farmers could also add information or
post questions to the system

39. Drawback: Cost
• Premium rate service:
Normal calling rate + ~5 KSh
• Can take 2-5 minutes from beginning of the
call to finishing listening to a section
– Can shorten time by providing information on how to
access things quickly
• Cost of SMS vs. Speech
– Time to read 160 characters is ~15 sec. which
translates to half the cost of the SMS

40. Ways of cutting costs
• Can provide landline phone at
Information Desks, or ”offline” version
of the service on a computer
• One call per Common Interest Group
• Possible to convince mobile operators to
make service very cheap at morning
hours?
• Get government and/or donors to
subsidise calling rates?

41. Other services
• SMS Further info
• Value added Information Service
– E.g. Commodities, Weather, Health,
E-government, Insurance, Banking
• TTS good for providing personal or
sensitive information

42. Other languages
• Possible for a linguist together with an
engineer to within 6-8 months produce a good
quality TTS system for a linguistically simple
language like Kiswahili
• Database at www.llsti.org with language
features that add complexity when building
TTS
• Can create limited-domain TTS before a full
TTS can be developed

43. Further testing and research
• Further tests on the interface of the system so
it is comprehensible by its target users
• Some improvements of the Kiswahili is
needed, e.g. for enabling simultaneous TTS.
• As 70% of the users stated a preference for a
speaker of the same gender, future
developments could include male and female
options in both languages

44. Conclusion of tests - Strengths
The findings suggest that despite low levels of
experience with Information and Communications
Technology:
• a system has been developed which meets the
satisfaction of the target group;
• the system is easy to use and;
• the target group would prefer to use this system instead
of some other resources that they already have
available.

45. Future development
• Extend to other crops and to livestock
• Possible to get local, personal and up-
to-date information, as well as
integration with SMS and e-mail
• Easy input and updates – enables
community involvement
• Improvements of voices – Kiswahili and
Kenyan English

46. Successful ICT projects
• Affordable
• Scalable
• Self-sustaining
• Sensible
• Simple ICT components
• Involvement of local community and
professional people

47. Access to information important
• The Kenyan government recognises
that the effective use of information and
knowledge is becoming the most critical
factor for rapid economic growth and
wealth creation, and for improving
people’s socio-economic standards and
quality of life

48. Thank you!
Asante sana!
Tack så mycket!