Pension department of the Rajasthan govt dispatches pension to the pensioners
under various schemes/categories. These are Widow/Physically-Handicapped/Old-Age
Pension, etc through postal money-orders or by direct credit in bank accounts on
a monthly basis. Most of these pensioners are located in rural areas and do not
have definite means of knowing when the dispatch/credit was made. Therefore,
they remained clueless and kept waiting for the pension to reach them. At times,
they even do not know that pension has been credited into their bank account and
kept waiting for the money-order to reach them.
The biggest challenge faced was the choice of communication medium. Sending
intimation by letters was prone to the same delays as faced by a money-order.
Since most of the beneficiaries were illiterate, they would again depend on
someone else to read them. Eventually it was decided to use telephones and
mobiles for reaching out to pensioners. To enhance the transparency in the
process and empower the pension beneficiaries with timely information, the
department decided to provide personalized information to them about their
pension sanction and credits on a monthly basis. So, this project for Automatic
Information Dissemination was conceptualized that would require the system to
dial each pension beneficiary who had a mobile or telephone number listed with
the department and play a message in Hindi stating the recipient's name and
relevant pension details. This process would be repeated each month.
QWhat was the technical complexity involved in implementing this Q Kindly tell us about the potential of this project ? |
For making phone calls and at the same time to save costs, they decided to
use technology instead of call center executives. And as a result they contacted
a Jaipur based IT company called Data Infosys to develop a text to speech engine
which can read through the database and call up numbers and read out the data,
amount and information to the pensioners. For this, Data infosys used open
source technologies to save costs. Festival tool was used as the base for the
text to speech engine. The biggest challenge was to read the database in English
and speak out the information in Hindi. But, as text to speech conversion was
using open source techniques, they were able to customize the code and add their
own keywords in Hindi.
![]() |
After the successful PoC test, the project was finally implemented as a
client/server software consisting of a very lightweight Web-based GUI front-end
from where the treasury operator can upload the database of recipients with
relevant pension details & mobile/telephone numbers of people who should receive
the AID call. The database is in a simple text format and can also be ZIPped to
conserve bandwidth. The front-end allows the operator to schedule the calling
immediately (in a few minutes) or the next day. He may also stop/resume the
calling process anytime, or upload the next database, as required.
Company Scenario |
Before Deployment |
|
After Deployment |
|
Implementation Partner |
Dr Ajay Data and Nitin Walia |
At the back-end the server components were used to store data on to an RDBMS,
serve the Web pages for the front-end, Hindi TTS engine to convert names and
words from Roman text to Hindi and an auto-dialer system to pick recipients from
the database, dial their mobile/telephone numbers and playback the customized
message to the recipient.
The backend server for this is a quad-Xeon with 2 GB RAM, 40 GB HDD,
FastEthernet NIU, fitted with a Tormenta ISDN PRI/E1 PCI card, housed in a data
center with an Internet leased line and an ISDN PRI/E1 line from a leading
telecom provider. The front-end is a regular PC with an Internet connection. The
server is running on Linux and using Apache as the web server.
The key challenge which they faced in this deployment was the format in which
the recipients' names should be provided. The department was maintaining names
only in Hindi using the format supported by the IST software they use for the
purpose. The department was notified that they would need the names in
Roman-Hindi mapping using the iTrans3 format. But their Roman data entry was
done based on common spellings, without using any phonetic standard. This again
posed quality issues with TTS synthesis of names and workarounds had to be
incorporated in software as well as TTS.
Finally, they had several issues with the efficiency of the TTS process where
it was taking too long and blocking the dialer process. They then had to
incorporate some smart scheduling coupled with caching techniques to ensure that
the TTS process did not delay or block the dialer process.