My IIIT Experience
I ll write about it in details sometime when i find some time. About the work i have done there you can see my resume. Only one this that i should mention about it is that it has left a bloody hell lot of confussed now. When i went there i was clear that i ll go for a job at the end of 4 year of BE and then after 2 years of job i ll go for MS. But now i really am badly confussed. What should i be doing ??
Ok here is the note that i am giving as an overview of my summer training report. Believe me there is much more to it. Will write some other time.
OVERVIEW
I have done my summer training at the end of 3rd year starting from 3rd June
2002 to 17th July 2002 at Language Technologies Research Center, IIIT Hyderabad.
Language Technologies Research Center (LTRC) is a premiere research lab in India
which aims at developing technologies dealing with language. This includes
technology pertaining to translation and other NLP areas, speech processing,
optical character recognition, etc. The main support for LTRC comes from Satyam
Computer Services Ltd. LTRC is presently headed by Dr. Rajeev Sangal who was
also the supervisor of my training.
During my stay at IIIT, I worked at the speech processing lab which is a part of
the sponsored wing of LTRC. I worked under guidance of Mr. S. P. Kishore who is
a Research Scientist with specialization in Speech Processing and Neural
Networks.
Speech processing lab was setup at LTRC in March 2002 and they have been working
on developing Speech Synthesizers for Indian Languages both in Limited and
Unrestricted Domains. When I reached LTRC they were working on Unrestricted
Speech Synthesizer for Indian Languages using a framework developed by Carnegie
Mellon University called Festvox. They had already developed Limited Domain
Synthesizer for Hindi and Telugu using the above mentioned framework.
During my interaction with Mr. Kishore in the first week of my stay at IIIT, we
explored the possibility for a suitable project for me. He mentioned the need to
have a very light weight limited domain synthesizer for Hindi and Telugu. Such a
synthesizer would make it possible to port the application they have developed
using the synthesizer to be used on Simputer, a handheld computing device. Mr.
Kishore gave me some idea and some notes of how he proposed to do the synthesis
and mentioned that it might just work. So I started working on developing his
idea with the aim to produce a light weight Speech Synthesizer both in terms of
memory requirements and processing requirements. We aimed at exploiting the
special properties of Indian Languages as the synthesizer was basically meant to
be used for Indian Languages.
Very soon I had a Light weight generic Limited Domain Speech Synthesizer for
Indian Languages ready. We tested the synthesizer with the speech corpuses
available with LTRC which they were previously using with Festvox framework. The
speech quality was excellent and also the synthesizer was very small (nearly 500
KB). Also it gave a satisfactory performance in terms of speed and memory
requirement. Now we needed to tie up this new synthesizer with the existing
applications and port the whole thing to Simputer.
The applications that were being developed were a Talking Clock for Hindi and
Telugu and a Travel Guide. I quickly browsed through the code of the Talking
Clock application and identified the statements that were required to be changed
to make it use the new synthesizer. Also some changes were made in the
synthesizer so as provide a suitable interface for the application to call the
synthesizer. Then I spent a day or two testing and debugging the talking clock.
I also check the defects in the speech corpus and labeling and correction were
made. Following this a complete application came up. Now we needed to port the
application as well the synthesizer to the Simputer. I spent a day exploring the
possibilities for this. I will discuss further on these in the report. Once we
had the application working on the Simputer, another round of testing was done
and some last bugs were eliminated.
After the limited domain synthesizer was finalized, Mr. Kishore suggested that
we move on to apply the principal involved with the Limited Domain Synthesizer
to Unrestricted Domain. I spent the last days of my stay at IIIT developing an
Unrestricted Domain Synthesizer for Indian Languages. The project is under
development and I hope to continue further on that work.
Besides this we worked on Unrestricted domain synthesizer that are on the way to be rolled on soon. And we thought of so many more things and as i am planning to back to IIIT this winter i hope more good work is on the way.