|
|
My motto:
To develop a system that will parallel the
breakthrough of "what Mouse is to Computer, Speech Recognition will be
to Mobile/Computing Devices"
Hi!!
I am Kshitij Gupta, and its a pleasure to have you over at my "virtual
home", the 8 billionth and ..... page on the WorldWideWeb! Within this
home lies another home... the home of Speech-on-Silicon,
an initiative for the creation of a low-power co-processor
for real-time, Large Vocabulary, speaker-independent, Continuous Speech Recognition System-on-a-Chip
(real-time LVCSR SoC, in short). I invite you to take
a few moments to know more about me and my pet project!
Before we get started, a quick two liner about who I am, where I come
from, what my interests are.... I hail from the 400 year old city of
Hyderabad, home to "HiTEC City" & "Fab City"
(India's first proposed Chip fabrication plant). I got my Bachelor's in
Electronics & Communication Engineering under
Osmania University and
Master's in Electrical Engineering from
University of Pittsburgh.
My current hobbies include Speech Recognition, Speech Recognition,
and Speech Recognition.
Sounds obsessive? Well, it kind of is! You see, my two year Research
Assistantship during my Master's was focused towards the design of a "dedicated hardware architecture"
for continuous speech recognition By creating a
fully pipelined custom hardware architecture, my
thesis breaks the myth
that multi-GHz processors are required for performing speech recognition
in real-time. It is shown that a design using fully continuous models,
performing full-precision computations, running at less than 100 MHz is
sufficient for processing a 1,000 word command & control based
application. It is further shown that this design requires a small
silicon footprint thereby implying a minimal increase in the overall
cost of the final product.
Therefore, for speech recognition
to be an integral part of our daily lives, a dedicated hardware
architecture with bit-level optimizations needs to be developed for
processing computation and bandwidth intensive speech recognition
algorithms. We call this Speech Silicon, or, as I like to
more generically call it, the
Speech-on-Silicon initiative. With rapid enhancements in chip
fabrication technologies resulting in ever-increasing transistor packing
densities on a single die, I believe, the day is not far when just like CPUs
& GPUs (Central/Graphical Processing Unit), there shall be very few
products shipped without a dedicated SPU (Speech Processing Unit).
I believe:
"Just
as CPUs & GPUs (Central/Graphical Processing Unit) are an integral part
of products today, there shall be very few products shipped without a
dedicated SPU (Speech Processing Unit) in the years to come."
As for some of my
other interests, I really
enjoy playing Tennis. Although its been difficult taking time-off of late, you can be rest
assured that, weather permitting, its right up there on my *possible*
to-do list! When it comes to food.... Idly & Dosa rule, period.
Had been enjoying a weekly meal of Dosa since the past 2 years, until a
few weeks back that is. They raised the price of the buffet - so doesn't
feel that worth it anymore. Hence, these days I am open to second thoughts -
whether to substitute my garam garam
Sunday lunch of Indian Dosa with thunda American or Chinese or
Italian or Mexican!
Research Interests:
Apart from Speech-on-Silicon, I am interested
in the design, implementation & verification of SoC-based designs using ASICs, FPGAs, DSPs and configurable
processors for applications ranging from embedded consumer
applications to high performance computing (especially in multimedia, networking and communications). I am particularly interested
in multi-processor architectures, compilers, parallel processing, re-configurable computing and memory design,
with emphasis on low-power design. A more detailed discussion on developing a dedicated architecture for speech recognition
can be found on the
research
page. Unfortunately, since this work is currently going through the patent process, I am unable to post my entire thesis online. However,
you can go over the Abstract, Contributions and Conclusions drawn from
my work [thesis].
Want to be heard? Well, you can..... fill-in the
feedback form and tell me
what you would like your cell-phone/PDA to recognize!
The NEXT "Killer
App": "Speech-enabled devices providing the 'ideal' user-interface"
|
