[Project-ideas] Language Model for Speech Recognition

[Sankarshan Mukhopadhyay]

On Fri, Mar 30, 2012 at 7:51 PM, Rabi Shanker Guha
<guha.rabishankar at gmail.com> wrote:

> I am willing to work on Language Model for Speech Recognition as my SoC
> Project. However since it is a research driven project I wanted to ask what
> approach to take?

It is less of "research driven" and more or of "risky" because you
have a higher chance of not being able to complete it or, failing.
That's about it.

> Approach 1: First implement known techniques such as N-grams, Smoothing and
> Corrections on N-Grams, HMM, etc. Then read up research papers and try to
> implement them and compare results.
> (If this is the preferred mode, can I have slots in my timeline dedicated to
> reading papers, and further slot saying 'Will implement the algorithm', not
> knowing precise details now)

The timelines, phases and milestones in your proposal are controlled
by you. The need for an explicit mid-term deliverable is something the
organization requires. So, you are indeed free to write a proposal
that allows you time to read up on the current state of the technology
before attempting to move forward.

> Approach 2: Proceed keeping in mind a fixed algorithm/technique before
> actually starting to code.
> (If this is the preferred approach, can someone direct me to the desired
> algorithm)

And, I repeat here again - no one is going to do your homework for you.

-- 
sankarshan mukhopadhyay
<http://sankarshan.randomink.org/blog/>