[Project-ideas] [GSoC 2013] Preliminary proposal on speech based query and result system

[Bhavani Shankar R]

On Wed, Apr 17, 2013 at 12:24 AM, srikanth ronanki
<ronanki.srikanth at gmail.com> wrote:
>> Hi Ronanki,
>>
>> Most part of the proposal is fine with me but I would be particularly
>> interested in the below
>>
>> a) Collection of data and extension of it to a more general sense
>> incorporating complex/compound sentences data.
>
> Collection of data is always a tedious task. I am used to it. I
> recorded a small corpus (400 words + 120 sentences) as a part of last
> year project as well. I can do it with my own lab members if required.
> I never did the second part. Extension of the data to incorporate
> complex/compound sentences seems a good idea. Definitely, I can try
> that given broad picture of what need to be done exactly.
>

What I was thinking was like extension of data to a more general set
broadly encompassing our daily speech styles and incorporating the
same.

Please excuse me if I'm not clear enough.

>> b) Voice activity detection algorithm you can use so that distortion
>> gets reduced
>
> Yes, we can use voice activity detection algorithm to reduce the
> distortion. Simplest way of doing it is to compute signal to noise
> ration (SNR) and detect voiced regions based on some threshold. Also,
> we can also use the time-efficient spectral subtraction methods based
> on VAD to reduce the noise as well if any.
>
> I worked on these lines to some extent earlier:
> http://web.iiit.ac.in/~ronanki/speech_enhancement/
>
> (sorry, if you couldn't play those files)
>

Had a look but unfortunately couldnt play the same.

>> c) extension to a global set for porting and extension into other languages
>
> Global phone set is some what similar to this type:
> http://homepage.ntlworld.com/stone-catend/trimain1.htm
>
> We can modify the latin symbols with phones that can be represented
> with English letters.
> For ex:
> Telugu - a అ, aa ఆ, i ఇ, ii ఈ, u ఉ, uu ఊ, rq ఋ, rrq ౠ, e ఎ, ee ఏ, ai
> ఐ, o ఒ, oo ఓ, au ఔ
> Bengali - ao অ, aa আ, i ই,ঈ, u উ,ঊ, rq ঋ, rrq ৠ, e এ, oi ঐ, o ও, ou ঔ
>

Fine by me.

>> d) Achieving word sense disambiguation
>>
> I don't have any great idea on this as of now. But, sure, I can work
> on this once I get to know about the corpus/text.
>

Does an annotator, something on the below lines help?

http://code.google.com/p/ytex/wiki/WordSenseDisambiguation_V08

>> and finally
>>
>> e) working 48 hra/week (which is equivalent to full time work)
>
> GSoC requires 30hrs per week and I know that it won't be pretty much
> same with my previous experience. 48hrs per week is the maximum time I
> can work on the project in any given week. Sometimes, I may take a
> week off during the project and some times I will work to the full
> potential. It all depends on with how much interest we are doing. I
> would be very happy to work on a project in Indian languages and to
> get a notable accuracy. I am looking forward to do this project and
> will definitely try my best to meet the goals and milestones as per
> the schedule.
>

Fine again. Thanks for your replies.


Regards,

-- 
Bhavani Shankar
Ubuntu Developer       |  www.ubuntu.com
https://launchpad.net/~bhavi