[Project-ideas] GSOC 2013 - Project : "Add a language model for speech recognition software for Bengali language"

[manish sharma]

Q. extending the dictionary for complex phrases/sentences ?

A. I have planned to use g2p tools like Phonetisaurus and sequitur-g2p and
right now I am analyzing which one i should use.

Q. MITLM ?

A. Thanks for this suggestion. Yes there are many other tools like MITLM
and CMUCLTK
for example SRILM, NGramLibrary, ya but SRILM cannot be used due to license
issues.
and yes mitlm is slightly better than cmucltk so okk i will use mitlm .

Q. Implement acoustic model.

A. I have decided to train the acoustic model by first collecting plenty of
data to train the model , I know that it will consume our time.


Manish Sharma
B.Tech,CSE ,IV year, Indian Institute of Technology Roorkee.
+91-7579048744


On Sun, Apr 14, 2013 at 11:25 AM, Bhavani Shankar R <bhavi at ubuntu.com>wrote:

> On Sat, Apr 13, 2013 at 3:22 PM, manish sharma
> <manish09.iitroorkee at gmail.com> wrote:
> > Hi !!
> >
> > A speech recognition engine has 3 components :
> >
> > 1) Language model.
> > 2) Acoustic model
> > 3) Decoder.
> >
> > As Each language language has distinct number of sounds and type of
> sound.
> > so we need to develop both an acoustic model and a language model.
> >
> > My plan how to develop a acoustic model and language model is shared with
> > this doc.
> >
> >
> https://docs.google.com/document/d/18gk39nrmSl6mOAYZ_zelVMnSPyS2-HY0CnLi03vxc44/edit?usp=sharing
> >
> > I have already shared it with you with a message
> >
> > GSOC 2013- Project : "Add a language model for speech recognition for
> > bengali language."
> >
> > Report is a little lengthy :).
> >
> > Looking forward for an early response.
> >
>
> Hi Manish,
>
> CMU sphinx looks fine for me. Just a couple of quick basic questions here:
>
> a) How do you think you can extend the dictionary for complex
> phrases/sentences (as a general view) and How do you think you can
> implement an acoustic model without much distortion, ensuring clarity
> for indic languages? (Since you have mentioned creation of acoustic
> model)
> b) What do you think about using a language model like MITLM along with
> sphinx?
>
> Regards,
>
>
>
> --
> Bhavani Shankar
> Ubuntu Developer       |  www.ubuntu.com
> https://launchpad.net/~bhavi
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ankur.org.in/pipermail/project-ideas-ankur.org.in/attachments/20130414/4b1e5852/attachment-0003.htm>