Workshop Suggestions

[Ahmed Toulan]

Hi Ahmed Sayed,

I have a masters degree in a field related to speech recognition
(Spoken term detection). From my experience, for the recognition
part only, we need to have a good database. I tried to  convince
Sebastien(Google's developers relations manager) to help create
an Arabic speech database but I never heard back from him.

Forgive me guys I will get a little bit technical about the subject :)

The basic definition of good database is that it should be phonetically
balanced and have sufficient repetitions for monophones and triphones.
So, IMO what we need is to collect a good database if we really
want to build a robust recognition engine.

Of course we can build a simple recognizer with a few hours of
recordings but it won't be robust enough for real world applications,
and we can use this as a starting point to keep ourselves motivated :)

And yes I totally agree with you. Using HTK requires a lot of glue
code to make it work together. I mostly used bash and python
for this purpose. We can create a simple IDE to save some time :D
I wanted to build that IDE when I was doing my masters, but of
course didn't have time to do both :)

About the workshop. I guess what we mostly need is to spread
the word and try to get help from different universities. I can contact
my professors @ Cairo universities and may be u can do the same
@ ur university and anyone else who might interested in the subject.

Who knows may be this team will produce the a standard speech
database that will be used by researchers in the future.

Can u please give us some technical specs of the database u have.
How many subjects ? How long ? How many male/female ? Who
owns the database ? ..etc

Best regards,
Ahmed Toulan.

On Mon, Jan 10, 2011 at 11:30 PM, Ahmad Sayed <ahmad.ahmadsayed at gmail.com>wrote:

> Dear Ahmed Araby,
>
> I understand your concerns, but I think i did not make myself clear, the
> goal of workshop is to give a short intro for each group and how they can
> contribute
> e.g
> Group 1 that are focus in data collection and transcription, all they need
> is hints on how they collect the data and the convention used in
> transcription.
> Group 2 that have programming skills, focus in writing code to do a file
> processing By the way a lot of glue code required to put things together
> here the python, perl myself use Java part
> .
> Group 3 people with background in this area merge the code and enhance the
> engine.
>
> I think we need about 5 hours to have a simple workable engine. the
> workshop will be a true team work challenge and every one share will do
> something he may be never expect to be done before.
> Using a pure opernsource tool, under linux
> to get out the engine done for simple task, please have a look at
>  http://www.voxforge.org/
>
> http://www.voxforge.org/home/dev/acousticmodels/linux/create/htkjulius/tutorial
>
>
>
> I hope I make my idea clear.
>
> Best regards,
> Ahmed Sayed
>
> --
> Ubuntu-eg mailing list
> Ubuntu-eg at lists.ubuntu.com
> Modify settings or unsubscribe at:
> https://lists.ubuntu.com/mailman/listinfo/ubuntu-eg
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.ubuntu.com/archives/ubuntu-eg/attachments/20110111/43817a7d/attachment.html>