Click here to register.

General Discussion

Flat
Several questions / ideas
User: kmaclean
Date: 11/16/2009 12:48 pm
Views: 2100
Rating: 6

email from Sam:

1. We could hold a fundraising / grant drive where we seek to raise five thousand dollars, and then use that money to pay people to give a half hour of speaking time in exchange for five dollars, giving a total of about five hundred hours in addition to the sixty we already have or
2. We could collaborate with Canonical LTD to advertise voxforge speech submission, and then get them to package the full speech recognition suite
3. Question: are five hundred hours enough to begin the process of having the speech recognition automatically segment audio books, as was suggested? Or is that process even feasible?
4. Is there a wide disparity in the amount of male to female dictation?

-Sam

--- (Edited on 11/16/2009 1:48 pm [GMT-0500] by kmaclean) ---

Re: Several questions / ideas
User: kmaclean
Date: 11/16/2009 12:49 pm
Views: 60
Rating: 5

My reply:

Hi Sam,

Thanks for the feedback, my replies follow:

1. We could hold a fundraising / grant drive where we seek to raise five thousand dollars, and then use that money to pay people to give a half hour of speaking time in exchange for five dollars, giving a total of about five hundred hours in addition to the sixty we already have or
2. We could collaborate with Canonical LTD to advertise voxforge speech submission, and then get them to package the full speech recognition suite
Do you have contacts at Canonical?
 
3. Question: are five hundred hours enough to begin the process of having the speech recognition automatically segment audio books, as was suggested? Or is that process even feasible?
It is feasible right now, but the process still requires manual intervention for out of vocabulary words.  I have used this Perl script to segment audio books: http://www.dev.voxforge.org/projects/Main/browser/Trunk/Scripts/Audio_scripts/AudioSegmentation (see inline docs)
 
4. Is there a wide disparity in the amount of male to female dictation?
Yes - mostly male

I'd like to post this in the VoxForge forum for additional opions, please let me know if that is OK.

Ken

--- (Edited on 11/16/2009 1:49 pm [GMT-0500] by kmaclean) ---

Re: Several questions / ideas
User: kmaclean
Date: 11/16/2009 12:50 pm
Views: 77
Rating: 6

Hey Ken:

Do you have contacts at Canonical?

In a roundabout way, yes. I'm friends with a forum moderator / package maintainer and I spoke with him on the phone for a bit -- he suggested that I could sticky a thread on the Ubuntu forums that would advertise Voxforge speech recognition. I won't lie and misrepresent how much clout he has: he has forum power, but not a whole lot else. However, the forum has a ton of motivated people that would probably help us out if they knew about the project. In this sense I *can* get voxforge advertised. Speech recognition is a long term goal for the Ubuntu project, and it feels like there hasn't been enough inter-project communication to really get the ball rolling on this.

The donations thread feels like it died in a kind of inconclusive state. I didn't even know about the Mechanical Turk possibility; what would be nice is if we only had to pay out to people upon verifying their uploaded data to be good and usable. If we had people rely on the audio java thingy solely, that would be ideal for mechanical turk. This can be talked about more thoroughly in the forum, though.

I've already segmented an hour long audio book (spoken by a woman), but for some reason the script for finding out-of-lexicon words doesn't work for me. I was hoping that by doing the grunt work of matching up text to spoken audio I would leave the other stuff for people more knowledgeable than I.

--- (Edited on 11/16/2009 1:50 pm [GMT-0500] by kmaclean) ---

Re: Several questions / ideas
User: kmaclean
Date: 11/16/2009 12:50 pm
Views: 765
Rating: 5

Hi Sam,

If you want to contact your forum moderator friend to give VoxForge more exposure, then please do...  Any help is greatly appreciated!

thanks,

Ken

--- (Edited on 11/16/2009 1:50 pm [GMT-0500] by kmaclean) ---

PreviousNext