Re: [g-a-devel] speech recognition and accessibility

[Henrik Nilsen Omma <henrik ubuntu com>]

Willie Walker wrote:
> Hey All:
>
> Just a couple comments...
>
> I believe Speech2Text is using the Julius
> (http://julius.sourceforge.jp/en_index.php?q=en/index.html) engine,
> which seems to be a decent system geared towards dictation tasks.  
>
> An interesting and related project is VoxForge
> (http://www.voxforge.org/), which is a system for gathering training
> data and making it available for people to train acoustic models.  IMO,
> this kind of work is part of what is dearly needed to help open source
> speech recognition work.  
>   
Indeed, and I think it would be very helpful if any new SR front-end 
also had the facility to record speech data linked to text for upload to 
VoxForge. An eBook recorder that LibriVox could use to record their 
stuff would be great so that data can be linked to the text as well.

> I'd be happy to engage you in a discussion about various
> approaches to this problem.
>   
The question as usual is 'where do we start?' I guess :)  -- I'll email 
you to arrange some talks on this perhaps.

Henrik