Re: [libreoffice-accessibility] Re: [libreoffice-users] Re: Sphinx - voice translation - Linux

On 12/2/2014 6:50 PM, Tom Davies wrote:
> Hi :)
> I am wondering if anyone here uses any speech recognition packages or has
> any idea about which ones work well.  I think the preference is for
> something working well on Linux but it might be interesting to hear about
> others.

I've been using speech recognition since 1994. I currently use 
NaturallySpeaking 13 and, if you want to use anything other than simple 
dictation, it's the tool of choice. Sadly, the company has its head up 
its hind end with regards to how it treats disabled people but they 
don't hesitate to tell you about their Section 508 compliance but don't 
mention that it's only if you're working in Microsoft Word.

If you want to work on the Macintosh they do have a product for straight 
dictation but very little support for creating your own speech user 
interfaces. People keep trying to use the Microsoft Windows speech 
recognition environment but, they keep coming back to NaturallySpeaking 
because of the recognition accuracy and greater ease-of-use.

 A few folks who try to do things with Google speech recognition but 
it's really aimed at dictating limited amount of tasks that you correct 
by hand on a mobile device. I put this in the maybe someday category 
especially since Google, like Microsoft and Nuance, doesn't talk to 
disabled users who really understand how speech changes the system.

The biggest problem with using speech recognition is not the accuracy. 
It's not even finding a good microphone which, is an expensive process 
both in terms of time and money, it's the social aspect. Think about how 
annoying it is to listen to someone talking on their phone hearing only 
half the conversation and all of the disjointed chaotic forms of speech 
that humans use with communicating with other humans. Now, imagine 
something 3 to 4 times worse but you have the sound of a person talking 
to a computer, correcting mistakes, and giving commands in a way that 
makes no sense to someone who knows the native language.

I've been wanting speech recognition on linux for a very long time. , 
I'm not the only one. Back about circa 2000, I outlined the idea that to 
put speech recognition on other platforms, you should have a dedicated 
box it does nothing but run the recognition engine, language and 
acoustic processing etc. with a client-side wedge that makes it possible 
to communicate the local context back to the recognition environment so 
that the appropriate actions could be activated by a user supplied grammar.

the primary advantage of this technique is that you would  control all 
of the hardware, audio channel, performance issues in a dedicated box to 
for speech recognition (which makes it damned inconvenient for 
everything else) and then you could control anything else for very 
low-cost. You wouldn't have to worry about porting between different 
operating systems or hardware architectures. You just had one platform 
you could control. In theory today, we could be talking about a mobile 
phone. NaturallySpeaking running takes up roughly 1 to 1.5 GB. Well 
within the range of a high-end smart phone.

At the time, it was possible but not practical as it would require a 
dedicated laptop or machine to run the recognition engine. Today, the 
only technological barrier to implementing this model is the ability to 
deliver a clean audio stream to a virtual machine and enough hooks in 
the target OS UI to let you do user interface automation and state 
detection.

this is not terribly difficult project if you have hands that work well 
enough to write code freely. However, as is usual for most accessibility 
projects, 99% of the barriers to implementation are 
political/social/resource rather than technical.  so I pick away at it 
in my spare time. I make painfully slow progress and after 15 years of 
thinking about it, I have the start of an implementation for the KVM 
virtual machine environment. I'm currently being hung up by a problem 
with using the host USB audio device inside the virtual machine.

So tread lightly in your desire for speech recognition. If I wasn't 
disabled, I probably wouldn't use it. 10 years ago I would've said 
otherwise.  10 to 15 years ago more applications were accessible by 
speech recognition. Today, with all the different toolkits that pay no 
heed to the needs of speech recognition from users, it's even less 
usability.

and I think that's enough for now. I will end with this. LO's 
accessibility for speech recognition users is on a par with most other 
Windows applications i.e. almost nil. For some reason I don't 
understand, Select-and-Say capability does work sometimes so there is 
that small blessing but for the most part, most of the features of LO 
are not accessible and cannot be made accessible with the current system.

--- eric

--
To unsubscribe e-mail to: accessibility+unsubscribe@global.libreoffice.org
Problems? http://www.libreoffice.org/get-help/mailing-lists/how-to-unsubscribe/
Posting guidelines + more: http://wiki.documentfoundation.org/Netiquette
List archive: http://listarchives.libreoffice.org/global/accessibility/
All messages sent to this list will be publicly archived and cannot be deleted