Artificial Intelligence Depot
Visiting guest. Why not sign in?
News, knowledge and discussion for the AI enthusiast.
FEATURES COMMUNITY KNOWLEDGE SEARCH  
Introduction to AI Robotics
Covers all the material needed to understand the principles behind the AI approach to robotics and to program an artificially intelligent robot for applications involving sensing, navigation, planning, and uncertainty.
More information at Amazon US UK
Talking to The Deaf
Digital Characters Do Their Thing
 
Talking to The Deaf

There was an article on BBC News two days ago discussing how Digital Characters 'Talk' to The Deaf. It covers the mergence of Speech Recognition and Character Animation in order to translate spoken english into sign language. This state of the art Artificial Intelligence application is going to be tested in post-offices in the UK. The benefit for the deaf or the hard of hearing is indeed obvious.

Technically, it's interesting to note that the system is based on "constrained phrases", which essentially means it tried to tackle a small set of very frequent sentences very well. It'd be interesting to see how well the system scales up, and if implementing learning into it would prove problematic.

934 posts.
Monday 04 March, 13:30
Reply
Deaf technology

The combination of the fields of natural language and character animation is one that is close to my heart at the moment (mainly because I'm up to my neck in a third year computer science project write-up involving them ;), and this looks like a great real world application of some of the research that has been going on the two fields.

It'd be interesting to know how much NL processing actually goes on behind the scenes, or whether (since it is restricted to a set of constrained phrases) it sticks to simple pattern matching on an input template.

I'd also be interested to see how well the speech recognition performs in a crowded environment with postal workers probably under stress and not wanting to have to bend down to talk deliberately and articulatory into the microphone screwed to their desk ;)

Andy

13 posts.
Wednesday 06 March, 19:43
Reply
Constrained Phrases

The quote from the article "constrained phrases" says it all.

In my opinion, they sample the microphone, and compute some sort of error tolerant hash of the audio stream. The closest match to this hash is fetched from the database, and the action is executed. I doubt theres much NLP going on there, if at all. Sorry to disappoint ;)

A very inflexible approach, but it's a good way to tolerate noise. All you need is a good way to compute the hash in a reliable fashion: some sort of dimensional reduction scheme would work wonders.

I doubt these early days of the technology will fully satisfy either party involved, but at least its a step in the right direction.

934 posts.
Thursday 07 March, 05:29
Reply