Quick and Dirty ... but possibly fun and useful little service GoogleSearch .. its a Text Publisher and is pretty good with the impressive NLP - heh I find it amusing and rather nice that Wikipedia turns out to be the source of most of the answers... BooYa ... 

There are a couple that did not reply with text I was expecting .. but it takes little effort to find out what is going on and correcting it.   The next thing to do on this is ask for pictures, and do image searches.

Will be fun to have a working pannous like back end again - lets hook the brain to it :)

 

I'd say this needs fixing .... let's buff it out !  This appears to be a call to pannous, and pannous in return is doing an invalid request to wikimedia.   Again, we should CUT OUT THE MIDDLE MAN and make this pipe shorter ... with fewer leaks :P

 

After a substantial refactor of SpeechRecognition interface, AbstractSpeechRecognition and  WebkitSpeechRecognition, the service appears to be worky.

Things left to do :

  • possibly offer an option between continuous and non-continuous mode. Currently, its not continuous and 'phrases' are quickly determined on a onend event.
  • throw up the error warning of zombie webkits when error event time events occur < 100 ms
  • still needs some polish

kwatters

4 years 10 months ago

I guess we probably need to be cautious about a screen scraping approach with a google search service.  that being said ... I love the idea of pulling in web and image search from google.  

we do have the ability to build our own indices.. if there was a corpus of knowledge that we wanted to ship with MRL.  Then it's a matter of mapping the user utterance to a query against it. 

Additionally, we probably want some more specific sort of knowledge sources..  we have (though I haven't tested them in a while)  weather forcasts and stock quotes.  I'm curious what other sort of sources would be nice...  movie times?  imdb?  email ? calendar? thermostats...   oh my..

it seems like the possibilites are endless.

 

I always liked the idea of a corpus composed of the shoutbox and mrl site would be exciting.
I'm pretty excited about the much more tighter modularity on services and their installation as resources under the resource directory (ie the index or whatever end structure we come up with coule be loaded as a module/service)