Community Feedback > Feature Requests

Speach Recognition

(1/4) > >>

Any thoughts on this?  Or is it just too unreliable to be useful?  My past attempts with other software were not that great. 

Wondering if this could be done via the iPad, Elve, and Command Fusion though.

I can't say that I am that eager about it since it would mostly likely not turn out well, but might be curious.

John Hughes:
Believe it or not I actually have a simple Speech Recognition driver I wrote a year or more ago using the Microsoft speech engine.  The reason I have not worked more on it is because it would give too many false positives so I didn't think it was reliable enough to continue development on using the current speech engine.

Sorry for the necrobump, but figured I'd post as I'm actually thinking of working on something similar.
I currently have a program I wrote for my current automation setup that's using the microsoft speech SDK along with the Kinect, and it has 0 false positives, even when a movie or music is on full blast. I was thinking of adapting this program to Elve (since I am upgrading my automation stuff this year and wanted to try out different products to see what direction I want to take ;) ) The biggest hurdle in my code planning is how I'm going to grab all the information on Lighting, scenes, etc. from the server and get it into the program for the dynamic grammar generation part (at the moment I was going to use the XML communication protocol and see how far I could get with that). Not sure if this is the route you went, but any advice you can offer would be great!

Great news you may be taking a stab at this. Reading up over on CT an upstart is getting notoriety for the same type of endeavor. I would try to help, but I am currently knee deep in another driver. Would love to see this functionality in Elve though. Maybe the user could map a keyword to desired device or scene, and that way the driver won't have to automatically try to configure it all. It does seem like a lot of work to pull all the device info and build keywords off of that. Also doesn't seem to be user friendly as you probably would have to use the device's full name, and many of mine have abbreviations. Like 'turn on FR-Fan-Light' or 'Turn off LR-Chndlr'.  :)

Yeah, in one of my other versions, it learned commands as it went; it would scan through devices and check your sentence, then do a weighted probabilistic matching to determine what the most likely desired result is. It then would confirm that it's guess was what you wanted, and would then store that data so it would recognize the command in the future. That version also allowed custom aliases to be defined for devices (which could also be learned on the fly by the system). I've read a little bit of the driver API, though I was considering trying to do this via the XML interface  and have this run as an app that interfaces with Elve. Still drafting up some ideas, but haven't really had time to work on it at the moment (lots going on at my job taking up lots of time haha). I think I've tried out the program you are thinking of, it's actually really impressive, though it's still pretty early in it's development and (in my opinion) lacking some desirable features still, though that can change of course. The natural Language Processing of that program is pretty awesome though, and I imagine it's far more advanced than what I'm doing, which is just clever tricks with Regex.


[0] Message Index

[#] Next page

Go to full version