![]()
Visiting guest. Why not sign in?
|
|
Computer Performs Tech-Support AI Taking Over Monotonous Task | |
Computer Performs Tech-Support
Quick mention of an article over on the NY Times entitled Need to Fix a Computer? Ask Another Computer. It discusses a piece of software that can do automated phone-based support. The program can perform speech recognition on the voice, and guide the caller through a set of context specific questions. This is essentially a tree-search algorithm, guided by the callers voice rather than getting him to type in number on the keypad. Despite the caller satisfaction still not being too high, this seems a viable option for night-time support. |
|
TuVox
There's a follow-up article on Internet News about the same topic: Start-Up Aims to Revolutionize Tech Support Call Centers Slightly more interesting to read than the first article posted. |
|
technology behind TuVox
"A small Mountain View, Calif.-based start-up called TuVox Inc., founded by two alums of Apple Computer, is aiming to change that by helping firms automate level one technical support calls with a combination of artificial intelligence and speech recognition utilizing the emerging Voice XML (VXML) standard...." -Internetnews AI is so broad, does anybody have info on exactly what technology tuvox is using, atleast some rough info? BTW, what happened to netTALK by Sejnowaski and Rogenberg, the famous application of NNs? Comments please. |
|
TuVox Technology
Hi vijay, welcome to AI Depot. The rest of this post is just speculation, so don't take it as anything else! The core of Tuvox is the speech recognition. This is some sort of pattern recognition for sure, but which specifically is hard to determine. Neural networks seem like a good choice since they can deal with noise and generalisation intrinsically. Based on that, you gather information about the problem of the user. Starting at the top of a tree, you ask him/her questions to refine the decision of which 'leaf' he is stuck in (i.e. what class of problem it is). Once this has been done, you can playback a recorded voice of a human to solve the problem -- actually speaking is not that important. All the messages can be pretty much recorded, and combined with the problem tree when that is created. |
|