May 17, 2012
Improvements to Speech Recognition Offers New Opportunities for IVR
Interactive Voice Response (IVR) systems have been evolving for decades. The latest spin on IVR technology, according to Plum Voice, is trying to solve a problem by developing something known as wake up words (WUW).
A recent Plum Voice blog highlighted that most IVR systems aren’t capable of determining which words out of those it hears are being directed toward it and which are not. A way around this is to establish a word or words that lets it know that the following correspondence is meant for the IVR to interpret and to react on.
Researchers at the Florida Institute of Technology are working on WUW to address the issue.At the forefront of any self-respecting IVR are excellent automatic speech recognition capabilities. One way to grow these capabilities is to develop a system by which humans can react more spontaneously with the computer or IVR.
The challenge in this process, however, is that IVR automatic speech recognition technology measured a couple of years ago shows that most conversations involve up to 200 words every minute with around one fault per minute. And that’s with the absolute best technology.
The WUW would trigger the machine to realize that the speaker is trying to interact with it, reducing the amount of faults.When humans address one another, the conversation often starts by addressing the person by name to alert them that they are being spoken to. The same could be done for computers or IVR systems. There is such a thing as “false acceptance” that could get in the way of this process, however. In order for the WUW to work, the IVR or computer must be turned on and ready to hear the WUW at all times. The WUW could be spoken but not intended for the computer or IVR.
Special attention must be given to the speech recognition system that specializes in detecting a word or phrase, and it must reject all other non-intentional mutterings of the key words and background noise.
This is akin to keyword spotting, but it must be able to discern unintentional trigger words. The developers of the WUW technology for computers and IVR systems are looking at the acoustic features of the recognition program utilizing a “triple-scoring” technique to zero in on the WUW and filtering out the rest of the conversation.
Researchers are showing some positive results – they have been able to use the WUW with 99.97 percent accuracy. The system has less than a one-percent fail rate in weeding out the non-intentional WUWs, which leads the researchers to believe they still have some tweaking to do.
Still, these advancements in IVR technology suggest there is hope for those self-service interactions that tend to lead to nothing short of frustration. If the digital to human interaction can be improved, customer satisfaction can increase as well.
Edited by Juliana Kenny