David Witham, with gotalk Ltd. in Australia, wrote:
I would like to point out what I consider to be inaccurate and out-of-date assertions made in Zeus Kerravala's article "Speech technology needs to be more natural to be useful."
In it he describes a quite common experience when dealing with speech-enabled automated systems. However, it's not the speech technology that is the problem – it's how the system is designed and deployed. The ability to construct speech applications using everyday grammar to vastly improve the efficiency of interacting with automated systems has existed commercially for more than five years. Generalised speech recognition (as opposed to technology used in mobile phones that simply matches sound patterns) is also quite good, although it can always be improved further.
There are many examples around the world where the technology has been deployed in the way Zeus wanted. The main problem is that the people in charge of delivering these systems are so interested in the sexiness of speech (to show how leading-edge they are) that the whole design process is subverted into nothing more than replacing DTMF key presses with words. This simply makes the application harder to use, more time-consuming and less accurate. At the other end of the scale, I know of a horse-racing telephone betting service that allows you to place a bet in one statement -- e.g., "I'd like to place a bet on the Doomben races on Saturday, race 5, horse 3, $10 on the nose." Imagine doing that with a traditional IVR.
So it's not speech technology that's the problem, it's making the right choices to design and deploy it intelligently.
Zeus Kerravala responded:
I think David's comments are fair. He's correct in that the technology is there to do what's needed today, but there are very few instances where the technology is actually deployed that way. David even states that it's the way the systems are designed and deployed that is the real problem. From the end user's viewpoint, though, it doesn't matter whether it's the technology or the deployment, the application needs an upgrade to be useful. So while I agree with David's comments, a big challenge for the industry is creating the actual proof points to show this is real.
Do you have comments you'd like to share regarding speech technology? We'd love to hear them. Send them to the SearchVoIP.com editors.
David Witham has worked for the past 10 years in the areas of prepaid and postpaid telephony application development and switching, IVR and speech-enabled systems, telecommunications billing, and IT system administration. He currently works in the engineering team at gotalk, looking after its VoIP switching infrastructure.
Zeus Kerravala is a regular contributor to SearchVoIP.com and manages Yankee Group's infrastructure research and consulting.
This was first published in June 2007