So does this Speech Stuff really Work?
Posted by: Ken Day @ 17/03/2010 12:02:47 PM
I was toying with the in-car GPS the other day and I set it up for spoken addresses. ‘Great’ I thought, I shouldn’t have to type in an address again. In fact, it seems to work fairly well although it is sensitive to lots of background noise which there often is inside the car.
This got me thinking about my second home computer and a program called Software Automatic Mouth (SAM) which is used to read text out aloud. It was very basic speech software, however 30 years ago I was using this to check word documents and programs I had written by reading back the Basic programs. It was kind of kitsch and had nothing on HAL.
In the last ten years speech technology has become more widespread, we see it in mobile phones, GPS systems, in some home theatre systems, telephone betting systems, ticket booking systems, telephone number lookup systems and there are plenty of examples of how governments may use the technology in ways we don’t talk about (as seen in the movies).
So why don’t we have more speech call routing systems in the market? Why do I have to put up with annoying IVRs, particularly the ones with more options than life itself? Finally why do so many speech call routing systems just not provide a good experience?
The answer to the first question is directly related to the last question. So many systems fail to perform and provide a good experience, therefore businesses do not want to implement a system that doesn’t deliver the business outcomes required.
Why do I have to put up with annoying IVRs is an easy one. The reality is that a business is attempting to classify calls into different needs in order to provide a better experience (the call is answered by someone who can help with the enquiry) and minimise the agent training costs. IVRs are generally designed along business / product structures first and then “tuned” to make them a little more customer friendly. The problem with IVRs is you cannot create a menu system based on what service customers think they want or would ask for.
So finally onto the subject of speech systems and why some don’t perform.
The first problem, is simply recreating the IVR where the only difference is that you expect your customer to say a particular word or phrase as one of your choices. A system may have a small number of allowable words or phrases that it looks for with each "IVR” option. If the customer’s choice of words does not match the pre-programmed words it falls into “Speech IVR” mode asking the customer to repeat a particular keyword from a list the system provides, or the systems starts asking questions and is looking for a “yes” or “no” answer (this type of system is referred to as Directed Speech). Recreating an IVR is not a positive customer experience, unless you are one of the lucky ones who get through with your first spoken phrases.
The second problem is the issue of recognising what customers say. A system needs to be designed, built and calibrated appropriately. This includes areas such as investigating telephony / speech system performance, gathering examples of the particular phrases and words that a business’ customers would actually say, choosing the right question to elicit appropriate responses from the customer, using open speech and a statistical language model and possibly consider changing business agent skill groups to align with customer requirements. Once a system is in place, it should be tuned to ensure that recognition rates remain high.
So going back to my first question, does this speech stuff work? The answer is yes, like anything else, what you get out of something depends on what you put in.
Comments
Leave a Comment
- Display Name:
- Comment:
-
Did You Know?
47% of those interviewed in a recent study were uncomfortable providing personal details over the phone to a call centre representative.