Dan Miller of Opus Research recently wrote a really interesting report on "conversational access technology," which he defines as "speech, real-time communications and service-oriented architectures." I thought his analysis and his conclusions were pretty enlightening. The report is called CAT 2007: Simplifying Search, Service Delivery and Customer Care.

As we prepare for the imminent arrival of the new year, it's a good time to think about applications of technology your call center deploys now that, until recently, were not readily available to most call centers. One of the best examples of these is speech recognition.

One of our friends from Opus Research sent us an interesting eyewitness account from a speech recognition tech event in New York.

Dan Miller, Opus's senior analyst, gave his take on Microsoft and Nuance's latest movie with Paul English -- we reprint it here (with permission) from its original source at the Opus Research website:

CBS News went to the SpeechTek trade show here in New York last week, and so did we. Sometimes it's fun to hear national news sources' take on the call center industry -- it can be both edifying and discomfiting -- but in this case CBS missed the point.

Last April, Peter Leppik of Vocal Laboratories, a Minnesota-based customer service survey firm, posted his list of the Top Ten Most Annoying Recorded Messages on the VocaLabs blog. Now, three months later, Leppik's wife tells him she's been asked by her employer to lend her voice to a short message that happens to be number ten on the list.

If you read the Q&A on our front page, you know that EIG's Rex Stringham is. And believe it or not, so are we. Sensational call center stories are hard to come by though, so when I saw the New York Times on Sunday, I was excited. William C. Taylor, founding editor of the hip business magazine Fast Company wrote a nice piece on English and the Great IVR Debate, but if you've been following Mr. English's media arc, you won't find any new ideas, just new commentators.

The most interesting parts of Taylor's article aren't about Paul English. The reason we're getting tired of Paul English is because he's a one-trick pony: he's got a complaint, and his function isn't to change anything -- only the call centers can do that. But the rest of Taylor's article is worth talking about because much of it echoes what we've been saying for a while.

There are three basic ways that hosted speech service providers price their services: per-minute, per-port, and per completed call. All three are measures of volume, but the last one takes the success of the call into account.

We asked Datamonitor's senior voice business analyst Daniel Hong a few questions about speech IVRs for our upcoming speech recognition service providers article.

Datamonitor came out with a report last July that a lot of speech applications vendors have been quoting. It says that hosted speech is growing, and growing fast. Will this be the end of premise-based speech apps? Not so fast, says Hong. Speech is growing in general, both on premise and hosting.

Enterprise Integration Group, Inc. (EIG) is a consulting firm that helps companies build better IVRs. Their mission: "EIG is the premier independent professional services firm specializing in simultaneously improving IVR utilization and customer satisfaction." EIG helps plan, design, and test IVR and automated speech recognition applications, among other services.

We asked EIG's president and co-founder Rex Stringham a few questions about hosted speech recognition services, and he responded via e-mail.

Will the Paul English anti-IVR machine ever stop? He launched a new website today, Gethuman.com, a dedicated forum for his IVR Cheat Sheet, tips on how to get to live agents, and -- best of all -- a full line of gethuman.com merchandise: t-shirts, bumper stickers, mouse pads, and teddy bears. Yes, cute, cuddly, anti-IVR teddy bears.

Coincidentally, we're working on an article for the April issue on Speech Recognition Service Providers. I've talked to a few vendors, and English's latest salvo is the perfect excuse to post some of our conversations with the people who make the systems that irk English so. In this entry, I'll focus on Peter Leppik and his customer satisfaction survey and testing firm, Vocal Laboratories. They don't make speech apps, but they make sure they work right.

Speech recognition technology seems to be about one of two things these days: replacing the human agent and connecting us to other technology when we can't (or don't want to) press buttons.

During IBM's press event for their "Future of Speech Day," I learned about talking cars, call routing systems that would allow me to yell and swear and still route my call correctly, Wake Forest University's experiments giving students pocket PCs, and the science of engineering a "pleasant transaction" through speech rec.

I've been reading a blog by the guys who run VocaLabs, a Minnesota company that does testing on customer service and IVR systems. They describe what they do as "collecting caller-focused data on companies' customer service operations."

A Guest Post By Brian Garr, IBM

Humans, despite all the visual and auditory cues we get from looking at a speaker, still have a word error rate of about 3%-5%, so for an un-seeing, un-hearing, un-thinking computer to understand speech, the task is monumental. IBM, after 35 years of work, and some 250 patents contributed by really smart people at IBM Research, concluded that a computer should be capable of recognizing human speech, and we are highly successful at it, to the point where we are deploying speech in cars, in handheld devices, and in the Contact Center. But will the pervasiveness of speech cause a revolution in the IT industry, or will it drive the evolution of IT and Contact Centers in terms of improving their ability to retain customers, bring down costs, and create new streams of revenue? Will speech revolutionize the way we humans interact with computers?

We're making it easy for you to get up to speed fast with all the information you need about speech recognition gathered into one handy place. Check out our new Info Guide: Speech Recognition for recent articles, roundups, trend analyses and vendor news.

I've received some interesting (and on occasion a bit heated) responses to my last blog, in which I expressed a little skepticism about the quality of most speech-enabled IVR applications. Walt Tetschner of ASR News, a longtime observer of the speech rec industry, is someone who doesn't pull his punches when he evaluates trends, developments, vendors, or, as I've now learned, bloggers. This is a most admirable trait in journalist. He seems to agree generally with my premise (there are too many bad speech apps out there), but he made some excellent points in some e-mailed comments that I'd like to share:

Walt suggested that I was far too focused on testing procedures, and not sufficiently attentive to flaws upstream in the design process. But my paraphrase perhaps doesn't do justice to his words. What he actually said was, "...the large number of poor speech-enabled implementations is directly related to the basic design process being flawed. Additional testing does not improve a poor design very much. Many of the poor implementations are performing precisely as they were designed and intended. Reality is that you cannot make chicken salad from chicken poop."

Walt's absolutely right (I tried that chicken salad thing, and it really doesn't work). Any design process that doesn't have the needs of the end-users in mind, and any set of testing specifications that doesn't address those needs, is doomed from the start. Good testing will not improve an otherwise bad application.

Still, usability testing should be an effective gateway to deployment, as well as an important step toward making post-deployment improvements. A good call center manager would not think of hiring an agent without putting him or her through a battery of skills tests, usually including simulated calls. I'd suggest that the same approach to testing speech recognition applications is healthy, if not always practiced.

Walt also said that it wasn't fair to compare a limited vocabulary enterprise app such as Avaya's UCC (which, by the way, I still find impressive) with more open-ended customer-facing speech rec apps. That's fair enough, so here's a more appropriate example that several readers have recommended to me:

Amtrak's Julie reservation and information system (1-800-872-7245) is a customer-facing speech app that is well designed, flexible, user-friendly, and properly tuned. She's very clear, she recognizes a variety of answers (she missed "You betcha!" as an affirmative response, but "Yup!" worked just fine), she offers important information (she told me that my hypothetical trip from Washington, DC to visit my uncle in Greenville, NC was subject to delays), and she's not shy in offering the option to speak with a live agent. All-in-all, Julie easily passes my "modified Turing test." Not bad for a four-year old.

The telephone directory assistance application from Tellme Networks also earns high marks in my book. If every speech app I dealt with were as good as these two, I would have written a much more positive, albeit more boring, blog. (BTW, Walt had an observation on these choices, too, he warns that preparing a speech interface that is stong on personality is expensive, and poses the risk that some people might not like the system. He, too, liked Julie, though.)

Some of my more astute readers have discerned that the purpose of my speech rec blog entry, and the related contest, was to identify the best practices in speech recognition, not to vent my ire. (OK, some of it was venting, too, but that was secondary;you try to get prescription info from Medco's IVR, and you'll want to vent, too.) In the best of all possible worlds, I'll receive so many first-rate speech app contest entries that I'll have to put my criticisms to rest for good.

Nothing would make me happier than having to eat some humble pie on this issue in the pages of Call Center Magazine. But, until that happens, I'm going to remain skeptical about the whole business.

by Allan Rosenberg

Posted
Monday, April 11, 2005
7:31 PM

Alan Turing, My Uncle Bobby's Voice, and The Quest For The Best Speech-Enabled IVR Apps