The Rocky Road to Speech Recognition

Strong internal, external resources needed for success The speech recognition industry is rapidly maturing. More than 10,000 commercial speech applications are now in use around the world, handling millions of calls every day. Speech recognition software costs less and performs better than it ever has in the past. A technology previously limited to the giants of customer service is now being adopted by many organizations of modest size that have more limited incoming call flows. Speech recognition is making its greatest strides today in organizations with 100 to 5,000 desktops and call centres of between 25 and 250 agents. So far, most of the contact centres using speech recognition are in travel, retail and financial services. Within the office, early adopters are using services such as voice dialing, appointment scheduling and unified messaging. The biggest advances are in hosted and call centre applications, according to Andrew Fisher, executive vice-president of NewHeights Corp., an Ottawa-based telecommunications software provider. Speech recognition delivers greater simplicity, and that means greater reliability. "You’re talking about a subset of commands, numbers, orders, a minimal number of commands, not a whole dictionary. You have much higher reliability, better user acceptance, huge ROI savings," Fisher said. But implementing speech recognition software is still not a cinch. Success depends on overcoming challenges in many domains, both technical and organizational, and marshalling the wisdom of many corporate departments. Let’s look at a few of the major barriers. First, there is no killer app in speech recognition, no technical breakthrough that guarantees its profitability in a wide range of businesses. Return On Investment (ROI) will be markedly different in every case: the key is to understand the ROI or business value of the planned application in a specific context. We know that speech applications can increase call completion rates and customer satisfaction, compared with conventional Interactive Voice Response (IVR) systems. But is the potential improvement worth the money and hassle? Everything depends on the business model of the user organization. "There are companies that will ask, ‘What is your speech strategy?’ I would suggest, ‘What is your automation strategy?’" says Tim Moynihan, Intel Corp.’s director of product marketing. "How does speech fit in as a component of your human contact centre, your Web portal and your telephony access, within the guidelines of your business model?" Numbers and BrandThere are two broad methods to evaluate speech recognition solutions, and they are not incompatible: by the numbers, or by the user organization’s brand. The first measures things like a reduction in call times or an increase in sales, while the second looks to the company’s image and reputation. How important is it that the organization projects the same personality in every locality and across every transaction? Most customer interaction happens on the telephone, and most often, the experience is a frustrating one. It makes sense to bring more value to every call, but this demands major investment. Vendors toss around numbers that show how important customer experience is to the bottom line, but it is not always easy to link satisfaction and loyalty levels to revenues. On the other hand, if hold times are getting out of control and callers consistently zero out of confusing menus, speech recognition may be a cost-effective solution. If the business model clicks, and the corporate infrastructure is up to date, there is no insuperable technical barrier to speech recognition. Even so, corporate structure may block effective implementation. If the various departments using and providing speech recognition—IT, sales, marketing, HR—do not work with a common vision, the stovepipes that exist inside the organization will probably duplicate themselves both in the voice solution and in the face it presents to the public. Technical ChallengeAnd the face of speech recognition does not always smile. Customers have learned to hate touchtone IVR systems, and they may be quick to despise voice solutions as well. In a conventional call centre, one agent speaks with one caller at a time. In a voice-enabled context, one voice speaks to everybody. First impressions may be critical to success: the technology may well have been installed perfectly and customers may still hate it. So fine tuning and careful scripting is vital. Applications are inherently complex. Speech capability calls for a platform that functions as a server, as speech software and as a network connection. Data feeds into the platform can come from ERP applications, from remote application providers, or from other servers. The applications on the platform require Advanced Speech Recognition (ASR) and Text-to-Speech (TTS) capabilities; there should be an advanced toolset for application development and maintenance; and, either on the platform itself or closely coupled with it, data gathering and assessment tools for effective management. The platform should, ideally, be able to work with both circuit-switched and VoIP networks. Moreover, even state-of-the-art solutions are still not perfected to the point of making voice applications fully interactive, capable of providing really fluid responses to callers and directing the conversation. Smooth dialogues with callers are still limited by vocabulary and recognition capabilities. "More progress has to happen with Natural Language Understanding and enlarged vocabulary applications to minimize that," says Ron Gruia, an analyst with Frost & Sullivan. Staffing IssuesThree factors have to come together for a successful voice implementation: the Voice User Interface, telephony, and linkage to the back-end infrastructure. Many of the barriers to speech-based telephony have been broken down, and the challenges are not so unfamiliar—many of them are really no different from those faced in a touchtone IVR application. The challenge really revolves around finding a person or a group of people who have the skill sets to span these three factors and drive the implementation forward. "It is still difficult on all fronts," says Intel’s Moynihan. "You may have a company with great Voice User Interface design and telephony expertise but they may be lacking in some of the database expertise, and they may have to work with somebody else. Maybe the enterprise has a unique back-end infrastructure." Getting competent outside help is not so simple either. "At the end of the day the value chain has become more fragmented," says Gruia. "So in the past you had just the IVR vendor who could come in and do the whole thing; and now you have a speech vendor, an application vendor, many different vendors. You need someone who can come in and bring the solution to market. You need a good integrator." Ongoing DevelopmentNor do the problems end with start-up: speech recognition is a process, not a product. The same development, testing and tuning that goes on before launch will inevitably continue afterwards, and as new applications are rolled out. Changes in the telephony or database environment may force changes in voice applications. Existing business processes may resist easy adaptation to a more rules-based structure. Organizations that are already collecting data from touchtone IVR systems may have to expand their categories to include a whole new set of parameters: ability to cope with noisy conditions; unfamiliar words and phrases; reasons for zeroing out to a live operator. "You have a pretty long application development cycle, so you have a lot of parameters that you have to take into account," says Gruia. "You have speech language complexities, format and platform integration, and audience. Noisy environments could still be an issue. The solutions are still perceived as being costly." A Certain StandardHowever, the trend of standards development is easing the task of implementation. In recent years, speech recognition has moved from craft to science, as standards began to prevail over the proprietary environment inherited from early IVR days. Today, the two main speech standards, Microsoft’s SALT (Speech Application Language Tags) and VXML (Voice eXtensible Markup Language), give speech buyers access to less expensive hardware, easily customized applications and, perhaps most importantly, easier integration with their Web, CTI and database back ends. Standards have allowed the development of toolkits for speech application development, shortening the time for development and testing, and allowing easier tuning and maintenance. Whether it is SALT or VXML, speech development now looks a lot more like Web development. Indeed, the closer voice and Web are linked, the better, because it greatly simplifies communication with back-end systems. Web-based communications and transactional tools have benefited from massive investment. Speech recognition can be an important channel to get more return from those resources. Meanwhile, the vendors are busy. In August, Avaya announced a new generation of IP-based speech self-service solutions for integration with Web services as well as more conventional environments. For its part, Microsoft announced it would push its Speech Server beyond call centres and IVR applications and into email and unified messaging, by integration with an upcoming release of Exchange Server. Nortel Networks’ Applications Centre Platform for IP-based contact centres integrates speech recognition, multimedia collaboration and unified messaging, tying them together with common toolkits and reporting applications. The trend toward modularization also helps. Many speech vendors offer dialogue components or code that can be reused in various applications with a standardized way of interacting. A credit card processing application, for example, can be written in a standardized way, and need not be redesigned. It would provide a more consistent user interface across applications within the enterprise and even across enterprises, so people would be able to interact with familiar speech applications in familiar ways. (For more on this trend, see "Speech Recognition Becomes Cheaper, Easier," in Telemanagement #227.) Speech of the FutureThe same standards that enabled the explosive growth of the World Wide Web may have a similar impact in the voice world. It takes no great leap of the imagination to envision callers connecting through an airline’s speech recognition system to a car rental system, for example, and having all their contact information move with them. As capabilities increase, organizations may well learn that they have many more ways to interact with customers than those that were available, feasible or cost-effective with live agents. "It enables business processes you wouldn’t have looked at before," says NewHeights’ Fisher. "All kinds of new ways of using this technology that previously you wouldn’t have had a call centre agent do. Now because you can automate database lookup, it can become a very seamless process, decrease your costs, and improve your turnaround time for business transactions. And the big win is making it much more pleasant for your customers and end users." The Perfect Host Hosted speech recognition is an attractive option for many organizations—and not only the small ones. Even when call volume is high, periodic and unpredictable surges in volumes may demand a daunting upfront investment. A hosted service will ease that problem, while offering lower maintenance, up-to-date enhancements and the ability to test drive a range of advanced applications at a reduced cost. A speech implementation requires close support. This factor may be boosting the attraction of hosted services: outsiders assume the risk to monitor, deploy and develop applications. Where enterprises don’t have the in-house expertise, and don’t want to buy equipment either, hosted services provide risk mitigation. Like many new-age technologies, speech recognition presents a puzzling mixture of opportunities and hazards. But its most distinctive feature today is the speed of change. Even organizations that do not wish to undertake speech recognition now will benefit from following its evolution closely. And in some industries, in some contexts, offering customers the considerate convenience of speech recognition is rapidly headed toward becoming a competitive necessity. Survey: Customers Prefer Do-It-YourselfIf your company’s call centre managers eschew new self-service methods like speech recognition for customers, they might be doing your firm a disservice, judging from the results of a survey from Genesys Telecommunications Laboratories Inc. The California-based contact centre software provider’s report, Contact Center Realities: Industry Insights for Success with Self-Service, indicates that customers are increasingly fond of and used to self-service mechanisms. The report says 85% of consumers think speech recognition is satisfactory or works very well, and 65% prefer voice self-service to touch-tone self-service. Nearly 60% of executives surveyed say meeting customer expectations is more important than reducing call centre costs; 62% of organizations using speech recognition have experienced improved customer satisfaction ratings. Genesys canvassed 200 North American customer-service execs and 300 consumers for its report. "This industry study confirms that self-service is a valuable tool for call centre management," says Wes Hayden, CEO of Genesys. Yellow Pages Opts for SpeechThis summer, the Yellow Pages Group in Toronto launched HelloYellow, which voice-enables the Greater Toronto Area’s 17 printed business directories. Without charge, callers can "voice up" hundreds of thousands of businesses, by category and by location. (Landline users who haven’t blocked Caller ID don’t even need to specify location.) "It is important for us in terms of sales, it is important in terms of market tests and branding, so there are a lot of important strategic issues for us," says Jean-Pascal Lion, vice-president of electronic directories for Yellow Pages Group. "It always boils down to the ultimate question: what is the most user-friendly experience?" Lion said he knew Yellow Pages Group was ready to invest heavily in speech recognition when the system was able to understand 75% of what customers said. "Below that, it is worthless," he said. "If you can get this guarantee that more than 75% of voices are recognized and there are no issues, fine." Advertisers readily accepted the product, Lion said. "Now we have a lot of customers on board, and we know we can sell it. Next January I will know the magic number, which is the renewal rate." Between now and then, he is counting on deals with carriers that offer HelloYellow service to 411 callers to drive enough telephone traffic to subscribed businesses.

The Rocky Road to Speech Recognition

News | 08/25/2005 4:00 am EDT

To keep reading, get a free trial.

Already a Subscriber?

Reuse & Permissions

Inside The Wire Report

Information

Hill Times Publishing

Get in Touch With Us