Design Inclusive Speech Systems

What should speech technology do for those who use it? The simple answer is that it should provide value. But the answer becomes much more complicated when you're trying to determine exactly what that value is, who should benefit from it, and whether or not that value has successfully been delivered. It becomes even more complicated when well-meaning organizations try to deploy a speech solution that's not only valuable to the organization but to its customers as well. In this case, customer value has to be determined. The challenge, though, is that value—like beauty—is in the eye of the beholder. What's valuable to one person might or might not be valuable to someone else.

If you ask a handful of customers what they expect from a speech-enabled system, they might tell you it should facilitate whatever task they're trying to accomplish—whether it's to pay a bill by phone, get driving directions from their mobile device, or dictate a message. But few might say they want the interaction to satisfy them. Yet customer satisfaction is increasingly the barometer that organizations use to determine the success or failure of a particular technology deployment.

Task completion is easy enough to measure. If a customer wants to pay a bill by phone and use a speech-enabled IVR to complete the transaction, then everyone should be happy, right? Not necessarily. What if the customer completes the transaction, but the experience is absolutely dreadful? In this case, the task's completion might indicate the interaction was a success, but the customer satisfaction score might suggest it's a failure. That's because when measuring customer satisfaction, there are usually other implicit expectations of the system that go far beyond task completion.

For example, many consumers prefer individual attention, but some personalization efforts can go too far and be considered creepy. Find out what some of these boundaries are and how not to cross them in our feature story "When Is Speech Technology Too Creepy?" by freelance writer Phillip Britt.

Also, if a company deploys a natural language system, callers might expect to be able to interrupt the IVR as they would someone in a normal conversation. Unfortunately, this could cause many existing systems to freeze up, send callers down the wrong path, or drop the call. But there are best practices to overcoming these obstacles; learn more in our Q&A with Nandini Stocker, senior voice interaction designer at Google (“The Art and Science of Error Handling”).

Finally, speech solutions should be accessible to people with disabilities, and not just physical impairments but cognitive ones. According to Deborah Dahl's column "Voice Should Be User-Friendly—to All Users," "Cognitive disabilities are both diverse and common; they include dyslexia, autism, ADHD, intellectual disabilities, aphasia, and dementia." To help, the World Wide Web Consortium has formed a task force to develop standards for Web accessibility for people with cognitive disabilities. This will become increasingly essential as more speech solutions integrate into the Web. Read Deborah's column for tips on reducing cognitive load in speech systems, or to find out how you can support the task force.

These examples suggest that, in terms of delivering value, organizations must look beyond the needs of the company and consider the satisfaction of everyone who uses the speech system.