Kortum P. (ed.) HCI Beyond the GUI. Design for Haptic, Speech, Olfactory, and Other Nontraditional Interfaces

Подождите немного. Документ загружается.

that may take the user’s attention away from the VUI, and background noise

that may trigger misrecognitions—or there may truly be an issue with the

interface. Error prompting must work for a ll users. A good practice for error

prompts is to allow users to barge in q uickly and self-correct (for users who

know what they want and simply need a chance to repeat it), but to give addi-

tional information to help users who were unable to give a good response at

the initial prompt.

6.6.5 Provide Landmarks

Because audio-only interfaces are nonpersistent, one of the most challenging things

for users is staying oriented. Using landmark prompts can serve as a way to

acknowledge a user’s choice and reinforce a sense of place. For example:

YSTEM: You can say: account balances, last contribution, transfers,

allocations, or life insurance.

SER: Transfers.

YSTEM:

Transfers.

You can say: hear pending transfers or make a transfer.

6.6.6 Use Hints and Reminders

Presenting information to users before they are in a context in which they can

make use of it is ineffective. The more usable alternative is to provide instruction

to users in small chunks at times when they are likely to need the information

(Hura, 2006). Hints are information presented to users to give them context-

specific instruction for the current dialog state or to inform them of shortcuts

or alternate methods of completing tasks in the system. For example, if a user

has reached the “order status” state in the application by navigating through sev-

eral menus, consider offering a hint at the order status module: “Next time, you

can say ‘order status’ at the main menu to get here quicker.” Reminders, not

surprisingly, are brief prompts that remind the user of a relevant bit of informa-

tion at a particular state in the dialog, such as “You can speak your account

number or key it in.”

6.6.7 Favor Models over Instructions

It is sometimes necessary to provide an example to users of how to form a

response. In these cases, it is often easier and more effective to simply model

the desired sort of response than to explain it. Consider two versions of a prompt

requesting date of birth to be entered as an eight-digit number. The second ver-

sion, using a model, is more concise and much easier to understand than the first

version.

6 Voice User Interfaces

222

Version 1:

Please enter your date of birth as an eight-digit number, with two digits

for the month followed by two digits for the day and four digits for the year.

Version 2:

Please enter your date of birth as an eight-digit number. So, March

14, 1986, would be 03141986.

6.6.8 Use Discourse Markers Liberally

Remember, this is a conversation! Discourse markers serve to connect utterances

within a dialog. Discourse markers help establish a sequence, change the topic,

provide acknowledgment, and convey the speaker’s attitude toward what’s being

said. This does not mean that the VUI should respond to every user utterance with

“All right” or “Okay”—this tends to make the VUI sound overly perky. Instead,

vary the discourse markers with the conversational context as in the following

conversation (discourse markers are in italics):

YSTEM: Main menu. You can say: balances, transfers, transaction history,

or loans.

SER: Balances.

YSTEM:

All right

, balances. To get you that information, we need to get

you logged in.

First

, please enter your account number.

SER: 123–45–678.

YSTEM:

, enter your password.

SER: 54321.

YSTEM:

Thanks

. The balance on account 123-45-678 as of May 23, 2007 is ...

6.6.9 Allow for Conversational Repair

Human conversation is full of small problems. Speakers hesitate, start an utter-

ance and then back off and start again, and mispronounce words; listeners misun-

derstand and request clarification. We do all of this without thinking about it and

without the conversation failing in a vast majority of cases. When there is a small

problem in a VUI interaction, users naturally fall back on their techniques to fix

conversations (recall the example of the user speaking more slowly and clearly

after being misunderstood).

In crafting error prompts, be aware of this tendency and write prompts and

grammars that accommodate the user. A common example that is easy to accom-

modate is prompting following an explicit confirmation. If what is being con-

firmed is wrong, users often want to correct the mistake in a single turn:

YSTEM: To confirm, you’re looking for flights from Los Angeles to Boston

on May 23rd.

SER: No, to Austin.

6.6 Design Guidelines

223

To accommodate such corrections, build grammars that include “no” plus any cor-

rection the use might offer (here, “no” plus city names).

6.6.10 Be Careful with Terminology

Using terminology that is familiar and comfortable for users is a basic guideline for

any automated system. In the VUI world, this guideline takes on more importance

because the interface is nonpersistent. If users do not understand a term on a

web page, they can rescan the options with minimal effort. In a VUI, if users miss

a term, they have to spend the time to listen to the entire list of options again.

Steer clear of jargon and branded or technical terms in a VUI unless users use

these terms.

6.6.11 Give Users an Escape Hatch

Users detest being held captive in an automated system. Most users are willing to

try automation to accomplish simple tasks, but when the automation fails, they

just want a real person. Users also tend to pick up the phone when they have

already established that they have a problem—meaning that a user may be enter-

ing his interaction with a call center VUI already certain that automation cannot

solve his problem. Businesses deploying speech often want to deflect as many

calls as possible away from live agents and want to completely prevent users from

requesting a live agent. These efforts are often counterproductive in that users

always find the escape hatch, and if you have made them work hard to find it,

they will be frustrated and angry when they get there. Take the middle ground:

Offer to transfer to agents after you have evidence that the user is having a prob-

lem such as repeated time-outs or no matches. It is helpful not only to count the

number of errors at a single dialog state but also to use a global counter that keeps

track of errors for the entire interaction.

6.7

CASE STUDY

A case study for a speech interface can be found at

www.beyondthegui.com

6.8

FUTURE TRENDS

What’s the next big thing for VUIs? This is necessarily linked to the next big thing

for speech technology overall. One trend that has emerged recently is voice

search—using speech recognition to specify a search term and then receiving

the output auditorily or displayed as text on some device. One highly anticipated

6 Voice User Interfaces

224

voice search application is 1-800-GOOG-411, Google Voice Local Search, from Goo-

gle Labs (

http://labs.google.com/goog411/

). Google is the search engine giant on

the Web. The Google Labs decision to offer a speech option for specifying searches

over the phone makes a portion of the Web accessible to mobile users in a way

that it has not been until now. Searching the Web using even the most advanced

smart phone is painful. Typing search terms on a tiny keyboard is overwhelm-

ingly more difficult than simply speaking the term. Speech-enabling search is a

logical step.

It is interesting to note that the user interface of 1-800-GOOG-411 is quite min-

imal. There is nothing flashy or chatty about it. The prompts are brief and to the

point and are not necessarily phrased exactly the way a person would phrase

things in a human-to-human conversation. The VUI designers at Google clearly

conceive of 1-800-GOOG-411 as a tool, not as a live operator emulator. The

prompts for 1-800-GOOG-411 are certainly pleasant and not at all terse, but they

represent a stylistic shift away from the more verbose, more human-aspiring

systems of the last five years.

As of this writing, 1-800-GOOG-411 is still described as an experimental ser-

vice and is limited to Yellow Pages–type searches for local businesses. A number

of the features of 1-800-GOOG-411 are novel and would be useful in broader

searches as well. For example, the user can browse the results of a search in mul-

tiple ways; using simple voice or touch-tone commands, users can navigate the list

of results and select the results they want. Alternately, users can choose to have

the list of results sent via short message service (SMS, or text messaging) to their

mobile phone. Google gives users the choice of modality (speech and touch-tone

for input, audio and SMS for output) throughout the application. This decision to

be modality agnostic should serve tomorrow’s increasingly mobile users well.

When the context of use is “anywhere, anytime,” the user interface must allow

users to select the best way to interact based on their current circumstances.

Speech technology is also poised to move into the home. Appliances, home

security systems, and “smart home” systems will soon be supplemented with

speech technology. In these cases, speech recognition algorithms are unlikely to

reside on the devices themselves, but will be housed on a central server to which

the local device communicates in real time. These developments will be propelled

initially by the advantages that speech offers in terms of accessibility for the

mobility impaired. Being able to speak commands in order to control lights, tele-

visions, heating, cooling, and the like, would provide control and freedom to indi-

viduals who today cannot manage these tasks independently. It is likely that

speech in the home will be a phenomenon (like curb cuts) in which an affordance

designed to assist a small group ends up benefiting the larger community in unex-

pected ways.

Finally, there is the promise of speech for mobile and handheld devices. Mul-

timodal speech, touch, and graphics devices have been promised by many

companies large and small for more than 10 years, but there has yet to be a true

6.8 Future Trends

225

multimodal breakthrough. The proliferation of mobile wireless devices with small

screens and audio input and output makes the possibility of real multimodality more

likely over the next 10 years. Like any technology, breakthroughs will come when

multimodality enables easy access to an activity that is cumbersome or impossible

today.

REFERENCES

Attwater, D., Ballentine, B., Graham, M., & Hura, S. (2005). Voice user interface design:

What works, what doesn’t. Paper presented at SpeechTEK West Conference.

Baber, C., & Noyes, J. (1993).

Interactive Speech Technology: Human Factors Issues in the

Application of Speech Input/Output to Computers

. London: Taylor & Francis.

Ballentine, B., & Morgan, D. (2001).

How to Build a Speech Recognition Application

. San

Ramon, CA: Enterprise Integration Group.

Blumstein, S., & Stevens, K. (1981). Phonetic features and acoustic invariance in speech.

Cognition

10:25–32.

Campbell, J. (1997). Speaker recognition: A tutorial.

Proceedings of the IEEE

85:1437–62.

Cohen, M., Giangola, J., & Balogh, J. (2004).

Voice User Interface Design

. Boston: Addison-

Wesley.

Denes, P., & Pinson, E. (1993).

The Speech Chain

. New York: Worth Publishers.

Furui, S. (1996). An overview of speaker recognition technology. In Lee, C., & Soong, F.,

eds.

Automatic Speech and Speaker Recognition

. Boston: Kluwer Academic, 31-57.

Gupta, P., & Gilbert, J. (2005). Speech usability metric: Evaluating spoken language

systems.

Proceedings 11th International Conference on Human–Computer Interaction,

Las Vegas, NV

(CD-ROM).

Grice, H. P. (1975). Logic and conversation. In Cole, P., & Morgan, J., eds.,

Syntax and

Semantics

, Vol. 3:

Speech Acts

. New York: Academic Press, 41-58.

Heider, F. (1958).

The Psychology of Interpersonal Relations

. New York: John Wiley & Sons.

Hura, S. (2003). Passing the test.

Speech Technology

8.5 (September/October).

Hura, S. (2005). Great expectations.

Speech Technology

10.6 (November/December).

Hura, S. (2006). Give me a hint. In Meisel, W., ed.,

VUI Visions: Expert Views on Effective

Voice User Interface Design

. Victoria, BC: Trafford Publishing.

Hura, S. (2007a). How good is good enough?

Speech Technology

12.1 (January/February).

Hura, S. (2007b). My big fat main menu.

Speech Recognition Update.

Hura, S., Polkosky, M., & Gilbert, J. (2006). Point, counterpoint: What three experts say

about speech usability.

User Experience

5:2.

International Organization for Standardization. (1998). ISO 9241:11. Ergonomic require-

ments for office work with visual display terminals (VDTs)—Part 11: Guidance on usabil-

ity. Geneva, Switzerland: ISO.

Klie, L. (2007). It’s a persona, not a personality.

Speech Technology

12.5 (June).

Kotelly, B. (2003).

The Art and Business of Speech Recognition: Creating the Noble Voice

Boston: Addison-Wesley.

Larson,J.,Applebaum,T.,Byrne,B.,Cohen,M.,Giangola,J.,Gilbert,J.,NowlinGreen,R.,Heb-

ner,J.,Houwing,T.,Hura,S.,Issar,S.,Kaiser,L.,Kaushansky,K.,Kilgore,R.,Lai,J.,Leppik,D.,

6 Voice User Interfaces

226

Mailey, S., Margulies, E., McArtor, K., McTear, M., & R. Sachs. (2005). Ten guidelines for

designing a successful voice user interface.

Speech Technology

9.7 (January/February).

Lea, W. (1980).

Trends in Speech Recognition

. Englewood Cliffs, NJ: Prentice Hall.

Massaro, D. (2006). Speech perception and recognition: Theories and models.

Encyclopedia

of Cognitive Science

. New York: John Wiley & Sons, 1-9.

McClelland, J., & Ellman, J. (1986). The TRACE model of speech perception.

Cognitive Psy-

chology

18:1–86.

Mori, M. (1970). Bukimi no tani. The uncanny valley. MacDorman, K. F., & Minato, T.,

trans.

Energy

7(4):33–35.

Nairne, J. (2002). Remembering over the short-term: The case against the standard model.

Annual Review of Psychology

53:53–81.

Nass, C., & Brave, S. (2005).

Wired for Speech: How Speech Activates and Advances the

Human–Computer Relationship

. Cambridge, MA: MIT Press.

Nielsen, J. (1993).

Usability Engineering

. San Diego: Academic Press.

Pinker, S. (1994).

The Langauge Instinct

. New York: HarperCollins.

Pitt, I., & Edwards, E. (2003).

The Design of Speech-Based Devices

. London: Springer-

Verlag.

Polkosky, M. (2005). Toward a Social Cognitive Psychology of Speech Technology: Affective

Responses to Speech-Based e-Service. PhD diss., University of South Florida.

Pruitt, J., & Adlin, T. (2006).

The Persona Life Cycle: Keeping People in Mind Throughout

Product Design

. San Francisco: Morgan Kaufmann.

Rabiner, L. (1989). A tutorial on Hidden Markov Models and selected applications in speech

recognition.

Proceedings of IEEE

77:257–86.

Reeves, B., & Nass, C. (1996)

The Media Equation: How People Treat Computers, Television

and New Media Like Real People and Places

. Stanford, CA: Center for the Study of Lan-

guage and Information.

Resnick, M., & Sanchez, J. (2004). Effects of organizational scheme and labeling on task per-

formance in product-centered and user-centered web sites.

Human Factors

46:104–17.

Rolandi, W. (2002). Do your users feel silly?

Speech Technology Magazine

, November/

December—available at

http://www.speechtechmag.com/Articles/ReadArticle.aspx?

ArticleID=29420.

Rolandi, W. (2007). The persona craze nears an end.

Speech Technology

12.5 (June).

Sacks, H., Schegloff, E., & Jefferson, G. (1974). A simplest systematics for the organization

of turn-taking for conversation.

Language

50:701.

Schegloff, E., Jefferson, G., & Sacks, H. (1977). The preference for self-correction in the

organization of repair in conversation.

Language

53:361–82.

United States Access Board. (2007). Section 508 Accessibility Standard—available at

http://

www.section508.gov/.

Shiffrin, D. (1987).

Discourse Markers

. Cambridge: Cambridge University Press.

Stevens, K., & Blumstein, S. (1981). The search for invariant acoustic correlates of phonetic

features. In Eimas, P., & Miller, J., eds,

Perspectives on the Study of Speech

. Hillsdale,

NJ: Erlbaum, 1-38.

Weinshenk, S., & Barker, D. (2000).

Designing Effective Speech Interfaces

. New York:

Wiley.

World Wide Web Consortium. (2007). Web Accessibility Initiative—available at

http://www.

w3.org/WAI/.

References

227

This page intentionally left blank

CHAPTER

Interactive Voice Response

Interfaces

Jeff Brandt

7.1

NATURE OF THE INTERFACE

Interactive voice response (IVR) interfaces are chiefly telephony interfaces. They

are the much maligned systems that you reach when calling a business and do not

connect to a live person, but instead hear a recording something like this: “Your

call is very important to us.... To continue in English, press 1.” Most people can

readily recall instances of terrible experiences with IVRs. IVRs, however, are not

inherently bad. More often than not, they are created that way by their designers.

This chapter will explore the challenge of designing IVRs that are usable and

useful, and it will offer techniques for testing IVRs that will give the designer con-

fidence that, once fielded, the IVR will fulfill its design goals and leave its users

smiling (or at least not cursing) (Figure 7.1).

The most common IVR application is voice mail. Another well-known exam-

ple is a 24-hour banking application where account holders can check balances

and transfer funds from one account to another using the phone. The user dials

a phone number associated with the bank, which is answered by an automated

system (the IVR), which immediately plays out a recorded message: “Welcome

to the Big Bank. Please enter your account number.” Using the keys on the tele-

phone, the caller enters an account number. The IVR receives these key presses

and can then check the entered account number against a database and continue

with the interaction.

It is the playing of recorded messages or prompts and the collection of

responses through key presses that define a user interface as an IVR. On a phone

the key presses are made on the dual-tone multifrequency (DTMF) keypad

(DTMF was also trademarked for a time as “TouchTone”). The principles of IVR

design can also be applied to other interfaces that combine the elements of an

IVR with others such as those in a self-service check-out system. These systems

typically play prompts to communicate with the shopper, and often have a

numeric keypad used for responses, just like an IVR. Additionally, self-service

checkouts have barcode scanners and frequently a touch screen that can both

display information and accept input.

In general, the psychophysical requirements from an IVR user are minimal.

The buttons on phones are designed to be used by people with a broad range of phys-

ical abilities (see Chapter 2 for more detail on haptic interfaces). Individuals are free

to select a phone with keys that match their physical needs for size, labeling, and

push force, so long as the phone produces standard DTMF tones. Users who can

carry on a conversation with another person on a phone are usually able to hear

the prompts from an IVR. Some IVRs provide a means for a user to adjust the volume

or pace of prompts. Users requiring amplification require it of all calls, not just calls

reaching IVRs. Callers with these needs typically make use of a phone with adjusta-

ble volume, or use a hearing aid and a hearing aid–compatible phone (see Chapter 5

for more detail on psychophysical considerations of auditory interfaces).

While the user’s physical capabilities must be taken into consideration, the

most challenging human characteristic to consider in IVR design is the cognitive

capabilities of the expected users of the system. Quite frequently the intended

user population is the general population, with its inherent range of cognitive abil-

ity. Differences between the user’s expected task flow, the language model she

uses to describe the flow, and the vocabulary presented by the IVR can cause dif-

ficulties. The serial nature of recorded prompts also presents a unique cognitive

challenge. The designer must consider the fact that a user must hear an option

and the associated key to press for that option, one following another, and may

have to try to hold in short-term memory one or more possible choices while

simultaneously evaluating the continuing options on a menu. The fact that most

FIGURE

7.1

Most people expect terrible IVRs.

Poor designs have conditioned users to dislike IVR—especially when they are

expecting a person. Users tend to be surprised when an IVR has been well

designed to meet their needs.

Source:

From “Blondie,” February 7, 2007;

King

Features Syndicate.

7 Interactive Voice Response Interfaces

230

IVR systems have a finite waiting period for a response before timing out and

hanging up creates a further cognitive load. Techniques to make IVRs more acces-

sible are discussed in Section 7.3.1.

7.2

TECHNOLOGY OF THE INTERFACE

Historically, IVRs were built on telephone companies’ (telcos’) class-5 switches

such as the Nortel DMS100 or the Lucent 5E. These switches were used to provide

the IVRs that the telcos themselves might need, and in some circumstances to pro-

vide IVRs for large corporate customers. Class-5 switches have many limitations

that must be considered when designing an IVR. Typically, recorded prompts have

length limitations—often as short as 17 to 20 seconds. Prompts may have to be

recorded over a phone handset as a live recording, or may have to be recorded by

the switch manufacturer onto an announcement card. Some keys on the phone

keypad may have reserved functions that cannot be changed, and the timing para-

meters may not be sufficiently adjustable. The ability to interface with external data

sources (information not on the switch or in the telephony network) is limited and

in some cases not possible or not in real time.

Later, IVR capabilities were built into smaller telephony platforms including

CENTREX platforms and key systems, allowing medium-sized businesses to create

their own IVRs. Many of the limitations imposed by the class-5 switches were

reduced on these platforms. Currently, IVRs can be built on hardware as small

as a consumer-grade PC. Outfitted with a telephony modem or a purpose-built

telephony line card and software, a system to serve one or more lines can be cre-

ated for less than $10,000. Line cards allow phone calls to connect to the computer

that is hosting the IVR and playing the prompts. Line cards are manufactured

for both analog (plain old telephone service [POTS]), and digital (ISDN or T1)

phone lines. Voice over Internet Protocol (VoIP) calls can be terminated through

the computer’s Ethernet/Internet network connection.

Voice eXtensible Markup Language (VXML) now allows IVR service providers to

host your IVR code on their hardware—eliminating the need for small- and mid-sized

businesses to maintain telephony servers. Current IVR technology affords the

designer virtually complete control over the interaction. The limitations present in

IVRs today are mainly those that are inherent to the technology. First, information

must be presented serially due to the auditory nature of prompts. Second, responses

are limited to the 12 keys on the phone keypad, hanging up, or doing nothing, which

eventually results in a time-out. Third, the keypad makes number entry easy, but

alphanumeric entry is tedious, slow, and error prone. Fourth, although the phone

supports full duplex audio (both sides can talk and hear), IVRs cannot “hear” anything

other than the DTMF tones produced by key presses—all the unkind words shouted

at IVRs fall on deaf “ears.” Speech recognition interfaces, where the system can “hear”

(and hopefully understand) what is said, were covered in Chapter 6.

7.2 Technology of the Interface

231