Kortum P. (ed.) HCI Beyond the GUI. Design for Haptic, Speech, Olfactory, and Other Nontraditional Interfaces
Подождите немного. Документ загружается.
7.6.14 Time-outs
When the user has taken no action after hearing a prompt, the system should
respond after the designated time-out period. The system response following an
unanswered prompt should take the form of
F First time-out or error: Repeat the prompt.
F Second time-out or error: Play the help message and repeat the prompt.
F Third time-out or error: Play the help message and repeat the prompt.
F Fourth time-out or error: Transfer to a customer service representative if
available; otherwise, play a polite message asking the user to try again later,
and then terminate the call.
Time-outs can be of variable length; at present, there is no single standard for
the amount of time to allow users to comprehend and act on a prompt before the
system responds. Factors such as the context of the prompt, the frequency that
the prompt will be used, and the complexity of the action should be considered
in determining the length of the time-out. A postmenu time-out is the time that
a user is given to initiate a response after a list of menu choices. The recom-
mended value for postmenu time-outs is five to six seconds. An interdigit key
press time-out is the time allowed between key presses in a multiple-digit entry.
It also functions as the terminating time-out for variable-digit entry. The recom-
mended value for interdigit time-outs is six seconds.
These are general guidelines and it is strongly suggested that time intervals
used in IVR interfaces be carefully tuned for the intended audience, as well as
for the situation in which they will be presented. Also, designers should be aware
that users whose phones have keypads in the handset (cordless and cellular
phones) may take longer to respond to prompts due to the time necessary to lower
the handset from the ear to access the keys. Finally, individuals with perceptual or
motor disabilities may also experience difficulty in responding to prompts, and
their specific needs should be considered.
7.6.15 Announcement Voice
The sound of the digitized voice should be consistent within the application,
as well as with all others offered by the company or organization. Whenever
possible, the same person should be used for the recordings, and the recordings
for a given application should take place in a single session. However, as it often
becomes necessary to record over several sessions (e.g., as changes and updates to
an application become necessary), it is understood that the same individual may
not be available when needed. When a different individual will do subsequent
recording, try to select a speaker in such a way as to minimize differences in pitch,
7 Interactive Voice Response Interfaces
262
accent, and other individual voice characteristics, as these differences can distract
users from paying attention to the content of the statements or prompts.
When recording spoken digits such as telephone numbers, the digits must be
recorded for each position in which that digit will occur. That is, any particular
digit will have different intonations depending on whether it appears at the begin-
ning, middle, or at the end of a spoken digit string.
Silence and pauses are just as important as the words being spoken. In the
following example, pauses occur at different places. In the first case, the inappro-
priate location of the pause might cause some users to “press 1 to send.” The
appropriate pauses in the second case reduce the possibility of that error.
Inappropriate:
“To listen <pause> press 1. To send <pause> press 2.”
Appropriate:
“To listen, press 1. <pause> To send, press 2. <pause>”
7.6.16 Techniques for Collecting Information
The techniques described below have been subject to multiple usability tests
(Miller, 1996) and are widely and successfully deployed.
Collect Personal Identification Numbers Securely
For security reasons, do not tell the user how many digits they are to enter. The
exception is when the user is changing a personal identification number (PIN)—
then the system
must
tell the user how many digits to enter. If the entered PIN
does not match the phone number and PIN pair maintained by the system for
the user, inform the user and reprompt for both phone number and PIN. Do not
voice back the erroneous data. Say instead, “The phone number and PIN you
entered do not match our records.”
Be Flexible When Collecting Phone Numbers
F
Never
ask user for a “10-digit phone number.” When a 10-digit phone
number is needed, prompt user with “Enter just the area code and
phone number” (or the regionally appropriate description of the phone
number).
F If the user enters a 1 or a 0 as the first digit, the system should ignore it
and collect 10 more digits.
F If the user enters the wrong number of digits, inform her: “You entered
only 8 digits.” Reprompt for the phone number. Do not repeat the short
set of collected digits.
F If the user leaves off the area code when asked for the area code and
phone number, prompt him to add just the area code. However, the
system should anticipate and accept either the 3-digit area code or the full
10-digit number as a response to this prompt.
7.6 Design Guidelines
263
Collect Days of the Week in Order
Prompt the user by reading the days, with each day followed by a number
corresponding to the day’s position in the week. Monday is day 1 and Sunday
is day 7. If “today” is a likely response, assign it to the pound key and offer it
as the first option; otherwise, start with Monday. “For today, press pound. For
Monday, press 1. For Tuesday, press 2.”
Be Flexible in Colle cting Times
F The system should accept input in military time (24-hour clock) format.
(However, the user should not be asked to enter military time.)
F Ask for hours and minutes, followed by 1 for A .M. or 2 for P.M.
F If the user leaves out the A .M ./P .M. indicator, prompt the user for it
again while confirming the hours/minutes part of the entry. For example, if
the user entered “2, 3, 0” corresponding to 2:30, prompt the user with “For
2:30
A .M., press 1. For 2:30 P.M., press 2. For a different time, press star.”
F The 0 (zero) key can still provide assistance, but only if it is the first and
only digit entered after the prompt for the time entry. After a time-out, the
user is either routed to a live agent or receives some other form of
assistance.
Define Time Periods with a Start Time and a Stop Time
F A start time and a stop time,
not
a start time and duration, define time
periods.
F The stop day is the same as the start day if the start time is earlier than the
stop time. On the other hand, if the start time is later than the stop time,
the stop day is the following day.
F Identify overlapping time periods and provide options to correct them.
7.7
CASE STUDY
A case study for an IVR can be found at
www.beyondthegui.com
.
7.8
FUTURE TRENDS
It is unlikely that people will be willing to live without voice mail in the future,
and so IVRs will continue so long as phones exist. In the next 10 to 20 years,
I would expect little to change in the field, as IVR technology is already relatively
mature. VXML has opened many doors, making it easier than ever to create and
operate an IVR. However, that has been a mixed blessing. Many easily created
7 Interactive Voice Response Interfaces
264
IVRs are poorly conceived and create headaches for their users. Promoters of
speech recognition systems have been predicting the demise of IVRs and the rise
of speech for the last two decades.
Currently there is little evidence that speech is about to take over the world,
or that IVRs are about to go away. IVRs, once built, tested, and put into operation
tend to be quite robust, and are fairly straightforward to update. Speech recogni-
tion systems require professional tuning and careful selection of grammar models
to be successful, and seemingly insignificant changes often require these steps to
be repeated.
One area for growth in the IVR field is standards. Currently there are few,
ANSI/ISO/IEC 13714-1995 being one, and in the experience of this author, exist-
ing standards occasionally promulgate suboptimal solutions. What would be of
great value to the designers of IVRs is a “tool kit” of proven solutions that could
be adopted and copied. The open-source movement and VXML have the possibil-
ity of creating a public library of usability-tested solutions to common IVR tasks
such as collecting a phone number, collecting a date and time, and entering a dol-
lar amount. With such a library in widespread use, users may be able to count on
some similarity across various IVRs rather then having to learn for each one “how
we do it here.”
Acknowledgment
The author is indebted to the many human factors professionals who have
advanced the field of IVRs. Researchers from Southwestern Bell/SBC, Pacific Bell,
and Ameritech have all made contributions to the body of knowledge that is car-
ried forward in the AT&T Human Factors Lab of today.
REFERENCES
American National Standards Institute (ANSI). (1992).
Standard User Interface to Voice
Messaging
. ANSI Working Document, Project 0976-D.
ANSI/International Standards Organization (ISO)/International Electrotechnical Commis-
sion (IEC). (1995).
User Interface to Telephone-Based Services—Voice Messaging Applica-
tions
. ANSI/ISO/IEC 13714-1995.
Carey, M. B., Chen, H. T., Descloux, A., Ingle, J. F., & Park, K. I. (1984). 1982/83 End
office connection study: Analog voice and voiceband data transmission performance
characterization of the public switched network.
AT&T Bell Laboratories Technical
Journal
63(9): 2059-2119.
Christodoulou, J. (1996). Strategies for Effective Voice Response Scripting. Draft. Austin,
TX: Southwestern Bell Technology Resources, Inc.
Davidson, J., & Persons, K. S. (1992).
Quickservice Billed Toll Inquiry: Technology Assess-
ment and Recommendations
. San Ramon, CA: Pacific Bell Human Factors Engineering
Laboratory.
References
265
Engelbeck, G., & Roberts, T. L. (1989).
The Effects of Several Voice-Menu Characteristics on
Menu-Selection Performance
. Boulder, CO: U.S. West Advanced Technologies.
Gardner-Bonneau, D. (1999).
Human Factors and Voice Interactive Systems
. New York:
Springer.
Gardner-Bonneau, D., & Blanchard, H. E. (2007).
Human Factors and Voice Interactive
Systems
, 2nd ed. New York: Springer.
Human Factors and Ergonomics Society (HFES)/ANSI. (2006).
Human Factors Engineering
of Software User Interfaces
. HFES/ANSI 200 (Canvass draft 2006). Santa Monica: HFES.
Marics, M. A. (1991).
Interactive Telephone Services: Designing for the User
. C3F.91.008.
Austin, TX: Southwestern Bell Technology Resources, Inc.
Miller, J. T. (1996).
An Interim Summary and Proposed Studies for Standardizing Interactive
Voice Response and Speech-Based User Interfaces for Advanced Intelligent Network
Services
. A3.126.96.79. Austin, TX: Southwestern Bell Technology Resources, Inc.
Pacific Bell. (1990).
Voice Messaging User Interface Forum: Specification Document
. San
Ramon, CA: Pacific Bell Human Factors Engineering Laboratory.
Pacific Bell. (1992).
Guidelines for Interface Design of Automated Response Units
. San
Ramon, CA: Pacific Bell Human Factors Engineering Laboratory.
Schumacher, R. M. (1992). Phone-based interfaces: Research and guidelines.
Proceedings of
Human Factors Society, 36th Annual Meeting
. Santa Monica, CA: Human Factors Society,
1050–56.
Schwartz, A. L., & Hardzinski, M. L. (1993).
Ameritech Phone-Based User Interface Stan-
dards and Design Guidelines
. Hoffman Estates, IL: Ameritech Services.
Shovar, N., Workman, L., & Davidson, J. (1994).
Usability Test of Alternative Call Director
Prototypes
. San Ramon, CA: Pacific Bell Human Factors Engineering Laboratory.
Simon, S., & Davidson, J. (1990).
Los Angeles CSO Audio Response Unit (ARU) Implemen-
tation: Customer Impact Assessment and Recommendations
. San Ramon, CA: Pacific Bell
Human Factors Engineering Laboratory.
Tourangeau, R., & Smith, T. W. (1996). Asking sensitive questions: The impact of data col-
lection mode, question format, and question context.
Public Opinion Quarterly
60:275–304.
Wickens, C. (1984).
Engineering Psychology and Human Performance
. Columbus, OH:
Merrill.
7 Interactive Voice Response Interfaces
266
8
CHAPTER
Olfactory Interfaces
Yasuyuki Yanagida
8.1
NATURE OF THE INTERFACE
Olfaction is the sense related to smell, and olfactory interfaces refer to devices
and/or systems that provide users with information through smells. Olfaction is
categorized as a kind of “special sensation,” for humans have a dedicated sensing
organ to detect the stimuli. Humans perceive smell by detecting certain molecules
in inspired air via the olfactory organ in the nose.
The olfactory interface is a relatively novel device. In modern computer sys-
tems, visual and auditory interfaces are highly developed, and there are numerous
ways for users to obtain information through visual and auditory channels.
Recently, haptic interfaces have been actively developed such that, in certain
environments, users are able to touch and feel virtual objects or environments.
Among the so-called “five senses,” only olfaction and gustation (sense of taste)
have been left unexamined. These unexploited sensations are chemical senses,
whereas the relatively well-developed sensory interfaces (visual, auditory, and
haptic) are related to physical stimuli. This fact makes it difficult to introduce
olfactory and gustatory interfaces in a similar manner to sensory channels based
on physical stimuli. However, researchers have recently begun to actively discuss
and develop the use of olfactory interfaces, including the introduction of an olfac-
tory display in virtual-reality (VR) systems (Barfield & Danas, 1996), and transmit-
ting olfactory information over telecommunication lines (Davide et al., 2001).
Figure 8.1(a) shows the structure of the human nose. In the nasal cavity, there are
three nasal turbinates separating the path of the air flow in the cavity. There is an
organ called the olfactory epithelium at the top of cavity, where smells are detected.
Figure 8.1(b) shows the detailed structure of the olfactory epithelium. Odor molecules
are captured by the mucus layer and then detected by olfactory cells.
Materials that have smells are quite limited. Carbon (C), hydrogen (H), oxygen
(O), nitrogen (N), phosphorus (P), sulfur (S), chlorine (Cl), bromine (Br), and iodine
(I) are the elements related to smell, but only the last three (halogens Cl, Br, and I)
have smells as single elements. Most molecules with smells are organic compounds
and have a limited range of molecular weight. Only molecules whose weight is
approximately 30 to 300 have smells; heavier and lighter molecules do not.
From the perspective of human–computer interaction, it may be interesting
to examine how many kinds of smell are required to synthesize arbitrary smells.
However, before discussing this important issue, let us review the case of vision.
It is said that we can “synthesize” arbitrary color by mixing three primary colors:
red, green, and blue. Physical light consists of electromagnetic waves whose
power is described as a continuous distribution function of wavelength or fre-
quency. This distribution is called a spectrum. Our visual system, however, might
regard light with different spectrums as a single color if the balance of power of
the three representative wavelengths (red, green, and blue) is the same. Hence,
the phenomenon of the three primary colors is not the nature of physical color
itself, but derives from the mechanism of visual sensation. This mechanism is
now used by various visual recording and display systems, including color televi-
sion sets, to transmit color information among remote places by coding color as a
three-dimensional vector instead of a continuous distribution function that can be
considered to have infinite dimensions. In other words, we can compress color
information by making use of our visual mechanism.
The question now is whether olfaction has a mechanism comparable to
the three primary colors—that is, are there primary odors? The answer is,
Olfactory epithelium
Olfactory bulb
Upper nasal turbinate
Middle nasal turbinate
Lower nasal turbinate
Nostril
(a)
Mucus layer
Olfactory cilia
Olfactory cell
Basal cell
Supporting cell
(b)
FIGURE
8.1
Nose anatomy.
(a) Structure of the nose. (b) Detailed structure of the olfactory epithelium.
Source:
From Ooyama,
Handbook of Sense and Perception
(1994).
8 Olfactory Interfaces
268
unfortunately, no. Humans can detect the smells of approximately 0.4 million
chemical compounds, and distinguish thousands (in some cases, tens of
thousands) of smells. Amoore (1970), in his theory of stereochemistry, sophisticat-
edly categorized various odors into seven major groups of smells. Some research-
ers considered that these seven groups correspond to the three primary colors, but
recent progress in olfaction research shows that this was not sufficient to explain
human olfaction. Later, Amoore increased the number of “primary odors” to
approximately 20 to 30, but this number has not been clearly tested.
A breakthrough in olfactory research occurred in 1991. Using an approach
from the field of genetics, Buck and Axel (1991) estimated the number of receptor
proteins in mice to be approximately 1,000. For this discovery, they received the
2004 Nobel Prize in physiology or medicine. Using recent data obtained in human
genome analysis, this number is estimated to be 350 in humans. Each receptor
protein responds to multiple molecules, resulting in the very complex functions
of human olfaction. Currently, this number (350) is considered to be an index that
shows the order of required odor components for obtaining arbitrary smells.
8.2
TECHNOLOGY OF THE INTERFACE
There are several technologies applicable to the development of olfactory inter-
faces. Because people smell by using their nose, odorants should be vaporized
and delivered to the nose. Thus, the major technical fields involved in creating
olfactory interfaces are the generation and delivery of smells.
8.2.1 Generating Smells
Most extracted smell sources are in liquid form (e.g., essential oils) or, occasion-
ally, solid substances soaked in liquid. However, humans can detect airborne
smells. This means that olfactory interfaces should include a device to vaporize
sources of smells of which there are several methods.
Vaporization
A straightforward way of vaporizing smells is natural evaporation. This method
can be used for liquid sources with relatively high volatility. By storing odor
sources in a small vessel, such as an essential oil adhered to cotton or porous cera-
mics, and letting air flow through the vessel, we can generally obtain scented air
in sufficient concentrations. If such natural evaporation is not sufficient, the mate-
rial can be heated to accelerate the evaporation. Certain types of odor molecules,
however, are easily destroyed by high temperatures, so heating of materials
should be avoided whenever possible.
Another way of accelerating the vaporization of a liquid source is by making a
fine mist of the liquid. One commonly used device is a sprayer. By continuously
8.2 Technology of the Interface
269
pushing the liquid out of a fine nozzle, the liquid becomes a fine mist that is eas-
ily vaporized. Electromechanical actuators can also be used to make a fine mist
from liquids. Piezoelectric actuators are often used to push a small liquid drop
out of the nozzle. With this method, the number of drops per second to be gen-
erated can be controlled so that the precise control of the aroma intensity can
be achieved.
Blending
There are two ways to bl end component odors at a specified ratio: in liquid sta-
tus and after vaporization. To blend odors while they are in liquid form, dilution
fluids (e.g., alcohol) a re typically used, because aromatic sources (essential oils)
are condensed and blending granularity would not be ideal if the undiluted solu-
tions are directly mixed. Diluted solutions containing essential oils are mixed in
drop units to make a mixed solution containing component odors at the desired
ratio.
Another way of blending component odors is to blend them after each odor is
vaporized. Air flowing through a vessel containing a liquid odor source becomes
saturated with vapor, which is then mixed with air at desired mass flow rate. This mix-
ing is usually performed by valves. Valvular systems that can continuously control the
amount of air flow are called mass flow controllers (MFCs). As commercially available
MFCs are often large and expensive, it may be difficult to integrate many MFCs
into odor-blending systems. Instead, a method to blend multiple component odors
by digitally controlling solenoid valves has been proposed (Yamanaka et al., 2002).
8.2.2 Delivering Smells
Once an odor source is vaporized, it should be delivered to the nose. The method
of delivery depends on the application style, which involves whether users move
or walk around, and how quickly the various smells should be delivered to the
user. The traditional way of enjoying smells involves diffusing aromas throughout
a room, where they are continuously enjoyed by users for a sufficient period of
time. This is the reason why traditional aroma generators are generally called “dif-
fusers.” Widespread, long-term aromas are sufficient for communicating environ-
mental or ambient information. Such smells do not disturb users explicitly, but
can still subtly communicate what is happening in the surrounding world. Most
of the aroma generators described in the previous section are categorized into this
type because they, by themselves, do not have an explicit function for controlling
the spatiotemporal distribution of smells.
This traditional way of enjoying aromas, h owever, may not be suffi cient if
we wa nt to make use of smells as a part of media technologies. If we consider
enjoying smells in conjunction with audiovisual entertainment such as movies,
TV programs, and interactive video games, the smell corresponding to each
8 Olfactory Interfaces
270
scene might need to be replaced immediately. The problem with traditional
“scattering-typ e” aroma generators is that they cannot erase the smell once it is
diffused within the space. Hence , a mechanism that localizes scent delivery to
a limited range (in time and/or in space) is required to use smel l as a p art of
multimodal media.
There are two approaches to spatiotemporal control of smells. One is to emit
the minimum amount of smell being delivered to the nose. Another is to incorpo-
rate smell elimination equipment into the system. The former approach has sev-
eral variations (see Section 8.3). The most direct way of delivering small
amounts of smell to the nose is to use tubes that carry scented gas from the scent
generator. If the scent generator is placed on the desktop, the user’s workspace
(movable range) is limited by the length of the tube. Therefore, making scent gen-
erators compact enough to be wearable is an effective solution (Yamada et al.,
2006). If the scent generators become extremely compact, they can be directly
attached to the nose to configure a direct-injection type of olfactory display
(Yamada et al., 2006).
Compact scent generators can also be mounted at the shoulders or chest
(Morie et al., 2003). Mounting scent generators to these positions reduces the
inconvenience of wearing devices and can result in natural human–computer
interactions. There have also been attempts to deliver small clumps of scented
air through free space without using tubes. By using the principle of a vortex ring,
launched from the aperture of a box, scented air can be carried through free space
(Watkins, 2002; Yanagida et al., 2004). This is an untethered approach for spatio-
temporally localized olfactory displays.
The second option for achieving spatiotemporally localized olfactory display
systems is to incorporate smell elimination equipment into the system. Without
such equipment, the emitted scent will be gradually diffused and reduced in den-
sity. Such natural diffusion is adequate if the amount of emitted scent is suffi-
ciently small. For relatively large amounts of scent, system designers should
also design smell-eliminating equipment. If the amount of the smell is moderate,
suction pumps and filters (such as a charcoal filter) can be used. For systems that
emit a massive amount of scent, ventilating functions should be incorporated into
the system. If these suction or ventilation functions are included in the system,
the scent-emitting subsystem is less critical in delivering smells to the user.
8.3
CURRENT IMPLEMENTATIONS
OF THE INTERFACE
Although olfactory interfaces are still at an early stage of development, several
successful implementations can be found. Systems are focused on generation of
aroma or spatiotemporal control of smells.
8.3 Current Implementations of the Interface
271