Speaking Alphanumeric Characters in a Telphone-based
Interactive Voice Application
Shirley Tobias
Lucent Technologies/Bell Labs
sbtobias@lucent.com
CHI '97 Workshop: Speech User Interface
Design Challenges
Short Biography
For the past eight years I have designed and evaluated phone-based user
interfaces for AT&T/Lucent voice processing and voice messaging products
and consumer service offerings employing both touch-tone and spoken input.
I am the author of the Intuity CONVERSANT Application Design Handbook,
which presents user interface design guidelines for voice response systems.
Design Problem
What is the most effective way for a user to speak alphanumeric characters
as data input to a telphone-based interactive voice application?
This problem manifests itself in applications where users are asked to
spell input, such as a name or catalog number. Although some systems
simply record users responses to prompts and later have the responses
transcribed, the design problem at hand assumes that the response will not
be transcribed by a person. To constrain the problem, let's assume that
the user does not have access to a touch-tone telephone to augment or
replace spoken input. Let's also assume that the interaction is completely
auditory; the user gets no visual feedback.
Issues of particular concern include:
- What form should the caller input take? Standard letters
(a, b, c. . .) or the military alphabet (alpha, bravo, charlie. . .)?
- Should the input be isolated (one at a time) or continuous?
- What is the most effective way to prompt for input (questions, tones?)
- How do users know when recognizer errors occur and how can users
correct them?
- How do users keep track of their place in the input string?
- Should users be able to use command words (e.g., wrong, stop, backup)?
If so, what are the best command words?
Imaginary Examples
- The user enters information with the military alphabet in a continuous
manner.
Please spell your last name.
sierra, mike, india, tango, hotel.
Did you say "smith?"
- The user enters information in isolated fashion (one item after each
tone). At any point the user can say "readback" to hear the input so far.
Should the user hear an error, the user can say "backup" to back up one item.
Please say one letter after each tone [beep].
[beep]
[beep]
[beep]
[beep]
[beep]
s,m,i,t,h.
CHI '97 Workshop: Speech User Interface
Design Challenges