More on Speech Interfaces

1
More on Speech Interfaces

• (lets take it slower and easier)

2
Common Mistakes

• Numbers are a sure of problem
• speechbuilder does not have modules/pkgs
• must build everything from scratch
• define digits, but make sure cover whole range
• Calculator example

3
Applications for Speech

• Killer Applications?
• Used in niche environments fighter pilots,
  telephone service, robotics
• Turn on lights, recognize dog talk (?)
• Computer games
• Why not more applications?

4
Speech Controlled Animation
5
Voice controlled animation

• A very fertile domain room for improvement
• mouse is very limited
• hard to specify parameters
• choose from list -- awkward when long
• one action and one parameter
• Other apps Mech. Design sketching

6
Testbed for other ideas

• Naming
• give basic object a name
• give composite object a name (macro)
• many parameters come from context (environment)
• differentiate between base object and
  instantiated object

7
(No Transcript)
8
Controlling Errors

• Two types of consequences to errors
• something useful (or interesting)
• something destructive (or boring)
• Who gets to decide?
• tolerating some errors --gt flexibility
• avoiding all errors --gt too rigid

9
Semantics

• Where does the semantics get checked?
• no consensus (speech, vision, sketch)

10
Our approach

• Command action and parameters
• error incompatible action and param
• dogs sit, run, lick, beg, bark
• cats sit, run, lick, sleep, purr
• Consider the error dog purr
• if cat is on stage, it purrs
• if dog is on stage, do random action
• random actor does random action

11
Considerations

• Really depends on the cost of error
• can action be undone easily?
• is the user getting frustrated?
• Rather than selecting at random
• choose the most likely action