Today and the Future of Wearable Agents Emmett Coin Director of Speech Research and Development Spee

1
Today and the Future of Wearable AgentsEmmett
CoinDirector of Speech Research and
DevelopmentSpeechTEK 2007 West, February 21,
2007
2
what I will talk about

• Definition of a wearable voice agent
• Overview of voice-based agents in logistics
• the subsystems in a real world voice application
• Some of the more difficult issues
• Futures
• Summary
• Conclusion

3
what is a Wearable Voice Agent?

• NOT just a voice app on a cell phone or PDA
• Voice dialing
• Stock quote
• One shot use
• Rather it IS a partner that
• Complements the task (a teammate)
• Adds value (faster, more accurate, less injury
  etc.)
• Used for extended periods of time (maybe all day
  long)
• Requires no (or very little) hand/eye time
• Is as small as possible
• Becomes Invisible (forget that the device is
  there)

4
some examples

• Currently
• Battlefield Translation
• Inspection Insurance, Q/C
• Logistics Distribution Centers
• Consumer GPS route computers
• Very Near Future
• Retail Extend Distribution to the Sales Clerk
• Consumer Organize lists and errands
• Industry Process Control

5
voice in logistics

• Distribution Centers
• The way we move the vast majority of products
  from manufacturer to consumer
• Moving from many homogeneous collections to many
  heterogeneous collections
• Many Suppliers (send product TO the Center)
• Many Stores (receive product FROM the Center)
• A massive repackaging task
• A sizable fraction of the cost retail products
• One of the biggest sectors of wearable voice
  agents to date

6
voice in logistics
7
a Selector talking to Jennifer

• The agent tells the human
• where to go
• what to select
• how many to select
• where to put the item(s)
• The human tells the agent
• Location checkstring
• Quantity selected
• If the bin is empty

8
a Selector Agent in the Refrigerator.
9
some things that just happened

• Selector was directed to product
• Location was verified
• Some product had unique weights entered
• Others had expiration dates to verify
• Selector needed the agent to repeat
• Selector was lifting (80 lbs), walking, driving,
  reading, etc. while talking

10
fast interaction

• Overlapped dialog
• Look ahead
• Independent use
• Eyes
• Hands
• Speech
• Natural corrections
• Low cognitive load

11
accommodation

• Linguistics
• Finishing each others sentences
• The classic barge-in
• The never (but maybe soon) seen interruption
• Expectation
• Predicting dialog flow
• When the response is marginal but expected
• Response is legal but how probable?

12
accommodation example agent side

• In conventional voice applications the prompts
  need to be clear and unambiguous.
• But for an agent co-worker this would be
  tedious.
• In the beginning a natural prompt speed is best
  for learning the routine.
• Later, however, natural will feel like
  slow-mo and must be snappier.
• Later still, the human and agent know each other
  well and just cut to the chase further shortening
  the prompt.

13
components of a voice agent

• Small device
• Light weight, long battery life, rugged
• Speech Technologies
• Recognition,Text-to-Speech and recorded waves
• Multi-Modal fits in here too.
• Dialog Management
• A core system that controls the goals of the
  interaction
• Connectivity
• The real work usually involves information
  external to the agent

14
simple view of a generic voice platform

• Most PDA-like platforms run some version of
  Windows CE or Windows Mobile
• They need full-duplex GOOD quality audio IO
• Enough cycles to do the ASR and TTS
• Low level control over power management

15
a more complicated view
16
some industrial hardware platforms
devicesSmall_3.JPG
devicesSmall_3.JPG
17
did I mention they have to be tough
18
would regular folks talk with a computer?

• Obviously Hands and Eyes free
• Grocery shopping
• Assembling a childs toy
• Cooking a new recipe
• We think differently (freely? Innovatively?) when
  we talk
• Talking is a low (perceived) cognitive load
• People get writers block more often than
  talkers block
• To off load and manage the fussy details of our
  lives

19
Futures

• The latest cell phones have the power to support
  a voice-based agent.
• They cost 1/10th of a present day industrial
  device
• It is just a matter of time before we talk TO our
  phone as well as ON it.

20
Summary

• Wearable voice agents
• Have been here for a while
• Proven and make good business sense
• Declining in cost
• Expanding the range of worker multi-tasking
• Can be effortless to use

21
Conclusions

• They are more places than you think
• They are REAL TOOLS not window dressing
• They are just in their infancy
• I am looking forward to my next new synthetic
  agent!

22

• Thank you!
• Contact
• Emmett Coin
• Director of Speech Research and Development
• coin_at_lucasware.com
• 724 940 7041
• www.lucasware.com