Tom Freeman
Sr. Vice President & Co- Founder
VoiceBox Technologies, Inc.
Copyright © 2001-2006 VoiceBox Technologies, Inc
The Four Elements ofConversational Speech
The Problem
• When speech interfaces actually work, they require learning on the part of the user
• Users ‘dumb down’ requests to the ‘irreducible minimum’
The Promise
• Immediate accessibility for ALL systems
• Never have to learn a system again
• Human-Machine interactions that feel ‘normal’
The Risk
• Without a clear definition of “Conversational”
The word ‘Conversational’ will be subject to:
• Exploitation• Disillusionment• Ultimately consumer cynicism
Conversational Speech
Four Elements– Casual Speech– Noise Tolerance– Intent Determination & Hypothesis Building– Cooperative Conversations™
Casual Speech
Features•Free Form Utterance
– Typical day-to-day language– Tolerant of infinitely variable human Emotional states, distractions and
serendipity
•One Step Access and Directed Dialogue– Complex and multi-variant queries
•Intent Inference– Completion from incomplete utterance
•Recognition of Expression Variants– Widely flexible action words
•Start, Stop, Restart, Stutter & Run-ons– Verbalized pauses and restarts
Noise Tolerance
• Meaningless words – Out of vocabulary words
• Removal on non-human noise
• Out-of-context words
Intent Determination &Hypothesis Building
• Inference of Intended operations from last
Command
• Multiple Words in Combination help reveal
Context– What about Traffic?
80s Rock and Roll Band
Direction to O’Hare
Review of Michael Douglas Movies
Cooperative Conversation™
• Creates a holistic Voice User Experience (VUE™)• Delivers on the promise of Conversational Language processing• A perceived successful conversation with a machine
Regardless of:length of timenumber of steps
Cooperative Conversation™
The parts:
1) Shared Knowledgea) Long Term
b) Short Term
2) Hypothesis Building / Intent Determination
3) Adaptive Responses
1) Shared Knowledge
Short Term – ‘session’ knowledge
current & recent recognition events
Long Term – ‘user-centric’ knowledge
topic knowledge
conversational history
•What are we talking about?
•Course Correction
•Build on what we know
2) Intent Determination &Hypothesis Building
3) Adaptive Responses
Understanding
+
Certainty
= Great User Experience!
Examples
“What's mumble the Admiral Twin Theater?”
Reply: "The movie schedule for the Admiral Twin Theater? Here you go ..."
Reply: "Did you want a movie schedule for Admiral Twin Theater?"
“What's mumble the Admiral Twin Theater?”
“What movies are at the Admiral Twin Theater?”
Reply: “Here’s what’s playing at the Admiral Twin"