View
216
Download
1
Category
Tags:
Preview:
Citation preview
VoiceXML:A Field Evaluation
By: Kristy BradnumSupervisor: Peter Clayton
Presented in partial fulfilment of the CS Honours Project
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Overview Objective of Research Background Aims & Motivation Methodology
Tools Results Conclusions Questions
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Objective of Research
My project in a nutshell:An evaluation of VoiceXML 2.0, using a range
of platforms, looking specifically at it’s maturity as a technology and it’s status as an industry standard.
Objective of Research
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Background
Overview of Speech Technology Overview of VoiceXML
HistoryRole
Background
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Overview of Speech Technology
Conversion between spoken word and binary Output
Previously: pre-recorded prompts Today: speech synthesis (TTS)
Input Previously: DTMF (pressing keys on the phone) Today: speech recognition (ASR)
Background >> Overview of Speech Technology
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
History of VoiceXML
AT&T Bell’s PhoneWeb project Lucent’s Phone Markup Language Motorola’s VoxML IBM’s SpeechML
VoiceXMLversion 2.0full W3C recommendation16 March 2004
Background >> Overview of VoiceXML
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Role of VoiceXMLBackground >> Overview of VoiceXML
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Motivation 2002
Mya Anderson’s Field Investigation of VoiceXML 1.0
New technology Unstable Unsuccessful
Now VoiceXML 2.0 = W3C standard Nortel: maturity increasing, widely accepted Jackson: already mature
Project Aims & Motivation
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Project Aims
Investigate these claims Examine:
maturity of VoiceXML 2.0 as a technology it’s status as an industry standard
Project Aims & Motivation
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Methodology
Approach Tools
WebSphereOptimTalkBeVocal Café
Analysis Cross-Platform Analysis
Methodology
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Approach
Iterative approachSet goalEvaluate outcomesDetermine next goal
ROSS prototypeRelevant to RhodesProduct secondary to investigation
Methodology >> Approach
Inadequate
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Tools
3 approaches [Seth]:BuyRentBuild
3 environments [Beasley et al]:HostedSimulatedWeb-based
Methodology >> Tools
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Project Tools
WebSphere Studio Application Developerwith Voice Toolkit
OptimTalk 0.9.1 BeVocal Café 2.5
Methodology >> Tools
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
WebSphere
IBM’s WebSphere Studio Application Developer with Voice Toolkit plug-in
“Buy” approach Voice Toolkit also includes other features:
CCXML developer NLU model maintenance
Call Flow Builder Grammar developer Pronunciation Builder
but Version problems
Methodology >> Tools >> WebSphere
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
OptimTalk
Simple VoiceXML platform Desktop standalone development environment Set of libraries interpret W3C SIF markup
languages Tailored towards research Command line application Requirements: microphone and speakers
Methodology >> Tools >> OptimTalk
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
BeVocal Cafe
Good background Web-based development environment Hosted platform “Rent” approach Tools:
VoiceXML CheckerVocal Scripter
Methodology >> Tools >> BeVocal Café
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Analysis
Each platform studied separately 15 examples for OptimTalk
Basic + Blackjack
10 projects for BeVocal Café Miller’s “10 Projects to Voice-Enable Your Web Site”
Methodology >> Analysis
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Cross-Platform Analysis
BeVocal Café’s projects in OptimTalk OptimTalk’s examples in the Café Run amendments through original platform
Methodology >> Cross-Platform Analysis
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Results
Platform IndependenceOptimTalkBeVocal Café
Grammars Design Considerations Platform Certification
Results
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Platform Independence
Environments: steep learning curve Extensible tag set
Limits platform independenceProprietary extensionsSome features added – some left out
Example code usually worked
Results >> Platform Independence
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
OptimTalkResults >> Platform Independence >> OptimTalk
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Results >> Platform Independence >> OptimTalk
OptimTalk
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
OptimTalk
Speech recognition component erratic Built-in grammars not supported in OptimTalk
type attribute of <field> Boolean = yes / no grammar Number = ?
No <say-as> Phone numbers (from database)
Results >> Platform Independence >> OptimTalk
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
BeVocal Café
More mature Very confusing error messages Many proprietary extensions
Results >> Platform Independence >> BeVocal Café
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Grammars
Built-in grammars not supported in OptimTalk
OptimTalk’s grammar not supported by BeVocal
W3C passes responsibility to SIFPlatforms should support ABNF of SRGS
Results >> Grammars
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Design Considerations
No long menus Pronunciation
Different voices for TTS ROSS by LH Michael vs Microsoft Sam
Be careful with ASR Ties in with grammar
No break in <enumerate> in OptimTalkSo lists run together
Results >> Design Considerations
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Platform Certification
VoiceXML Forum’s Platform Certification ProgramTest Suite v1.0 ≈ 700 test programsTo check compliancy with VoiceXML 2.03 platforms passed (in September)
NVP VoxPilot Open Media Platform VoiceGenie Platform
Results >> Platform Certification
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Conclusions
Learnt a lot about Speech technology Language does seem fairly mature now
Fewer extensionsMore complete as a standard
Still not quite stable
Conclusions
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Conclusions
Give platforms time to catch up Still hurdles in development
Especially in South Africa Possibly better for commercial enterprises
Not for research But improving all the time
Conclusions
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
Future work
Windows vs Linux comparison
Look at W3C Speech Interface Framework as a whole
VoiceXML 2.1 is on it’s way
Conclusion >> Future Works
------------ VoiceXML: A Field Evaluation ------------ Kristy Bradnum – Computer Science Honours 2004
QuestionsQuestions ???
Recommended