Presentation Topics

1
Presentation Topics

• System Description
• System Components
• Speech Technology
• Positioning Technology
• Weighted Graph Algorithm
• Implementation
• PDCS Model
• Managing Data
• Structuring Code
• Demo
• Perspectives
• Relevant Links

2
System Description

• The Position Dependent Communication System
  (PDCS) is a location-based service incorporating
  both positioning and speech technology.
• The PDCS is a well documented open-source project
• Existing code can easily be expanded
• Relevant design patterns have been implemented
• Unified Modeling Language (UML) and Javadoc have
  been used to document the project

3
User scenario

• The destination is recognized by a speech
  recognizer
• The physical position is tracked by a positioning
  server
• The route (shortest path) is computed
• The route is communicated to the user on a map as
  well as a synthetic speech telling in which
  direction to go

4
System Components
Speech Technology
The following features must be taken into account

• Speaker-dependent versus speaker-independent
  recognition
• Isolated-word versus continuous-speech
  recognition (c.f. grammar)
• Quiet versus noisy environment

5

• Java Speech API (JSAPI)
• IBM Speech for Java (implementation of the JSAPI)
• IBM ViaVoice (speech engine providing both speech
  recognition and synthesis)

The XML based Java Speech Markup Language (JSML)
makes the synthesized speech sound more natural,
while the Java Speech Grammar Format (JSGF) is a
dictation grammar used by the speech recognizer.
6
Positioning Technology

• Wireless Local Area Network (WLAN)
• Ekahau Manager includes calibrator functionality
  (recording signal strengths from access points)
• Ekahau Positioning Engine
• Ekahau Client

7
Weighted Graph Algorithm
Arbitrary relationships among data objects can be
expressed in a weighted graph model. An example
of a directed weighted graph (in the PDCS the
weighted graph is undirected) could be as follows
V 1, 2, 3, 4 E (1, 2), (1, 3), (2, 4),
(3, 2), (4, 3) G (V, E) w(1, 2) 7, w(1, 3)
1, w(2, 4) 5, w(3, 2) 2, w(4, 3) 9
8
PDCS model
Implementation
The PDCS model consists of both a physical and
logical model (c.f. terminology)
9
Managing Data
The PDCS model is directly related to the data in
the database

• The physical model corresponds to the Floor, Room
  and Door tables in the database
• The logical model corresponds to the
  AbstractPosition and Distance tables in the
  database

10
mysqlgt select from Area --------------------
--------------------------------- areaId
floorId roomId xStart yStart xEnd yEnd
------------------------------------------
----------- 1 2 NULL
954 690 1428 1212 2 2
NULL 954 1212 1428 1436 3
2 NULL 1428 1212 1912
1436 4 2 NULL 1912
1212 2400 1436 5 2 NULL
2400 690 2872 1186 6
2 NULL 2400 1186 2666 1436
7 2 NULL 2666 1186 2872
1436 8 2 NULL 2872
1186 3628 1436 9 2 NULL
3628 1186 4138 1436 10
2 NULL 4138 1186 4794 1436
11 2 NULL 4794 690 5038
824 12 2 NULL 4794
824 5266 1186 13 2 NULL
4794 1186 5266 1436 14
2 3.08 954 0 1912 690
15 2 3.09 1912 0 2872
690 16 2 3.10 2872
0 3342 1186 ()
select areaId from Area where roomId
3.09 select areaId from Area where floorId
2 and xStart lt 2000 and xEnd gt 2000 and yStart lt
500 and yEnd gt 500
15 will be returned as the areaId in both cases!
11
Structuring Code

• Dividing the classes into packages makes it
  easier to debug, while each package can be tested
  individually (c.f. each package can be considered
  as a plug-in)

dk.funkenstein.graph dk.funkenstein.service dk.fun
kenstein.position dk.funkenstein.protocol dk.funke
nstein.speech dk.funkenstein.model.physical dk.fun
kenstein.model.logical
12
Demo
PDCS example
In this scenario John has asked the system how to
find Thomas. The system responds with the
shortest path according to Johns specific
position.
13
Examples on speech synthesis

• IBM ViaVoice
• Justin
• Jane
• ATT Natural Voices
• Charles
• Crystal

14
Perspectives

• Loading new packages (grammar, physical and
  logical model) during run-time
• The static data (c.f. destination) could be
  extended with dynamic data (c.f. other users)
• The map (c.f. PDCS model) must be constructed
  automatically instead of manually
• Enrolment of each user
• Scaling the PDCS by making the logical model
  hierarchic
• The experience from designing and implementing
  the PDCS has been used in the first prototype of
  the DELCA (Dis-Embodied Location-specific
  Conversational Agents)

15
(No Transcript)
16
Relevant Links
Position Dependent Communication
System http//www.itu.dk/people/thomasbodin/itu.h
tml http//www.itu.dk/people/thomasbodin/PDCS/dem
o/index.html Speech Technology http//java.sun.
com/products/java-media/speech/ http//www.alphaw
orks.ibm.com/tech/speech http//www-106.ibm.com/d
eveloperworks/ibm/library/i-voice/ http//www.nat
uralvoices.att.com/demos/ Positioning
Technology http//positionserver.itu.dk/ http//
www.ekahau.com/ DELCA (Disembodied
Location-specific Conversational
Agents) http//lacomoco.itu.dk/index.jsp http//
mobilitylab.itu.dk/ http//www.itu.dk/research/de
lca/