A General Architecture for Self-Adaptive AmI Components Applied in - - PowerPoint PPT Presentation

a general architecture for self adaptive ami components
SMART_READER_LITE
LIVE PREVIEW

A General Architecture for Self-Adaptive AmI Components Applied in - - PowerPoint PPT Presentation

AG Softwarearchitecture Softwarearchitecture AG A General Architecture for Self-Adaptive AmI Components Applied in Speech Recognition Holger Klus ICSE 21.05.2006 21.05.2006 AG Softwarearchitektur 1 Outline Outline Outline


slide-1
SLIDE 1

21.05.2006 AG Softwarearchitektur 1

AG AG Softwarearchitecture Softwarearchitecture

ICSE 21.05.2006

A General Architecture for Self-Adaptive AmI Components Applied in Speech Recognition

Holger Klus

slide-2
SLIDE 2

21.05.2006 AG Softwarearchitektur 2

Outline Outline

  • Outline

– Application Scenario – Our three types of adaptation – Integration of speech recognition into our AmI- scenario as one type of adaptation – Outlook

slide-3
SLIDE 3

21.05.2006 AG Softwarearchitektur 3

Track Section Uphill Even Downhill

Motivation Motivation Example Example – – Assisted Assisted Training (1) Training (1)

  • Training group of (racing) cyclists
  • Given track profile
  • Cyclists have different qualities regarding the track sections
  • Each cyclist has a given overall physical condition and an individual training plan
  • Training effect depends on the speed of the group and the position of the cyclist

within the group formation

Main goal of the system

  • Optimization of the single training effect
  • Optimization of the group training effect
slide-4
SLIDE 4

21.05.2006 AG Softwarearchitektur 4

Motivation Motivation Example Example – – Assisted Assisted Training (2) Training (2)

pedal power sensor Bluetooth / WLAN GPS GSM AmI PDA wind sensor MICA Nodes pulse sensor trainer Internet-based training database

slide-5
SLIDE 5

21.05.2006 AG Softwarearchitektur 5

Motivation Motivation Example Example – – Assisted Assisted Training (3) Training (3)

Mobile Phone Display Application Pulse Control Headphone LCD Display

set limits show pulse and warning (text) monitor pulse get current pulse

Pulse Sensor

(Currently) explored adaptation variants

  • Service Implementation Adaptation

Self-adaptation of speech recognition service

  • Service Usage Adaptation

Change from a visual UI to a speech generation UI service implementation

  • Service Configuration Adaptation

New component containing new services plugs into the training system and results in a reconfiguration of the overall system from single bicycle to group bicycle training

Application Group Control

group training

slide-6
SLIDE 6

21.05.2006 AG Softwarearchitektur 6

Service Service Implementation Implementation Adaptation Adaptation – – Speech Speech Recognition Recognition

  • Speech Recognition in the AmI-Prototype

– Why

» Natural communication with the system » But important: High recognition rate in changing environmental conditions such as

Different speaker (young, old, female, male) Background noise Mental and physical stress of the speaker

– Our goal

» Input of upper and lower pulse boundary using speech input » Take into account the physical stress of the user

– Realization in cooperation with the BUTE (Budapest University of Technology and Economics)

» Using Hungarian speech recognition engine

Task: Mapping on German speech

» Take into account current stress of the speaker using German speech recordings » Feature: Speech recognizer is runnable on „small“ devices like

PDA, Handy, …

slide-7
SLIDE 7

21.05.2006 AG Softwarearchitektur 7

Speech Speech Recognition Recognition

Speech Signal Matching Algorithm Recognized Phrase Reference Model

slide-8
SLIDE 8

21.05.2006 AG Softwarearchitektur 8

Speech Speech Recognition Recognition under under Physical Physical Stress of Stress of the the Speaker Speaker

  • Generic Approach

– Stress Recognition

» Current research topic

Automatic stress detection

» In our scenario

Stress detection using the pulse value of the speaker

– Compensation of stress features

» Manipulation of speech-statistics at runtime

High computational power required

» Switching between different speech-statistics which have been created offline

slide-9
SLIDE 9

21.05.2006 AG Softwarearchitektur 9

Architecture Architecture of a

  • f a Self

Self-

  • Adaptive

Adaptive Component Component

Sensor Network Self-adaptive AmI component Context Providing Context Processing Context Adaptation ContextProvidingInterface ContextAdaptationInterface Actor Network

Stress recognition and stress level Decision of how to react on a specific stress level Switching between statistics Pulse Sensor Self-adaptive speech recognizer

slide-10
SLIDE 10

21.05.2006 AG Softwarearchitektur 10

Outlook Outlook

  • Service configuration adaptation

– First version is already implemented

  • Description techniques for „Quality of

Service“

– Especially required by

» Service Usage adaptation » Service Configuration adaptation

– What is the best service at the moment? – Criterion can change during runtime!

slide-11
SLIDE 11

21.05.2006 AG Softwarearchitektur 11

Questions Questions and and Discussion Discussion

Any Questions ?