2018: Technology in Labour for Rebirth of HAL-9000 series Some contributions of ETRO
May 11, 2015
2018: Technology in Labour for Rebirth of HAL-9000 series
Some contributions of ETRO
2
The HAL-9000 series (1968)
2001: A Space Odyssey - Stanley Kubrick, Arthur C. Clarke
The spaceship discovery
HAL-9000
3
HAL-9000 Ambient Intelligence System (1968)
HAL-9000
Monitors its surroundings, the ship and its crew Analyses sensors, images and soundsConverses in natural spoken languageDesigned to assist the user
4
AMI-9000 Ambient Intelligence System (2018)
AMI-9000
Monitors its surroundings, the ship and its crew Analyses sensors, images and soundsConverses in natural spoken languageDesigned to assist the user
5
AMI-9000
AMI-9000 Ambient Intelligence System (2018)
Monitors its surroundings, the ship and its crew Analyses sensors, images and soundsConverses in natural spoken languageDesigned to assist the userIntelligence: expression/emotion/body language awareness
6
Current research @ ETRO-DSSP
Speech modificationtime scaling, intelligibility enhancement, automatic lip synchronization, voice conversion
Speech synthesisFlemish TTS, hierarchic TTS, AV TTS, expressive speech synthesis (emotions)
Noise suppression (single channel, contact microphones)robust speech recognition
Microphone array techniquesdistant recording, blind source separation
Speech disordersdiagnostics, remedy
7That’s Right
AIBOSBA
Emotion Recognition
Emotions in speechalter pitch, timing, voice quality and articulation.
Emotional speech rec.classify statistical measures of acoustic features into classes.
Two approachesSegment Based (SBA) & AIBO (Sony)
8
X-database study of emotion recognition
Obtained state-of-the art recognition scoresEmotion recognition = database dependentClassifiers can be learned on joined databaseUse of higher level features might helpUse arousal detection as case study
24%46%42%35%Baseline X-DB
23%53%45%54%X-DB
51%34%42%32%Baseline
54% H:67%Human 85%67%82%Literature
64%75%69%87%Best score
DanishBerlinBabyEarsKismet
Go to Demo
9
Current research @ ETRO-IRIS
USER NATURAL INTERACTION
-motion-speech
-expression
USER INTERACTION SYNTHESIS
- estimating the facial animation parameters
- face model adaptation
- mouth visualization
- emotion feature extraction
- audiovisual speech segmentation
- morphing a 3D head
- data-driven feedback
- animating an avatar
- detection & tracking of body and face
USER INTERACTION ANALYSIS
10
Facial Analysis & Synthesis
GesturesMotion estimationPose and structure variationsEye gaze and expressions
Multi-modal speechEnhancement by mouth images Animated avatar
and takes into account the natural face motions
11
Expression Analysis
Facial Action Coding Systema muscle-based method to measure facial movements w.r.t. Facial anatomy , widely used in Psychology
Each Action Units (AU) represents one visibly distinguishable facial change (46 AUs for facial appearances, 12 AUs for gaze direction and head pose).
Face expression = Co-occurrence of several AUsA parametric model combining several AU’shas been built for expression analysis
What is hidden behind a face expression?the temporal course of the muscle activities (intensity of muscle contraction/relaxation versus time). Information to recognise concealed emotions (e.g. deception). AUs are purely descriptive. FACS provides a dictionary to interpret the corresponding emotions
recognition/synthesis
Visual input
Face processing unit
Expression processing unit
AngerSurprise
12
AMI-9000 Ambient Intelligence System (2018)
Smart buildings (surveillance)Care for the elderly at home (assisting, security)Personalised personal assistants (understand the user)VIN: adapt according to the user’s state-of-mindEmbodied conversational agents for education