1 / 13

human language technologies

human language technologies. hlt. Data Collections & Studies WP4 - Emotions: Faces. Collection and annotation of audio-visual databases. extensive data collection, both at KTH and ISTC/IRST using opto-electronic systems reflective markers placed on the subject’s face

nemo
Download Presentation

human language technologies

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. human language technologies hlt Data Collections & Studies WP4- Emotions: Faces

  2. Collection and annotation ofaudio-visual databases • extensive data collection, both at KTH and ISTC/IRST • using opto-electronic systems • reflective markers placed on the subject’s face • capturing of dynamics of emotional facial expressions with very high precision. • eliciting technique: using movies to elicit facial expressions denoting emotions on watching subjects attempted – not promising • extraction technique: extract expressive behaviour directly from movies and television talk-shows  attempted – not promising

  3. KTH - first year DATABASE 1 • 15 emotions and attitudes were recorded (acted) anger, fear, surprise, sadness, disgust, happiness, worry, satisfaction, insecurity, confidence, questioning, encouragment, doubt, confirmation and neutral • Semantically neutral utterances, 9 utterances per expression DATABASE 2 • 6 emotional states confident, confirming, questioning, insecure, happy, neutral • VCV & VCCV nonsense words • CVC nonsense words • Short sentences • Common ITA-SWE set (abba, adda, alla, anna, avva) DATABASE 3 • Spontaneous dialogue

  4. Eliciting technique: information seeking scenario Focus on the speaker who has the role of information giver The speaker whose facial and head motion is to be recorded seats facing 4 infrared cameras, a digital video-camera,a microphone and his/her interlocutor. Collection of audio-visual databases: interactive dialogues (KTH)

  5. ISTC & IRST - first year • 6 emotional states (Ekman’s set) + Neutral • Anger, Disgust, Fear, Happiness, Sadness, Surprise • 3 intensities (Low, Medium, High) • “isolated” emotional expressions • VCV nonsense words (aba, ada, aLA, adZa, ala, ana, ava) • good phonetic coverage of Italian • Long sentence (“Il fabbro lavora con forza usando il martello e la tenaglia” – “the smith works with strength using the hammer and the pincer”) • common ITA-SWE set(VCCV nonsense words: abba, adda, alla, anna, avva) • “concatenated” emotional expressions • VCV nonsense words, in pairs, with different emotions • e.g. (aba)Neutral – (aba)Happy

  6. Results • ISTC/IRST: 1573 recordings • 798 single emotional expressions (7 emotional states, 3 intensities – L, M, H) • 672 concatenated emotional expressions (in pairs, 3 emotional states - Anger, Happy, Neutral - medium intensity) • 57 long sentences (7 emotional states, 3 intensities) • 46 instances of the common ITA-SWE set (3 emotional states, medium intensity) • KTH: 1749 recordings (database 2) • 828 VCV words (138 x 6 emotional states) • 246 CVC words (41 x 6 emotional states) • 645 sentences (270 neutral + 75 x 5 emotional states) • 30 instances of the common ITA-SWE set Total: 3322 recordings

  7. Qualisys recordings: Swedish db – 2nd year • 75 sentences with Ekman’s 6 basic emotions + neutral • Dialogues to analyze communicative facial expressions: • 10 short dialogues in a travel agency scenario • 15 sentences uttered with a focussed word, with the 6 expressions used in corpus 2 + angerExample:Båtenseglade förbi Båten seglade förbi Båten seglade förbi

  8. See Poster!!! Audio-Visual Italian Database – 2nd year (IRST) : A database of human facial expressions (1) • Short videos containing acted kinetic facial expressions (video length: 4-27 secs.) • 8 professional actors (4 male and 4 female). • Each actor played Ekman’s set of six emotions (happy, sad, angry, surprised, disgusted, afraid) + neutral • Actors were asked to play each emotion on three intensity levels (Low -Medium – High) Total: 1008 short videos (= ~ 2h 50’)

  9. See Poster!!! A database of human facial expressions (2) • Facial expressions recorded in two conditions: • “utterance” condition: actors played emotions while uttering a phonetically rich and visemically balanced sentence. • “non utterance” condition: actors played emotions without pronouncing any sentence. • Both video and audio signals were recorded. • After collecting the corpus: Data Selection • Validation of the emotions played • Video selection based on the accordance among judges. “In quella piccola stanza vuota c’era pero’ soltanto una sveglia”, <FEAR>, <HIGH> <DISGUST>, <HIGH>

  10. Annotation of audio-visual databases: interactive dialogues ANVIL: tool for the analysis of digitized audio-visual data • Orthographic transcription of the dialogue • Annotation of the facial expressions related to emotions and of the communicative gestures (turn-taking, feedback and so on) • The annotation is performed on a freely definable multi-layered annotation scheme, created ad hoc for the specific purposes. • These levels go from a less detailed to a more detailed analysis • Annotation is performed on several main tracks, which are displayed, on the screen in alignment with the video and audio data

  11. Annotation (cont’d) glad

  12. Evaluation Studies (IRST) • Experiment 1: Comparison of emotion recognition rates from natural (actor) videos with different types of synthetic (synthetic face) videos, in different animation conditions [reference person: Fabio Pianesi – pianesi@itc.it] • Experiment 2: Cross-cultural comparison of emotion recognition rates from Italian and Swedish natural and synthetic videos [reference person: Fabio Pianesi – pianesi@itc.it] • Experiment 3: as for Experiment 1 but using • three regions of the face • only one animation condition (script based) [reference person: Michela Prete – prete@itc.it]

  13. Papers on Evaluation Studies • J. Beskow, L. Cerrato, P. Cosi, E. Costantini, M. Nordstrand, F. Pianesi, M. Prete, G. Svanfeldt, "Preliminary Cross-cultural Evaluation of Expressiveness in Synthetic Faces". In E. André, L. Dybkiaer, W. Minker, P. Heisterkamp (eds.) "Affective Dialogue Systems", ADS '04, Springer Verlag. Berlin, 2004. • E. Costantini, F. Pianesi, P. Cosi, "Evaluation of Synthetic Faces: Human Recognition of Emotional Facial Displays ". In E. André, L. Dybkiaer, W. Minker, P. Heisterkamp (eds.) "Affective Dialogue Systems". Springer Verlag, Berlin, 2004 • E. Costantini, F. Pianesi, M. Prete "Recognising Emotions in Human and Synthetic Faces: The Role of the Upper and Lower Parts of the Face". To appear in Proceedings of IUI 2005: International Conference on Intelligent User Interfaces. San Diego, California, 2005.

More Related