Recognizing Emotional Expression in Game Streams
summary
A multimodal system labels emotional events in game streams using neural facial-expression analysis, transcript sentiment, voice-emotion models and low-level audio features (pitch, loudness), fused by a temporal neural network trained on human annotations. Dataset covers 17 videos, ~2000 annotated events from Unravel and Unravel 2 streams. Performance reaches 70.7% accuracy (comparable to ~68% inter-rater agreement) and 80.4% accuracy for the five most intense events; excels at amusement/excitement, struggles with subtle puzzlement.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:11Contributions
- 1:46Data Set
- 2:15Criteria
- 3:00Classification
- 4:26Automatic Detection
- 5:46Gamma Stream
- 8:57Conclusion
- 9:48Discussion
related talks
59mTechnical Workshop: Sarcastically Speaking: Unlocking Multi-modal Sentiment Analysis With NLP and Facial Expressions
21mEmotion Estimation for Social Robotics
11mAmbient Intelligence and Emotionally Intelligent Devices
32mDecoding Emotions from Art to Memes
34mEnhancing Customer Experiences with Emotional AI,
06mThe Empathetic Car: Exploring Emotion Inference via Driver Behaviour and Traffic Context
43mDeciphering animal emotions using machine learning and vocalisations
23mPrediction of music pairwise preferences from facial expressions