Emotion Recognition - Google Science Fair 2011 Video

Akash: Hello everyone. I am Akash Krishnan.
Matt: And I am Matt Fernandez. We would like to thank Google for coordinating and hosting this science fair. Here is a demo of our project in action, and we hope you find it insightful and entertaining.
--Demo Video--
Akash: Our system works by preprocessing useless silence regions out of the signal and then extracting the frequency and energy levels in speech. We then determine these ranges for each emotion. Each emotion has different frequency and energy ranges, and so we can classify an audio signal based on where the audio signal's energy and frequency values fall under.
Akash: In addition to analyzing how one speaks, we also analyze what words are being said. We apply speech recognition to get the words, and then use parts of speech tagging to get the subject, verb, and object regions. We create separate models of frequency and energy ranges for each of these three regions, and keep track of the tone of the speech based on the negative and positive words used.
Matt: The best baseline performance is 40.9%, and after we added our major contributions and novel steps, we get about 59.8% average performance for five different emotions.
Matt: Among many others, emotion recognition can be used for robotics, gaming, and helping autistic children. We would also like to extract features via phonemes and evaluate real-time performance.
Matt: Thank you for watching, and we hope you enjoyed it. If you have any questions, please feel free to email us.

Category:

Education

Tags:

• Emotion
• Recognition
• Google
• Science
• Fair
• 2011
• robot
• valley
• computer
• math
• county
• technology
• smart phones
• talk
• sneak peek
• programs
• speech
• video game
• defence
• electronics