Important Events

User login

Feeds

Syndicate content

Contact details

Contact WCL

Tel: +30 2610 996496
+30 2610 996480
Fax: +30 2610 997336
+30 2610 991855
Email: raniadou [AT] upatras.gr
 
 
Electrical and Computer Enginnering Department
 
University of Patras
 

Demo: Emotion recognition from speech

Voice portals, Smart-home systems, Call routers, Access control applicationsOne research field of the Artificial Intelligence Group is emotion recognition from speech signal. Our work deals with the modeling and the detection of humans’ affective states, focused on the recognition of emotions on real-world data. The integration of an emotion recognition component in a variety of applications:

  • Voice portals
  • Smart-home systems
  • Call routers
  • Access control applications

can create user-friendly interfaces and provide the means for more successful interaction experiences.

The emotion models built utilize the SOCRATES real-world emotional speech corpus. The speech data consist of recordings collected during the interaction of naïve users with a smart-home dialogue system. Annotation of the speech data with respect to the uttered command and emotional state was performed.
Test site of the smart-home dialogue system

Test site of the smart-home dialogue system

The experimental setup and methodology that we followed corresponded to a typical operational scenario of the smart-home dialogue system deployed at our test-site. Specifically, the smart-home system provides, via speech interaction, access to information, entertainment devices, and control of intelligent appliances.

Emotion recognition component

Emotion recognition component

The Emotion Recognition Component relies on the state-of-the-art audio parameterization technique and utilizes Gaussian Mixture Models for training emotion models.

Currently. recognition results for negative emotions confirm previous research and highlight  the difficulties when dealing with genuine formulations.

DET plots for the detector of negative emotional states for speaker dependent (dashed line) and speaker independent emotion recognition

 DET plots for the detector of negative emotional states for speaker dependent (dashed line)
and speaker independent emotion recognition
(JavaApplet in action)

For more information please contact: Nikos Fakotakis or Theodoros Kostoulas
Web based demo created by: Charalampos Tsimpouris

References
[1] T. Kostoulas, T. Ganchev, I. Mporas, N. Fakotakis, "A Real-World Emotional Speech Corpus for Modern Greek", LREC’2008, Morocco, May 28-30, 2008.
[2] T. Kostoulas, T. Ganchev, A. Lazaridis, N. Fakotakis, "Enhancing Emotion Recognition from Speech through Feature Selection", TSD 2010, Lecture Notes in Computer Science, Springer Berlin / Heidelberg, (accepted).