Funded by FCT
Start date: March 1, 2012
End date: February 28, 2015
VOCE shall develop methods and algorithms that enable the online classification of stress from live speech with the goal of providing feedback cues to the speaker in real-time to improve his communication skills. The work will focus on detecting and classifying stress in speech by leveraging advanced signal processing and machine learning techniques, complemented with psychological analysis of different aspects of stress perception.
Starting from the communication model in this figure, we will first develop a methodology to observe speech and annotate it with objective and subjective measures of stress. We will use public speaking scenarios like presentations and talks in academic environment. The gathered data will be used to assess which speech signal features correlate most with stress, and which are most useful for the task of automatic classification of stress in speech. Finally, we shall develop a system for providing bio-feedback to the speaker in real-time through an adequately designed interface, so that the whole system can be used to improve public speaking in a PC and a mobile platform. A further step is the study of the generalisation of the developed algorithms to other oral communication scenarios.