Advances in Anger Detection with Real Life Data

We report on the progress with respect to an emotion-aware voice portal concerning several directions. A comprehensive new data collection has been carried out and gives new insight on the nature of real life data. The labeling process and the structure of the data will be discussed. Experiments with the anger detector on that data indicate that the acoustic features based on voicing don’t play an important role for this data. It shows once again, that classification problems are highly data specific as this data contains a lot of noise. Also using Support Vector Machines instead of Gaussian Mixture Models as classification algorithm turned out to give better results with respect to anger recall. As a consequence for these findings we plan for a new classifier that’s more flexible with respect to choice of features and algorithm in order to adapt for different data.