Analysis of robustness of attributes selection applied to speech emotion recognition

The paper presents the analysis of the robustness of an attributes selection method applied to speech emotion recognition. The features used were extracted by the front-end ETSI Aurora eXtended of a mobile terminal in compliance with the ETSI ES 202 211 V1.1.1 standard. On the basis of the time trend of these parameters, over 3700 statistical attributes were extracted to characterize semantic units of varying length (sentences, words and generic chunks). Using the WEKA (Waikato Environment for Knowledge Analysis) software the most significant attributes for the classification of two emotional states were selected using the CFSSubsetEval-BestFirst method. The results of classification, obtained using NaiveBayes models, were obtained using intra-corpus and inter-corpora experiments on four different speech corpora performing 4000 trainings and tests. On the basis of these results we can study the robustness of the attributes selection method.