Personality, likability, and pathology are important speaker traits that convey rich information beyond the actual language. They have promising applications in human-machine interaction, health informatics, and surveillance. However, they are less researched than other paralinguistics phenomena such as emotion, age and gender. In this paper we propose a novel feature selection approach for speaker trait classification from a large number of acoustic features. It combines Fisher Information Metric feature filtering and Genetic Algorithm based feature selection, and fuses several elementary Support Vector Machines with different feature subsets to achieve robust classification performance. Experiments on an INTERSPEECH 2012 Speaker Trait Challenge dataset show that our approach outperforms both baseline approaches.
Index Terms: Paralinguistics, Speaker Trait Classification, Personality, Likability, Pathology, Genetic algorithm, Fisher Information Metric, SVM
Bibliographic reference. Wu, Dongrui (2012): "Genetic algorithm based feature selection for speaker trait classification", In INTERSPEECH-2012, 294-297.