Investigation of hierarchical classification for simultaneous gender and age recognition

Abstract:

For a successful speech-controled human-machine-interaction individualized models are needed. If the system is designed to run with many users for short times each, a complete user adaptation is not useful. A possible solution would be to use user-group pre-adapted models and recognize the group the actual speaker belongs to in the very first beginning of the interaction. In this paper we present investigate different methods to recognize age and gender groups with an hierarchical model to improve the recognition rate. We could prove, that our method could get adequate results on a four class problem compared with classical approaches.


Year: 2012
In session: Spracherkennung
Pages: 58 to 64