Combine multiple time-delay HMEs for speaker identification

In this paper, we present a scheme of combining multiple time-delay nets with the structure of the hierarchical mixture of experts (HME) in order to complete a multi-scale analysis for the temporal data. This method extends the technique of combining multiple classifiers based upon the same static input to sequence processing. We have already applied the proposed method to a real-world problem, called text-dependent speaker identification. Based upon the experiments, we demonstrate that the speaker identification system based upon the combination of multiple time-delay HMEs achieves better performance than the one based upon an individual time-delay HMEs.