Towards better capturing inter-speaker information by active learning for speaker identification

In our previous work (2000), we proposed a novel way to capture inter-speaker information with a neural network for speaker identification system. In this paper, an active learning algorithm is presented for improving the generalization ability of the neural network towards better capturing inter-speaker information. Comparative results on the KING speech corpus show that our method leads to a considerable improvement, in particular, for the narrow-band set