Based on the initial clusters center generated by the LBG in every speakers feature space,every sample can be classified,so every clusters area can be parallel covered by the PONN.

• 该方法以LBG算法生成每个说话人特征空间初始的聚类中心,对本类样本按聚类中心分类后,用前馈优先级神经网络(PONN)对每个聚类区进行并行覆盖。