Abstract: According to a brand-new biomimetics pattern recognition theory, a new method realized with neural network to fulfill speaker recognition is presented, which takes advantage of the complicated coverage formed by high-order neural network to construct area of different speakers in feature space to recognize. Relevant experiments show that, this new method can achieve higher recognition rate than traditional methods by training of small-number samples.

Key words: Biomimetics pattern recognition; Speaker recognition; High-order neural network

摘要： 根据一种全新的仿生模式识别理论，提出了采用神经网络实现并完成说话人识别的新方法。该方法利用高阶神经网络形成的复杂包络在特征空间中构造不同说话人的覆盖区域达到识别目的。相关实验证明，这种新的说话人识别方法只要通过少量样本的训练即可达到比传统方法更高的识别率。

关键词: 仿生模式识别；说话人识别；高阶神经网络