摘要： 提出了一种基于最小分类错误准则的概率神经网络的训练算法。实验结果表明, 该系统及其MCE学习算法在20个说话人辨认应用中利用5s清晰语音获得98.9%的辨认率，利用15s电话语音获得86.2%的辨认率。

关键词: 说话人辨认, 概率神经网络, 最小分类错误

Abstract: This paper presents an efficient training algorithm for probabilistic neural networks using the minimum classification error criterion. Experimental results for speaker identification indicate that the system and its MCE training algorithm attain 98.9% identification accuracy using 5 second clean speech utterances and 86.2% accuracy using 15 second telephone speech utterances with a 20 speaker population.

中图分类号: 

• TN912.3