摘要
提出了一种可用于说话人识别的神经阵列网络，它以仅完成两类模式区分的小型网络作为子网络，再将单个子网络组合成阵列形式来完成多类模式的区分。文中给出了阵列网络的构成及搜索算法，并使用径向基函数（ＲＢＦ）阵列网络进行了文本无关的说话人识别的研究。实验显示，对 ２０名说话人，用 ５秒语音训练， ２秒语音识别时，该方法可达到 ９８％的正确识别率。
A neural array networks for speaker identification is proposed in this paper. It is able to convert the complex problem of N-catalog classification into a set of simple problem of 2-catalog classification. The architecture and searching algorithm of the neural array networks are described. The speaker identification with RBF(radial basis function) array networks is discussed emphatically. Experiments showed that this method can identify speakers well. The speaker identification correctness reaches to 98 % for 20 speakers when trained with 5 second speech utterances and tested with 2 second speech utterances.
出处
《声学学报》
EI
CSCD
北大核心
2000年第3期230-234,共5页
Acta Acustica
关键词
神经阵列网络
文本
说话人识别
语音识别
Algorithms
Arrays
Classification (of information)
Neural networks
Text processing
