论文部分内容阅读
针对基于人工神经网络的说话人辨认系统所存在的问题,提出了一种基于模糊最小二乘支持向量机(LS-SVMs)技术的两级分类说话人辨认系统。第一级对说话人进行大类预分,第二级则在大类范围内实现具体说话人的辨认。当有新的说话人加入时,只需要增加与新说话人相关的若干个二级分类器。系统仿真实验表明,在训练样本时长取3-11s,说话人由32人增至36人时,本文方法实现的系统训练时间明显降低,识别率更高。
Aiming at the existing problems of speaker identification system based on artificial neural network, a two-level classification speaker recognition system based on LS-SVMs is proposed. The first level carries out the major category pre-division for the speaker and the second level realizes the identification of the specific speaker within the broad category. When a new speaker joins, only a few second-level classifiers associated with the new speaker need to be added. The system simulation results show that when the length of the training sample is 3-11s and the number of speakers is increased from 32 to 36, the system training time realized by this method is obviously reduced and the recognition rate is higher.