基于FCM和FSVM的说话人辨认

针对经典支持向量机对非目标样本没有拒绝能力,且应用于说话人辨认时存在不可分区域的问题,提出一种基于模糊C均值聚类和模糊支持向量机的多级模糊说话人辨认方法。首先利用模糊C均值聚类方法对特征向量进行聚类,减少样本的数目,加快模糊支持向量机训练速度。最终由FSVM得出判决结果。并通过仿真实验验证了该方法的有效性。

Speaker Identification Based on FCM and FSVM

For the sake of solving the problem of conventional SVM which has no ability to reject non-target sample, and unclassifiable audio data exists when the conventional SVM was utilized to make classification in the speaker identification simultaneously. A novel hierarchical fuzzy speaker identification method based on fuzzy c-means (FCM) clustering and fuzzy support vector machine (FSVM) was proposed in this paper. Firstly, the FCM clustering technique is utilized to partition the whole training dataset into several clusters which has its own cluster center. And then, FSVM is trained by the cluster centers to make final decision and process the unclassifiable data. Experiment results show that the proposed method heightens identification accuracy of system remarkablely compared with the baseline SVM speaker identification system.