论文部分内容阅读
支持向量数据描述(SVDD)将多类样本数据每一类用各自的超球来界定,显著降低了二次规划计算复杂度,更易于解决多类分类问题,因此在语音识别研究领域越来越受到广泛关注,本文针对语音样本分类中特征向量重叠和更新等问题,对现有的SVDD多类分类算法进行了改进,一方面,根据样本所在空间位置,构造超球重叠域决策函数;另一方面,基于类增量学习,实现超球类支持向量的动态改变。仿真实验结果表明,本文所提方法明显缩短了建模时间并且具有更好的识别性能。
Support Vector Data Description (SVDD) defines each class of multi-class sample data by its own hypersphere, which significantly reduces the computational complexity of quadratic programming and makes it easier to solve multi-class classification problems. Therefore, in the field of speech recognition, In this paper, we focus on the problem of overlapping and updating eigenvectors in speech sample classification, and improve the existing SVDD algorithm. On the one hand, according to the spatial location of samples, we construct the super-sphere overlapping domain decision-making function; In the aspect, based on class incremental learning, the dynamic change of hypersphere class support vector is realized. The simulation results show that the proposed method can shorten the modeling time significantly and has better recognition performance.