论文部分内容阅读
传统声纹变化识别方法对初始值敏感、在迭代过程中易陷入局部最优、对声纹变化识别的准确率低、而且算法复杂。提出一种外加激光源扫描的声纹变化识别方法研究,首先构建三维激光源扫描定位系统的几何模型,对声纹变化区域进行精确定位和噪声滤除;基于人工鱼群与模糊动态均值聚类相结合的算法,获取聚类数目和中心位置,从声纹信号中提取未知特征的矢量集,实现对声纹变化的识别。实验证明提出的方法,能够有效地解决传统算法易陷入局部最优的问题,节省训练时间、提高声纹变化识别的准确率。
The traditional voiceprint change recognition method is sensitive to the initial value, easy to fall into the local optimum during the iterative process, and has low accuracy in recognizing the change of the voiceprint, and the algorithm is complex. In this paper, a method of voiceprint variation recognition based on laser source scanning is proposed. Firstly, the geometric model of 3D scanning laser locating system is constructed, and the exact location and noise filtering of the voiceprint variation area are established. Based on the artificial fish population and fuzzy dynamic mean clustering A combination of algorithms to obtain the number of clusters and the center position, from the vocal signal extracted from the unknown vector set of features to achieve changes in voiceprint recognition. Experiments prove that the proposed method can effectively solve the problem of traditional algorithm easy to fall into the local optimum, save training time, and improve the accuracy of voiceprint change recognition.