摘要
高斯混合模型(GMM)已广泛地应用于文本无关的说话人识别系统,该方法具有简单高效的特点。但如果GMM模型的高斯混合分量的数目比较多时,整个模型运算的复杂度会比较大。针对这个问题,提出将聚类算法和传统的高斯混合建模结合起来从而优化高斯混合模型,能够有效地提高说话人识别的速度。实验结果验证了这种算法的高效性。
Ganssian mixture model （GMM） has been widely used for text- independent speaker recognition. This method has simple and efficient character. However,if it has a large number of Gaussians in GMM, it leads to a large complexity of computation. To solve this problem, proposes a new method which combines classical GMM with clustering algorithm to optimize the GMM for reducing the complexity of computation. Experimental results demonstrated that our approach was quite efficient to reduce the complexity of computation.
出处
《计算机技术与发展》
2009年第4期35-37,40,共4页
Computer Technology and Development
基金
"985工程"二期"信息技术"创新平台资助项目(0000-X07204)
关键词
说话人识别
高斯混合模型
聚类算法
speaker recognition
Gaussian mixture model
clustering algorithm
