论文部分内容阅读
高斯混合模型（ ＧＭＭ）是当今说话人识别的一种流行算法，但 ＧＭＭ的训练的目标是使似然度最大，并不能产生识别性能最佳的模型。本文提出了ＧＭＭ ＋ＭＣＥ（最小分类错误）的模型来解决这一问题。并通过实验证明了其有效性。
The Gaussian Mixture Model (GMM) is a popular algorithm for speaker recognition today, but the goal of GMM training is to maximize the likelihood of likelihood and to not produce the model that best identifies the performance. This paper presents a GMM + MCE (minimum classification error) model to solve this problem. The experiment proved its effectiveness.