中文说明：
文章展示了基于高斯混合模型的语音频谱预测方法。频谱预测可能在传包过程中预防丢包这方面起到大作用。期望最大化算法用两倍或三倍的连续语音因素来测试模型。模型被用来设计第一，儿等指令预测量。预测表用频谱分配状态来估计并和一个简单的参考模型对比。最好的预测表得到一个平均频率扭曲值是0.46dB小于参考模型
English Description：
This paper presents a speech spectrum prediction method based on Gaussian mixture model. Spectrum prediction may play an important role in packet loss prevention. The expectation maximization algorithm tests the model with two or three consecutive speech factors. The model is used to design the first, second and other instruction predictors. The prediction table is estimated by spectrum allocation state and compared with a simple reference model. The best prediction table can get an average frequency distortion value of 0.46db, which is smaller than the reference model < br / >