HMM2小波变换的参数估计

The parameter estimation of HMM2 based on wavelet transformation

在观测噪声和马尔可夫链不相互独立的条件下提出了一种新方法来解决通过小波变换后的二阶隐马尔可夫模型(second-order HMM:HMM2)中高斯混合模型的参数计算问题.这种方法可以不必根据变换后的数据对系统参数进行重新估计,只需利用变换后输出的小波系数直接计算即可,避免了保留所有训练数据的繁琐复杂计算过程.

Given the condition of the mutual dependence of the observation noise and the Markov chain, a new method to solve the parameter problems of Gaussian Mixture models in second-order Hidden Markov Models（HMM2） is proposed when the input signals are transformed by wavelet. Hereby we do not need to reestimate the system parameters according to the transformed data, whereas we can make a direct calculation with the output transformed wavelet parameters. The new method simplifies the calculation process avoiding keeping all the data.