嵌入式语音系统信息采集算法研究

Research on Information Extraction Algorithm for Embedded Speech System

针对嵌入式语音系统的特点,把信息采集算法创造性地结合并应用在嵌入式语音系统的语音信号的采集中。结合了Baum-Welch算法、SA-HMM2算法和改进的Viterbi算法。信息采集主要有四个阶段:预处理、模型优化参数的获取、模型的训练以及信息的采集。首先是对输入的模拟信号进行预处理,滤波、语音格式和标准化等等;接下来通过BaumWelch算法确定新的模型优化参数;然后,再通过SA-HMM2算法进行训练,获取全局的最优参数模型;在最后通过改进的Viterbi算法得出最佳状态序列。

According to the characteristics of embedded speech system,the information extraction algorithm is applied to embedded speech system creatively.Combined with Baum-Welch algorithm,SA-HMM2 algorithm and improved Viterbi algorithm.There are four stages of information acquisition,the acquisition of pre process,the acquisition of the state transfer sequence,the training of the model and the collection of information.Firstly,the input analog signal is pre processed,filtered or the speech format is normalized.Then the VIPS is used to get the status of HMM2.