摘要
传统的语音处理技术忽视了语音中可能存在的非线性结构,因而限制了处理能力的进一步提高,通过语音预测模型的基于最小均方误差的自适应预测分析,讨论了数据集平均对LMS算法瞬态特征的影响和步长对均方误差的影响。通过LMS滤波器的过渡特性及其学习曲线的实验仿真,证实了小步长理论的结果与实验结果之间的紧密一致。
Traditional models for production of speech neglects nonlinear structure known to be presented in the speech signals, and this manifests itself an inferior ability to discriminate speech sounds. In this work, through the adaptive prediction analysis based on the least mean square error for speech prediction model ,the influence of data serial average on LMS arithmetic＇s transient feature is discussed and analyzed,as well as influence of step on lean mean square error. Besides,transfer character of I.MS filters and learning curves from simulation that manifest small step theory laboratorial results is sufficient identical with small- step theory.
出处
《现代电子技术》
2007年第17期4-5,13,共3页
Modern Electronics Technique
基金
教育部回国人员科研启动基金资助([2004]527)
