LM算法在神经网络语音识别中的应用

输入语音信号中声音的特征提取和分类识别可以通过多层前馈神经网络大量学习实现,但基于误差反向传播的前馈神经网络(BP神经网络)标准算法收敛速度慢,在训练中效率不高。采用一种快速稳定的Levenberg-Marquardt算法进行语音识别,通过对语音信号的预处理、特征提取和网络结果优化,建立了网络训练样本集,用MATLAB进行了仿真,仿真结果表明,该算法优于传统的BP算法,具有更好的收敛性。

Application of LM Algorithm in Neural Network Speech Recognition

A multi-layer feed forward neural network was designed to achieve feature extraction and recognition from speech signals inputted with much study. However, the normal BP algorithm had slow convergence rate and low efficiency .The paper adopted a fast and stable Levenberg-Marquardt algorithm for speech recognition, preprocessed speech signals ,extracted features, optimized results of network and set in the training sample of neural network. The simulation was carried out by MATLAB.As it turned out, the algorithm is superior to traditional BP algorithms and have better astringency.