一种面向语音识别的新型神经网络

A Speech Recognition oriented New Neural Network

提出了一种新型神经网络模型，描述了该网络的工作原理和训练方法以及识别算法。为克服神经网络对时序信号建模能力差的缺点，引入了非线性分段处理和代表帧特征提取方法。最后介绍了根据这一模型所设计的一个汉语语音识别系统，试验表明该网络在汉语语音识别方面具有较大的潜力。

A new neural network is proposed in this paper.The training and recognition algorithms are described.In order to improve the ability of modeling the temporal signal,a preprocessing approach named nonlinear segmentation and representing frame feature extraction are employed.Based on the new neural network,a Chinese speech recognition system is established and has achieved high performance in experiments.