神经网络在语音音调识别中的应用研究

文章提出了一种基于神经网络的二级语音音调识别方法，采用常用的多层感知器结构BP反向传播算法．对非特定人的音调进行建模及识别。通过对网络输入矢量进行均值差处理，改善了非特定识别的效率，并且加速了网络训练的进程。根据BP网络得出的一级识别结果，再以音长超音段信息作为辅助条件进行二级识别，加速了网络学习进程．提高了识别率。

Research on the Artificial Neural Network in Chinese Tone Recognition

In this paper, the method of two-step Chinese tone recognition based on neural network is proposed. It adapts BP algorithm to model and recognize speaker-independent tone. The input vector is processed by means of mean difference to improve speaker-independent tone and speed up NN training. The outputs of BP is processed by the length of tone, so the recognition rate is improved.