基于动态时间规整和神经网络的方言辨识研究

Dialect identification based on dynamic time warping and neural network

汉语方言辨识技术的研究不仅有利于提高方言语音识别系统的识别效率,而且对于公安部门的刑事侦查等方面都具有非常重要的应用价值。以湖南方言作为研究对象,对不同方言特征的差异及方言辨识中特征参量的合适选取进行了深入研究。针对语音信号具有很强的随机性而神经网络的输入结构相对固定等特点,提出了基于动态时间规整和神经网络的方言辨识方法。实验结果表明,选取相同的特征参数时对不同类别或不同声调的方言的辩识率不同。

The research of Chinese dialect identification is not only conducive to improving the efficiency of dialect speech recognition system,but also important in the criminal investigation department for public security.Hunan dialects have been selected as a research object in this paper.The difference of characteristics between various dialects and how to choose appropriate parameter have been studied thoroughly.Because the speech signal has the very strong randomicity and the input structure of neural network is firm relatively,the dialects identification technology based on a mixed cascade neural networks of time alignment network and BP neural network is proposed.The experimental results show that for the different dialects and different tone ,the identification rate is not the same when the same characteristic parameter is chosen .