Abstract: A kind of model combining the BP neural network with the Adaboost is proposed to identify isolated words of Hunan dialect speaker-independently in this paper. In order to reflect the dynamic properties of dialects and the characteristics of vocal tract, LPCC, MFCC and their  first-order differential coefficients are combined together as dialects characteristic coefficients. Multiple BP neural networks are used as weak classifiers for dialect initial identification, and then a strong classifier is constructed from these weak classifiers based on Adaboost iteration algorithm to obtain the final identification results. The experimental results show that this hybrid model has stronger robustness and higher recognition rate than the pure BP neural network under relatively low signal to noise ratio.

Key words: dialects identification, mixed characteristic parameters, auto-adapted Boosting, Back Propagation（BP） neural network

摘要： 着眼于非特定人孤立词湖南地区的方言辨识，提出一种将BP神经网络和Adaboost算法相结合的辨识模型。为反映方言的动态特性及其声道特性，采用LPCC、MFCC和各自一阶差分系数相组合作为方言特征系数。利用多个BP神经网络作为弱分类器对方言进行初步辨识，借助Adaboost迭代算法将这些弱分类器组合起来构成强分类器，得出最终辨识结果。实验证明，该混合模型较单纯的BP神经网络具有更强的噪声鲁棒性和较高的识别率。

关键词: 方言辨识, 混合特征参数, 自适应Boosting, 反向传播（BP）神经网络