基于DBF的汉语方言自动辨识

在汉语方言辨识中,传统的声学特征是语音信号的谱特征的参数化表示,常常包含说话人、信道、背景噪声等冗余信息,针对上述问题将深度神经网络(Deep Neural Network,DNN)引入特征提取之中,提出了与音素层面相关的深度瓶颈特征(Deep Bottleneck Feature,DBF),尝试从特征层面抑制方言冗余信息的影响.最后在实验部分对瓶颈层的位置,节点数目进行了讨论,结果显示,深度瓶颈特征相对于传统声学特征能够取得更高的识别率.

Based on DBF of Chinese dialects automatic identification

In Chinese dialects identification, the traditional acoustic features are the parametric representation of spectral features of speech signals, and often include redundant information such as speaker, channel and background noise. To solve these problems, the deep neural network (DNN) is introduced into feature extraction, deep bottleneck feature(DBF), which is related to the phoneme level, tries to suppress the influence of dialect redundant information from the feature level. Finally, the position and the number of nodes of the bottleneck layer are discussed in the experimental part. The results show that the deep bottleneck feature can achieve higher recognition rate than the traditional acoustic feature.