基于改进的长短期神经网络的贵州方言辨识系统的设计与实现

汉语方言的辨识能为案件侦破提供重要的线索,为了对贵州方言进行辨识,本文设计并实现了贵州方言辨识系统,该系统采用Client/Server与Browser/Server相结合的架构,其用户端采用Matlab实现并具有改进的长短期记忆神经网络算法,主要用于方言的辨识和方言语音样本的采集。方言样本采集于贵州省6个地区,首先提取语音样本与口头禅的梅尔频率倒谱系数MFCC,然后每份语音样本MFCC后面加上相应地区的口头禅MFCC,最后通过奇异值分解得到该系统的输入数据。该系统的网站主要用于训练数据的储存与修改,采用ASP.NET技术并利用C#、JavaScript和T-SQL等编程语言实现。实验结果证明贵州方言辨识系统是高效的,让用户获得极大的方便和客观统一的方言辨识结果。

Design and Build of identification system of Guizhou dialect

Chinese dialect identification may provide important clues for forensic investigation. This paper constructs an identification system of Guizhou dialect, which combines Client/Server and Browser/Server architecture. The user side of the system is mainly used for identification of Guizhou dialect based on improved Long Short-Term Memory (LSTM) and dialect voice samples collection, which is achieved by Matlab. Firstly, the authors extracted Mel frequency cepstral coefficients (MFCC) from speech samples and regional pet phrase collected from six regions in Guizhou province, then added corresponding regional pet phrase after each voice samples. Finally, the singular value was decomposed from voice samples as the input data of the identification system of Guizhou dialect. The website of the system is mainly used for storage and modification of training data, which is achieved by programming languages of C#, JavaScript and T-SQL based on ASP.NET. The experimental results show that the identification system of Guizhou dialect is efficient for identification of Guizhou dialect, which enables users to achieve great convenience and accurate, objective and unified identification results of Guizhou dialect.