Research On Chinese Dialect Discrimination Based On Distinguishing Features

Dialect is a special linguistic variant with valuable historical and linguistic research value.Chinese dialects,as the outstanding intangible cultural heritage of the Chinese nation,should not disappear with the increasing popularity of Mandarin.The identification of Chinese dialect types is a key step of advancing the intelligent processing of Chinese dialects,and has important practical significance in the protection and inheritance of dialects.At present,the existing dialect identification models mainly focus on how to extract text-level features such as effective morphology and syntax in dialect texts.Studies on distinguishing dialect words and dialect pronunciation features are relatively rare.Considering that current deep learning methods can extract effective features from the supervised information,this provides a feasible method for extracting distinguishing features of dialects.Based on these observations,the paper proposes two different and effective dialect distinguishing features for dialect language identification:(1)Dialect distinguishing features,this paper first builds a Chinese dialect text classification model using the attention mechanism,then selects the representative distinguishing dialect words by using attention weights,and then integrates the word vector features and underlying acoustic features of the distinguishing dialect words.Finally,distinguishing dialect features are extracted from the fused features to identify the types of Chinese dialects.The experimental results on the benchmark Chinese dialect corpus show that the fusion feature is better than using only dialect text or underlying acoustic features.(2)Dialect pronunciation features,this paper first constructs a Chinese phoneme recognition model,then uses the identified phoneme sequence,a kind of supervised information,to extract dialect pronunciation features from the underlying acoustic features,and finally extracts distinguishing dialect features from dialect pronunciation features to discriminate dialect types.The experimental results on the benchmark Chinese dialect corpus show that dialect pronunciation features play an important role in dialect recognition.Finally,this article integrates the above work,and designs a client-server architecture-based intelligent speech processing platform,which can effectively perform dialect type recognition and Gan dialect speech recognition,while incorporating functions such as voice intelligent chat.In general,this article has conducted in-depth research on key technologies in Chinese dialect identification,proposed some solutions to related problems,and designed corresponding algorithms and experiments.Experiments on the benchmark Chinese dialect corpus show that these methods proposed in this paper help to improve the performance of Chinese dialect recognition,while reducing dependence on largescale corpora,and provide effective references for similar studies.In addition,this article is not only helpful for exploring the ways and models of traditional cultural research and protection,but also for the subsequent development and use of sound resources in Jiangxi Province.