Endangered Tujia Language Speech Enhancement Research Based on Improved DCGAN

来源 :第十八届中国计算语言学大会暨中国中文信息学会2019学术年会 | 被引量 : 0次 | 上传用户:wilson_rui
下载到本地 , 更方便阅读
声明 : 本文档内容版权归属内容提供方 , 如果您对本文有版权争议 , 可与客服联系进行内容授权或下架
论文部分内容阅读
  As an endangered language,Tujia language only rely on oral communication.There must exist noises in the process of collecting Tujia language corpus.This paper studies an end-to-end speech enhancement model based on improved deep convolutional generative adversarial network(DCGAN)to extract nearly pure Tujia language speech in noisy environment.Due to the low resource nature of Tujia language,using Chinese corpus as an extension of the Tujia language can effectively solve the problem of insufficient data.The speech enhancement function of the Tujia language was realized using the end-to-end method that consists of symmetric encoding and decoding.By modifying the loss function and network hierarchy parameters,adding the spectrum normalization and imbalanced learning rate made the model more stable during the training process.The experimental results show that the speech enhancement method proposed in this paper can achieve better noise reduction effect on the Tujia language dataset than traditional speech enhancement algorithm and neural network enhancement algorithms.
其他文献
In the e-commerce websites,such as Taobao and Amazon,interactive question-answering(QA)style reviews usually carry rich aspect information of products.To well automatically analyze the aspect informat
Natural Language Inference(NLI),which is also known as Recognizing Textual Entailment(RTE),aims to identify the logical relationship between a premise and a hypothesis.In this paper,a DCAE(Directly-Co
The neural components in deep learning framework are crucial for the performance of many natural language processing tasks.So far there is no systematic work to investigate the influence of neural com
Legal Cause Prediction(LCP)aims to determine the charges in criminal cases or types of disputes in civil cases according to the fact descriptions.The research to date takes LCP as a text classificatio
会议
Natural language inference(NLI)aims to predict whether a premise sentence can infer another hypothesis sentence.Models based on tree structures have shown promising results on this task,but the perfor
We present a Chinese judicial reading comprehension(CJRC)dataset which contains approximately 10K documents and almost 50K questions with answers.The documents come from judgment documents and the que
会议
Native ad is an important kind of online advertising which has similar form with the other content in the same platform.Compared with search ad,predicting the click-through rate(CTR)of native ad is mo
学位
学位
学位