基于小波去噪与语音分离的同声传译系统设计  

Design of Simultaneous Interpretation System Based on Wavelet Denoising and Speech Separation

作　　者：王萌[1] WANG Meng(Xian yang Normal University,Xianyang Shanxi 712000,China)

机构地区：[1]咸阳师范学院,陕西咸阳712000

出　　处：《自动化与仪器仪表》2025年第1期300-303,308,共5页Automation & Instrumentation

基　　金：陕西省教育厅项目《14世纪英国女性与文学研究》(09JK278)。

摘　　要：同声传译作为一项涉及语言翻译和语音识别的复杂任务,被英语翻译行业广泛应用。但该模型在实现过程中面临着实时性差、翻译准确率不高和难以处理文化差异和特定领域知识等缺陷。因此研究尝试基于语音分离技术来实现同声传译模型,并基于深度学习算法提高模型的翻译准确率。同时研究引入小波去噪算法对原始信号进行去噪处理,使模型能更准确地提取语义,提高模型翻译的准确率。实验结果表明,提出基于小波去噪与语音分离的同声传译模型在中英文数据集上的翻译时延分别为0.29 s和0.32 s,因此该模型具有较好的实时翻译能力,同时提出模型英译中的平均准确率为88.96%;中译英的平均准确率为90.47%,均高于对照模型的翻译准确率,因此证明提出模型具有一定先进性。Simultaneous interpretation,as a complex task involving language translation and speech recognition,is widely used in the English translation industry.However,the model faces shortcomings such as poor real-time performance,low translation accuracy,and difficulty in handling cultural differences and specific domain knowledge during implementation.Therefore,the research attempts to implement a simultaneous interpretation model based on speech separation technology,and improve the translation accuracy of the model based on deep learning algorithms.At the same time,the study introduces wavelet denoising algorithms to denoise the original signal,so that the model can more accurately extract semantics and improve the accuracy of model translation.The experimental results show that the proposed simultaneous interpretation model based on wavelet denoising and speech separation has translation delays of 0.29s and 0.32s on Chinese and English datasets,respectively.Therefore,the model has good real-time translation ability,and the average accuracy of the proposed model in English translation is 88.96%;The average accuracy rate of Chinese to English translation is 90.47%,which is higher than that of the control model,so it proves that the proposed model is progressiveness.

关 键 词：小波去噪 语音分离 同声传译系统 深度学习 

分 类 号：TN929[电子电信—通信与信息系统] TP39[电子电信—信息与通信工程]