基于音素语义基的语音语义知识库构建方法  

A Semantic Knowledge Base Construction Method from Speech Phonetic Semantic Bases

作　　者：陈思旭 刘辰尧 郭杰洁 张艺檬 王凤玉 许文俊[1,3] CHEN Sixu;LIU Chenyao;GUO Jiejie;ZHANG Yimeng;WANG Fengyu;XU Wenjun(The State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China;School ofArtificial Intelligence,Beijing University of Posts and Telecommunications,Beijing 100876,China;Peng Cheng Laboratory,Shenzhen 518066,China)

机构地区：[1]北京邮电大学网络与交换技术全国重点实验室,北京100876 [2]北京邮电大学人工智能学院,北京100876 [3]鹏城实验室,广东深圳518066

出　　处：《移动通信》2024年第2期117-122,134,共7页Mobile Communications

基　　金：国家自然科学基金“语义通信性能评估方法与验证系统”(62293485);国家自然科学基金“基于流式数据处理的语义高效传输方法研究”(62301069)。

摘　　要：语义通信以保证信息含义的成功传递为目的,被视为下一代通信的潜在技术之一。然而,当前面向语音语义通信系统的语义知识库研究尚不成熟。为此,提出了一种基于音素的语音语义基模型,并基于双向长短期记忆网络提取音素语义基。进一步,通过将音素语义基与知识图谱的文本知识结构结合,构建了一种集语音含义及语音特性为一体的语音语义知识库框架,该框架生成具有知识结构的音素语义基,以此作为发送方和接收方之间共享的背景知识,支撑语音语义信息的高效、简约表征。仿真结果表明,相较于经典的语音语义通信系统,该框架在不同信噪比(SNR)下的语音质量(PESQ)和可懂度(STOI)都有显著提高,其中,在信噪比相同的条件下,基于所提语义知识库的语音语义通信系统在PESQ方面性能提升了8%,有效验证了所构建语义知识库的有效性。Aiming to ensure the successful transmission of the information meaning,semantic communication is envisioned as one of the potential technologies for the next-generation communication.However,the semantic knowledge base for speech semantic communication systems has not been well investigated.In this paper,a phoneme-based semantic base(Seb)model is proposed,and a bidirectional long short-term memory network is adopted to extract the phoneme Seb.Furthermore,by combining the proposed phoneme-based Seb with the text knowledge structure of the knowledge graph,a phoneme-based semantic knowledge base framework is constructed with the integration of phonetic meaning and phonetic characteristics.The proposed framework generates the phoneme-based Sebs with knowledge structure as the shared background knowledge between the transmitter and the receiver,facilitating efficient and concise representation of phonetic semantic information.Simulation results show that the proposed framework outperforms the classic speech semantic communications in terms of perceptual evaluation of speech quality(PESQ)and short-time objective intelligibility(STOI)under different signal-to-noise ratios(SNR).Under the same SNR condition,the performance of the speech semantic communication system based on the proposed semantic knowledge base is improved by 8%in terms of PESQ,which effectively verified the effectiveness of the proposed semanticknowledge base.

关 键 词：语义基 语义知识库 语音传输 

分 类 号：TN912.31[电子电信—通信与信息系统]