一种语音信源的语义压缩编码方法  

A Semantic Compression Coding Method for Speech Sources

作　　者：郭杰洁 刘辰尧 张艺檬 许文俊 别志松[3] GUO Jiejie;LIU Chenyao;ZHANG Yimeng;XU Wenjun;BIE Zhisong(School of Information and Communication Engineering,Beijing University of Posts and Telecommunications,Beijing 100876,China;The State Key Laboratory of Networking and Switching Technology,Beijing University of Posts and Telecommunications,Beijing 100876,China;School of Artificial Intelligence,Beijing University of Posts and Telecommunications,Beijing 100876,China)

机构地区：[1]北京邮电大学信息与通信工程学院,北京100876 [2]北京邮电大学网络与交换技术国家重点实验室,北京100876 [3]北京邮电大学人工智能学院,北京100876

出　　处：《移动通信》2023年第4期71-76,共6页Mobile Communications

基　　金：国家自然科学基金(62293485)。

摘　　要：语义压缩编码方法通过开辟信息表征新维度,能够显著提高压缩效率,因而受到广泛关注。旨在通过语义表征与编码实现语音信源的高效压缩:首先,面向语音信源内容、韵律、音调、音色四方面特性,提出多层语义表征框架,实现语音语义的综合表征;基于此框架,利用矢量量化方法与哈夫曼编码方法,构建语音语义知识库,进一步提高语音压缩效率。仿真结果表明,相较于经典低比特率语音压缩方法,所提方法能够显著提高信源压缩效率。Semantic compression methods have raised wide attention due to the superiority in improving the compression efficiency,which is achieved by opening up new dimensions of information representation.The paper aims at developing an efficient compression algorithm for speech sources through semantic representation and coding.Specifically,the characteristics of speech,including content,rhythm,pitch,and timbre,are jointly considered to construct a multi-layer semantic representation framework,facilitating comprehensive semantic representation of speech.Based on this framework,a knowledge base of speech semantics is constructed using vector quantization and Huffman coding methods to further improve the compression efficiency.Simulation results show that the proposed method significantly improves the source compression efficiency compared with the conventional low-bit-rate speech compression methods.

关 键 词：语义压缩 语音编码 语义知识库 深度神经网络 

分 类 号：TN912.31[电子电信—通信与信息系统]