G.718宽带语音和音频编解码算法的研究及实现

发布时间：2018-04-10 22:50

本文选题：嵌入式语音编码 + G.718　；参考：《安徽工业大学》2016年硕士论文

【摘要】：随着移动通信技术和网络通信技术的不断提高,当前语音压缩编码技术正朝着宽带、变速率和嵌入式的方向发展。ITU-T于2008年6月制定了全新的宽带变速率嵌入式语音和音频编解码标准G.718,该算法迎合了当前语音编码技术的发展趋势,采用了多项新技术,具有结构灵活,合成语音质量高,能够将有线网络通信系统和无线网络通信系统互联互通等优点,在音/视频会议、3G、4G、多媒体流等场合具有良好的应用前景。本课题在深入理解G.718编解码原理的基础上,讨论了G.718算法L1~L5层的结构特点。分析了G.718核心层L1层以及ACELP增强层L2层使用的预处理、信号分类、ACELP等关键技术,比较了G.718同其他宽带语音算法核心层编码的异同点。研究了G.718中L3、L4、L5层使用的变换域编码技术,针对全频带信号输入下,对语音特性信号采用的代数矢量量化技术以及对音频特性信号采用的频带选择波形增益编码技术和阶乘脉冲编码技术。介绍了G.718变换域编码使用的MDCT技术,引入了一种基于FFT的MDCT快速算法。由于G.718算法对不同类型输入语音采用不同的编码模式,导致其算法非常复杂,而且ITU-T提供的C源代码主要侧重于严谨地描述算法,使得算法执行效率差,无法实时实现。本课题通过解读C源代码中各函数的功能、各模块之间的关系及各模块的复杂度分布,提出了基于DSP全汇编实现G.718语音编解码器总体开发方案,结合DSP C55x汇编语言特点,制定了各模块编写方案。在维持C源代码总体框架不变的前提下,编写了G.718编解码器各函数模块的全汇编程序。为了提升全汇编代码的执行效率,采用循环体优化、并行指令等汇编指令优化方法对编写的全汇编代码进行优化。为了测试全汇编优化后G.718编解码器的还原语音质量,本课题基于TMS320VC5505 EVM设计了G.718的实时测试系统,经过WB_PESQ客观评估方法测试,G.718编解码器还原出的语音质量达到实时通信要求。
[Abstract]:With the continuous improvement of mobile communication technology and network communication technology, the current speech compression coding technology is moving towards broadband,In June 2008, ITU-T formulated a new broadband variable rate embedded speech and audio coding and decoding standard G.718. this algorithm caters to the development trend of current speech coding technology and adopts many new technologies.It has the advantages of flexible structure, high synthetic voice quality, and the ability to connect wired network communication system with wireless network communication system, etc. It has a good application prospect in audio / video conference, multimedia stream and so on.Based on the deep understanding of G. 718 encoding and decoding principle, this paper discusses the structure characteristics of G.718 algorithm L1~L5 layer.The key technologies such as pretreatment and signal classification in L1 layer of G.718 core layer and L2 layer of ACELP enhancement layer are analyzed. The similarities and differences between G.718 and other wideband speech algorithms are compared.In this paper, the transform domain coding technique used in L3 / L4 / L5 layer in G.718 is studied.The algebraic vector quantization technique for speech characteristic signal, the gain coding technique for frequency band selection waveform and the factorial pulse coding technique for audio characteristic signal are adopted.This paper introduces the MDCT technology used in G.718 transform domain coding, and introduces a fast MDCT algorithm based on FFT.Because G.718 adopts different coding modes for different types of input speech, the algorithm is very complex, and the C source code provided by ITU-T mainly focuses on describing the algorithm strictly, which makes the algorithm execution inefficient and can not be implemented in real time.By interpreting the function of each function in C source code, the relationship between each module and the complexity distribution of each module, the overall development scheme of G.718 speech codec based on DSP is put forward, and the characteristics of DSP C55x assembly language are combined.The program of each module is worked out.On the premise of maintaining the whole frame of C source code, the whole assembly program of each function module of G.718 codec is written.In order to improve the execution efficiency of full assembly code, the optimization methods such as loop volume optimization, parallel instruction and so on are used to optimize the whole assembly code.In order to test the restored speech quality of G.718 codec after full assembly optimization, a real-time test system of G.718 is designed based on TMS320VC5505 EVM. The quality of speech restored by G.718 codec is tested by WB_PESQ objective evaluation method to meet the requirements of real-time communication.
【学位授予单位】：安徽工业大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TN912.3

【相似文献】