语音编码算法G.729研究及其DSP实现.pdf

摘要
摘要
现代语音通信希望占用频带窄,语音质量高,这就需要一个低速率、高质量的语音
编码方案。国际电信联盟(ITU)1996年公布了一种共轭结构代数码激励线性预测(CS-
ACELP)的8kb/。,高语音质量的优
点,但复杂度高,实时实现困难。随着DSP(Digital Signal Processor)技术高速发展,芯
片计算能力不断提高,。本文的主要工
。
本文首先简单介绍了语音压缩编码技术,
基本理论。主要有语音信号产生数字模型、线性预测分析模型、Levinson递推算法、矢
量量化和感知加权滤波器。
,以及实现算法的硬件开发平台TMS320
C5416DSK 和TMS320C5509EVM ,对硬件平台上要用到的硬件资源也作了介绍。然
后介绍了软件的设计,主要包括硬件初始化,软件初始化,编码解码以及调用中断,通
过使用DSP/BIOS建立配置文件,结合DMA(直接存储器访问)和McBSP(多通道缓
冲串口)进行语音数据传输。程序的优化是工作的重点,采用汇编器优化,C语言级,
汇编级和算法级优化,主要方法包括开环基音搜索改进,固定码本搜索改进,MA(滑
动平均)预测器选择改进,循环程序改进,软件流水,使用内联函数,基本运算函数改
进等。实际测试结果表明,,而输出合成语音仍然
保持了很高的质量,。
最后对所做的工作进行了总结,并给出了下一步改进的目标。
关键词:; CS-ACELP; DSP;算法优化;固定码本搜索; 自适应码本搜索.
– I –
万方数据
Abstract
Abstract
The narrow bandwidth and good speech quality are all required in modern speech
requirement needs a codec with low rate and high quality. ITU
published algorithm in algorithm is based on Conjugate Structure
Algebraic Code Excited Linear Prediction(CS-ACELP) at 8kb/s. It has the advantage
of low delay and high speech quality,but algorithm is plex and difficult
to realize in real-time. With the rapid development of the DSP technology, -
putational capability of the DSP is improved greatly. Now a single DSP chip has the
capability to realize plex algorithm in real- to optimize
algorithm and realize it on DSP in real-time is the key work in this paper.
Firstly this paper gives a brief introduction of the speech coding, then introduces
some basic theories used in the algorithm, including the digital model of the
production of speech signal, the Linear Prediction Coding model, Vector Quantization,
Levinson algorithm and Perceptual Weighted Filtering.
Secondly this paper discusses some important details of algorithm, then
introdu