CELP算法在语音信号处理的应用精选.doc

CELP算法在语音信号处理的应用 摘要 语音信号处理属于信息科学的一个重要分支，随着通信技术的发展以及互联网语音实时传输技术的迅速发展的需要，为了使同样的信道容量传输更多路的语音信号和节省存储空间,语音编码技术取得了很大的进展。其中线性预测编码（CELP）是中、低速率下比较成功的编码算法。 CELP算法采用分帧技术进行编码，帧长一般为30ms，并将每一语音帧分为4个子帧，把在每个子帧内搜索到的最佳的码矢量作为激励信号。CELP使用的是FS-1016标准，该标准基于合成分析（ABS）、感知加权矢量量化（VQ）和线性预测（LP）等技术。它采用10阶线性预测滤波器来模拟语音信号的短时频谱或声道特性，并且采用了两种码本：自适应码本和随机码本。自适应码本用来模拟长时信号或基音的周期激励；固定的随机码本用来逼近经过了矢量量化的短时线性预测和长时预测后的预测误差。译码端合成部分的激励信号由自适应码本和随机码本经过增益加权后得到，感知加权滤波利用人耳的掩蔽效应来提高语音的质量。 研究表明，基于FS-1016的CELP编码器能较好地还原语音的时间波形包络，但合成语音的波形图与原始语音波形图有一定的差别，也就是说存在一定的失真现象，所以合成语音质量只能达到通信质量。合成语音在共振峰和基音周期结构上与原始语音比较一致，他们反映了语音信号的主要特征，所以合成语音具有比较良好的可懂度。 关键字：CELP算法；自适应码本；固定码本 Abstract Speech Signal Processing information is an important branch of information science.With the needs of the development of communication technology and the rapid development of the Internet voice transmission in real-time technology, in order to make the same channel capacity of the voice signal path transmite more and save storage space, considerable progress of voice coding technology has been made. Among those the linear predictive coding (CELP) is a successful low-rate coding algorithm.  CELP algorithm uses sub-frame coding,the length of a frame is usually 30ms, and each voice frame is divided into 4 sub-frames, regarding the best code vector signal searched in each sub-frame as an incentive. CELP using the FS-1016 standard, the standard bases on synthetic analysis(ABS), perceptual weighted vector quantization (VQ) and linear prediction (LP) techniques. It uses 10-order linear prediction filter to simulate the short-time spectrum or channel characteristics of speech signal, and it uses two types of code: adaptive codebook and random codebook. Adaptive code book is used to simulate the long signal or cycle incentive of pitch; fixed random code is used to close the short-term linear prediction after vector quantization of and the forecast error after long-term forecast .The synthesized excitation signal of decoding part gains from the adaptive codeb