CODE-EXCIDE LINEAR PREDICTION VOCODER WITH SPLIT CODEBOOK

• Main Authors: Chou Chin- Ting, 周錦廷
• Other Authors: Ching-Kung Lee
• Format: Others
• Language: zh-TW
• Published: 1998
• Online Access: http://ndltd.ncl.edu.tw/handle/70298990025438266061

碩士 === 大同工學院 === 電機工程研究所 === 86 === The fundamental limitations of effectively in the vector quantization (VQ) system are computational complexity and memory requirements. Among many codebook search methods, the conventional method of full search will spend a lot of computation in searching the optimal codeword. One approach to reducing storage and computation requirements is to improve the search method or change it's structure of encoder. Code-excited linear prediction (CELP) with split codebook and partial distance search (PDS) method are Split codebook and CELP coding are combined to form an efficient low-bit rate speech coding system. The resulting system uses an excitation sequence splitted into shorter vector in order to selecting the synthesis filter excitation sequence rapidly. When taking the mean square error (MSE) every time, another fast search method, PDS algorithm first mentioned by Bei and Gray at 1985, will be uesd.Simulations are performed for encoding rates between 4.8kbps ~ 16kbps. Informal listening tests indicated that this new structure is about equally speech quality compared to that of conventional CELP. Memory decreases a lot and the computational complexity will reduce to about 1/32.