Third International Conference on Spoken Language Processing (ICSLP 94)

Yokohama, Japan
September 18-22, 1994

Complexity Reduction Methods for Vector Sum Excited Linear Prediction Coding

Sung Joo Kim, Seung Jong Park, Yung Hwan Oh

Department of Computer Science, Korea Advanced Institute of Science and Technology, Taejon, Korea

The vector sum excited linear prediction (VSELP) coding gives high quality of synthetic speech at bit rates as low as 4.8kbps, but its computational complexity is prohibitive for real-time applications. In this paper, we propose three methods to reduce the computations in VSELP coding. First, we use the overlapped sparse codebook for the basis vectors. Second, we introduce the preprocessing step to the stochastic codebook search procedure. It decides some combination coefficients of basis vectors using heuristics so that the search space decreases. Third, some candidates are preselected before the adaptive codebook search procedure by comparing them with the ideal excitation sequence. We develop a 4.8kbps coder using all the proposed methods and perform the quality test. It has been shown that the proposed coder retains good quality of synthetic speech and it is more than twice as fast as the original coder.

Full Paper

Bibliographic reference.  Kim, Sung Joo / Park, Seung Jong / Oh, Yung Hwan (1994): "Complexity reduction methods for vector sum excited linear prediction coding", In ICSLP-1994, 2071-2074.