A New Deterministic Codebook Structure for Celp Speech Coding

Abstract

Low bit rate, high quality speech coding is a vital part in voice telecommunication systems. The introduction of CELP (1984, Codebook Excited Linear Prediction) speech coding provides a feasible way to compress speech data to 4.8 kbps with high quality. However, the formidable computational complexity required for real-time processing has prevented its wide application. Using our codebook, we reduce the computational complexity of codebook search, which originally accounts for 2/3 of the computational complexity, to almost nothing; while preserving the same good speech quality. This tremendous reduction in computational complexity is achieved by replacing the traditional stochastic codebook by an artificially constructed deterministic codebook. After a careful study of the minimization of vector quantization distortions, we found that although randomness is usually observed in speech residuals; it is not necessary to use a noise-like stochastic codebook to encode the speech residuals. As long as the code vectors were distributed uniformly over a sphere, very small VQ errors can be achieved. The most significant advantage of using this deterministic codebook is extremely fast codebook search. After this reduction, we have an algorithm about 5 MIPS. It can be handled by even inexpensive DSP chips, while maintaining the same high quality. Besides extremely simple encoding and decoding schemes, this codebook also provides optimal error tolerance property and it doesn't require codebook storage. We hope our contribution can finally make CELP speech coding a widely applicable technology.