Vector quantization for LSF coding and codebook storage

This paper presents an effective vector quantizer for very low bit rate speech coding. Based on vector quantization for line spectral frequency (LSF) parameters, it reduces the storage and searching complexity of codebooks. In this paper, the mean of LSF parameters is removed and then differential split vector quantization method is applied for vector quantization. To reduce the storage and searching complexity of codebooks largely, the storage structure of code word is optimized and code precision is lowed. Experimental results demonstrate that the average spectral distortion is less than 1dB, and spectral leakage is not more than 4dB. The quantizer meets the basic requirements of transparent quantization well.