Speech compression using tree structured Vector Quantization

The science of obtaining a compact representation of a signal while maintaining all the necessary information is known as Compression which can be basically classified in two types, Lossless and Lossy compression. Lossy compression can be further classified in two types, namely Scalar Quantization (SQ) and Vector Quantization (VQ). SQ involves processing the input samples individually using some distortion measure while VQ involves processing the input samples in groups into a set of well-defined vectors using some defined distortion measure. VQ since about 1980 became a popular technique for source coding of image and speech data [1]. The direct use of VQ suffers from a serious complexity barrier. The classical technique of Tree Structured Vector Quantization was introduced by Buzo et al. [2]. This paper explains the TSVQ design approach for speech compression with compact codebook.