Date of Award

Spring 1996

Document Type

Thesis

Degree Name

Master of Science in Electrical Engineering - (M.S.)

Department

Electrical and Computer Engineering

First Advisor

Ali N. Akansu

Second Advisor

Yair Shoham

Third Advisor

Nirwan Ansari

Abstract

Speech coding has been of interest to communication specialists for years. A number of technologies for describing and transmitting the speech spectral envelope have been studied. Among them, the Vector Quantization (VQ) of Line Spectral Frequencies (LSF) is receiving more attention because of its good rate-distortion performance. Typically, about 30 bits are assigned to code 10-dimensional LSF vectors with a resulting spectral distortion of less than ldB. However, 30-bit full-search VQ is totally impractical in terms of both computational complexity and memory space. Various standard suboptimal, low-complexity VQ techniques have been used for coding the LSF's in the literature. The most commonly used method is the split VQ, where the 10-dimensional LSF vector is typically partitioned into three sub-vectors of sizes 3, 3 and 4. Each sub-vector is then independently coded. This method reduces the complexity and required memory space significantly. However, the price paid is a compromise in performance. Reduced performance is inherent in most low-complexity VQ systems.

In this thesis we propose a simple fast-search VQ of the LSF's to be used on top of the split VQ (i.e., in each of the sub-vector domains). The main trait of the proposed method is that no suboptimal codebooks are used and there is no further reduction in performance. In each sub-vector domain, a full-size optimally trained codebook, typically of size 1024, is searched using a fast-search algorithm. The result of this search is identical to that of a full search, yet, only about 25% of full-search complexity is needed.

Share

COinS