Sixth European Conference on Speech Communication and Technology
In this paper, the use of Karhunen-Loeve transform (KLT) and discrete cosine transform (DCT) is studied for encoding of the line spectrum frequency (LSF) parameters at variable bit rate (VBR). For VBR coding, scalar quantization (SQ) is used with Huffman coding. The basic idea in developing these schemes is using linear transform to exploit the strong intraframe and interframe correlation of LSF parameters to increase the performance of the SQ. First, several adaptive KLT schemes are developed to increase the efficiency of the KLT. To reduce the interframe correlation efficiently, two-dimensional transforms and moving average prediction of the transformed coefficients are also investigated. It is shown that these schemes introduce as good as or better performance in the examined bit rates compared to other methods in the field of LSF coding.
Full Paper (PDF) Gnu-Zipped Postscript
Bibliographic reference. Lois, László (1999): "Variable length coding of transformed LSF coefficients", In EUROSPEECH'99, 1455-1458.