Fixed-length segment coding of LSF parameters

Evgeni Yakhnich, Yuval Bistritz

Research output: Contribution to conferencePaperpeer-review


This paper presents a method to attain very low bit-rate compression of speech spectral envelope. It is based on fixed-length segment coding. The method utilizes Temporal Decomposition (TD) technique for the compact representation of segments of Line Spectrum Frequencies (LSF) vector followed by split matrix quantization. The TD technique is modified to fit fixed-length segment coding. Computation is reduced by using fixed event functions and because event positions determination requires only a simple search within the short fixed-length segment. Weighted Euclidian distance is used as cost function to better approximate the spectral distance measure. The method achieves low bit rates without significant increase in computation cost. The method has been implemented on coding the spectral envelope for the MELP coder and showed its viability.

Original languageEnglish
Number of pages4
StatePublished - 2002
Event7th International Conference on Spoken Language Processing, ICSLP 2002 - Denver, United States
Duration: 16 Sep 200220 Sep 2002


Conference7th International Conference on Spoken Language Processing, ICSLP 2002
Country/TerritoryUnited States


Dive into the research topics of 'Fixed-length segment coding of LSF parameters'. Together they form a unique fingerprint.

Cite this