A Finite State Vector Quantizer for New Bit Rate Speech Compression

This paper presents an enhancement technique to compress speech signal based on LPC-10 model. Line spectrum pairs (LSP), as alternate representation of LPC presents some useful properties for quantisation and transmission, is used to model the speech signal in each frame. Artificial Neural Networks, based on a finite state vector quantizer technique, is investigates the characteristic of LSP coefficients to reduce the bit rate of the system. This technique shows an efficiency reduction of bit rate. The results of this experiments show that this technique can reduce the bit rate approximately 5.5 -37% and also provides the excellent formant frequency similar to the original speech.