Speaker adaptive vector quantisation of LPC parameters of speech

Recently, it has been shown by Schroeder and Atal (see Proc. of ICASSP-85, p.937-40, 1985) that good quality speech at rates as low as 6 kbit/s can be achieved with CELP and its derivatives. However, in bringing down the bit rate even further these coding schemes have resorted to allocating fewer bits to the quantisation of the LPC parameters. It is known that CELP and its derivatives are sensitive to LPC parameters quantisation errors, and recently various schemes have been proposed to overcome this degradation. The authors describe how speaker adaptive vector quantisation (SAVQ) can be applied to the quantisation of LPC parameters and assess its performance when incorporated into a low bit rate coding scheme of 4.8 kbit/s