Fractal modelling of residues in linear predictive coding of speech

This paper describes a novel approach of fractal modelling and coding of residuals for excitation in the linear predictive coding of speech. This work was motivated by reducing the bit rate to 1200 bps, while maintaining a good quality of speech. Linear prediction based speech coders differ primarily in the modelling of the residual. The design trade-off in the modelling of the residual is between quality and bit-rate. In this paper fractal modelling is used to model the residual. We show that fractal modelling reduces the bit-rate while maintaining quality. A 6 kbps speech coder was implemented using the piecewise self-affine fractal model. The new coder has a signal-to-noise ratio of 10.9 dB. An informal subjective measure found the perceptual quality to be comparable to that of the 13 kbps GSM coder.