L.P.C. speech recognition with random reference correlation

A correlation estimation algorithm, based on random reference correlation (r.r.c.), is applied to the problem of speech recognition. The inherent ill-conditioning of the estimated autocorrelation matrix is dealt with by pre-emphasis and ridge regression. Distance measures are defined and used to characterise the performance of the recognition. Results with the ten-digit dictionary show that the new method is comparable with the conventional one in terms of recognition performance but much more efficient in terms of implementation.