Signal classification in digital telephone networks

This paper considers the problem of classifying voice-band data, facsimile, and speech signals, using discriminant functions that are extensions of those described by Benvenuto (see IEEE Trans. on Commun., vol.41, no.4, p.539, 1993). Both the central second-order moment of the rectified signal, and the low order lags of the estimated autocorrelation sequence are computed. These discriminants are used to perform classification in 128 ms. with 0% misclassification for most signal types considered. Algorithm enhancements improve classification performance, reduce computational complexity, and extend the voice-band data subclassification abilities.