A novel pattern learning and recognition procedure applied to the learning of vowels

The ability of a set of simple predicates to capture characteristic patterns in a parametric representation of vowels in continuous speech was investigated with the aid of an efficient conjunctive pattern recognition and classification system. The results compare favourably with those produced by a cluster-based minimal Euclidean distance technique, run over the identical training and test samples. The predicates used are similar to auditory receptive fields. I General Introduction One of the most challenging problems in the construction of Speech Understanding Systems* is that of finding an inexpensive but accurate characterization of phones in terms of the initial, parametric, representation of the speech wave. A good labeller would reduce the amount of effort expended by the complex heuristic knowledge sources (such as those for syntax and semantics) when they are required to reduce the uncertainty due to poorly performing bottom-up phone and word recognizers. Regrettably, variability in the realization of phones in continuous speech militates against any simple-minded approach to labelling and no method has yet been forthcoming which is both accurate and inexpensive at run-time. It is our belief, TThis research was supported in part by the Defense Advanced Research Projects Agency under contract no. F44620-73-C-0074 and monitored by the Air Force Office of Scientific Research. 2 Burge & Hayes-Roth