Coarticuration of a sequence of vowels is normalized by a linear model of the coarticulation. An auditory response of V_0 in V V_0 V type connected vowels is modeled as a linear function of the acoustical features of the preceeding V, V_0, and the following V. The coefficients of the function are determined by the method of least squares using training data. The error rate of the recognition test using the model is lower by 7.5% than that of the test not using the model.
展开▼