A spectral prediction model is studied in the context of discriminating between the stop consonants by place of articulation. We study the predictor in both scalar and vector modes and show the sensitivity of both these models to sequence direction. The paper highlights the role of coherence in regulating the directionality dependence of performance. The predictor model is discussed from the multivariate autoregressive process point of view to throw some light on the observed interesting behaviour of the model.
Bibliographic reference. Rao, P. V. S. / Raveendran, R. (1995): "On the dual role of sequence directionality and coherence in a spectral predictive discrimination model", In EUROSPEECH-1995, 1377-1380.