Recognition of ambiguous vowels by generation of dynamical coherence in oscillator neural networks

Abstract

The authors have reported previously that a model of oscillator neural networks shows high ability in speaker-independent vowel recognition. In the present paper, the authors elucidate how ambiguous vowels are recognized in the model. First the authors demonstrate that their computer experiments on the identification of synthetic vowels well agree with psychological experiments. Second, it is shown that vowels with contextual variation are reliably recognized by the addition of feedforward of constraints to the authors' model. It is concluded that the principle of generation of dynamical coherence in auditory information systems gives flexible ability in the recognition of vowels consistently with psychological observation.