Flexible vowel recognition by the generation of dynamic coherence in oscillator neural networks: speaker-independent vowel recognition

摘要: We propose a new model for speaker-independent vowel recognition which uses the flexibility of dynamic linking that results from synchronization oscillating neural units. The system consists an input layer and three layers, are referred to as A-, B- C-centers. signals time series linear prediction (LPC) spectrum envelopes auditory signals. At each time-window within series, A-center receives extracts local peaks envelope, i.e., formants, encodes them into groups independent oscillations. Speaker-independent characteristics embedded connection matrix in B-center according statistical data Japanese vowels. associative interaction reciprocal between A- B-centers selectively activate global synchronized pattern over two centers. C-center evaluates activities among formant regions give selective output category five Thus, flexible ability dynamical features is achieved capability present was investigated demonstrated remarkable vowels very similar human listeners, including misleading In addition, it showed stable unsteady robustness against background noise. optimum condition frequency oscillation discussed comparison with stimulus-dependent synchronizations observed neurophysiological experiments cortex.