Temporally Dynamic Neural Networks for Speaker Independent Phoneme Recognition

Abstract

An alternative view of neural network based phoneme recognition based on multiresolution signal processing which incorporates noncausal context is suggested. Also some suggestions are made regarding target and error weight functions to improve performance and simplify training. Based on these observations a temporally dynamic neural network with self recurrent links of different delays is suggested and tested on speaker independent unvoiced plosives, (p,t,k), recognition task with input feature vectors derived from an auditory model.