SPEECH RECOGNITION WITH HIERARCHICAL RECURRENT NEURAL NETWORKS

A hierarchical recurrent neural network (HRNN)for speech recognition i s presented. The HRNN is trained by a generalized probabilistic descen t (GPD) algorithm. Consequently, the difficulty of empirically selecti ng an appropriate target function for training RNNs can be avoided. Re sults obtained in this study indicate the proposed HRNN has the advant ages of being capable of absorbing the temporal variation of speech pa tterns as well as possessing effective discrimination capabilities. Th e scaling problem of RNNs is also greatly reduced. Additionally, a rea lization of the system using initial/final sub-syllable models for iso lated Mandarin syllable recognition is also undertaken for verifying i ts effectiveness. The effectiveness of the proposed HRNN is confirmed by the experimental results.