Continuous speech recognition based on a self-learning neuro-fuzzy network

This paper presents a hierarchical neural networks system to continuous speech recognition. The system formulates the speech recognition as two-phase procedure. In the first phase, the hyperrectangular composite neural networks (HRCNNs) is proposed to classify the speech signal into three different types. These are silence, consonant, and vowel. In the following phase, the HRCNNs are utilized to recognize each consonant and vowel. The novel class of HRCNNs is used to classify frames. The HRCNNs integrate the rule-based approach and neural network paradigms, therefore, this special hybrid system may neutralize the disadvantages of each alternative. The values of the parameters of trained HRCNNs can be utilized to extract both crisp and fuzzy classification rules. In our experiments, continuous reading-rate of Mandarin balanced sentences were utilizing to illustrate, the performance of the proposed speech recognition system. The effectiveness of the proposed system is confirmed by the experimental results.