Synthesis and performance analysis of a Recurrent Fuzzy Multilayer Perceptron for speech recognition

A novel speech recognition method has been proposed which combines the capabilities of a Recurrent Fuzzy Multilayer Perceptron (MLP) to the existing Mel Frequency Cepstral Coefficients (MFCC) model, synthesized using JAVA. Performance analysis of the proposed recurrent fuzzy MLP relative to a speech recognition system has been shown using MATLAB. Owing to its short-term memory effect in addition to inherent neuro-fuzzy model advantages, the simulation results obtained were significantly better than offered by a crisp neural network.