Fuzzy neural network for phoneme sequence recognition

In this paper, we present a novel speech recognition system based on the use of the fuzzy neural network for 2D phoneme sequence pattern recognition. The self-organizing map and then learning vector quantization are used to organize the phoneme feature vectors of short and long phonemes segmented from speech samples to obtain their phoneme maps. The 2D phoneme response sequences of the speech samples are formed optimally on the phoneme maps by the Viterbi search algorithm. These 2D phoneme response sequence curves are used as inputs to the fuzzy neural network for training and recognition of speech utterances. Simulations indicate up to 91.7% accuracy on 0-9 digit-voice recognition can be obtained.