An implementation of short-timed speech recognition on layered neural nets

Static models in the form of state transition probability matrices representing short speech units such as syllables which correspond to Chinese utterance of isolated characters are adopted as learning patterns of layered Neural Nets, called MLP (multilayer perceptron). One of the major problems in neural net implementation of speech recognition is how an associative network usually with a fixed architecture, such as a number of input-output units and internal representation units, can deal with the sequential nature of speech. A static model is proposed as the input of a multilayer perceptron machine; a network architecture and learning algorithm are also introduced to implement isolated word recognition on MLP