Neural Networks Based Phonetic Recognition of Isolated Chinese Phrase

The technique of phonetic recognition of isolated Chinese phrase is studied. By combining the time alignment algorithm with neural network technique, a phonetic recognition system based on mixed cascade neural networks is established. The system is composed of two different neural networks. The former is a time alignment network used to solve the time alignment problem and to extract fixed dimension feature vectors from input speech signal. And the latter is a BP network used as the neural network classifier. In the experiment of phonetic recognition for isolated Chinese phrase, the correct recognition rate of this method is up to 98.25%. The experimental results demonstrate that the method can not only solve commendably the time alignment problem by using a neural network but also improve obviously recognition performance. Compared with other ANN based phonetic recognition systems by using the linear time alignment method, both the correct recognition rate and the training speed of the proposed method are improved.