On segmentation and recognition of connected digits based on neural network model

Summary form only given, as follows. An automatic segmentation and recognition system based on a neural network model is proposed, and a backpropagation learning algorithm is employed to establish the networks. The main new idea for segmentation is to classify the energy, spectrum, and pitch-period transitions that occur at the boundaries between syllables. These feature transitions are used as the input patterns of the neural network . The segmented syllables are then used as the basic units for recognition by being fed them into the well-trained network. Ten digits (0-9) and syllables spoken in Mandarin are used in the speaker-independent phase for segmentation experiments, and only ten digits are used in the speaker-dependent phase for recognition experiments. With an average speaking rate of 150 digits per minute, a segmentation accuracy with a coincidence rate of 95.7% and recognition rate of 97.2% can be achieved.