论文部分内容阅读
We present a multi-target approach in network training for isolated word recognition. In this approach, each word was designated to produce output at more than one-target in the output laycr of the network during training.The specification of the target vector was mode based on the statistical or cluster structure of training patterns. Experiments consisted of classification of simulated data and isolated spoken words indicated that the multi-target approach improved the overall performance of the network.
We present a multi-target approach in network training for isolated word recognition. In this approach, each word was designated to produce output at more than one-target in the output laycr of the network during training. Specification of the target vector was mode based on the statistical or cluster structure of training patterns. Experiments consisted of classification of simulated data and isolated spoken words indicated that the multi-target approach improved the overall performance of the network.