Understanding speech recognition using correlation-generated neural network targets

摘要: Training neural networks with variable targets for speech recognition systems has been shown to be effective in improving word accuracy. In this correspondence, a new and simple method estimating given training pattern is presented. It uses estimated correlations between different output nodes of network create set each pattern. Experimental results show that the error reduced by more than 20% when these correlation-based are compared conventional zero/one squared-error cost function. Performance approaches high-performance hidden Markov model (HMM) recognizers but requires far fewer parameters.