Research on Power Dispatching Speech Recognition Based on Improved Transformer

Now, speech recognition [1] has developed rapidly. Transformer network model architecture is constructed by encoder and decoder. has become popular and applied to speech recognition. The Transformer model has achieved very good results in ordinary daily scenes. In the production environment of power dispatching, speech recognition requires low error rate and faster recognition speed. Therefore, some improvements have been made to the Transformer model. Then, the core attention mechanism in Transformer is mainly introduced. According to the characteristics of voice dataset in power dispatching, the model is improved and the attention mechanism is lightweight. Finally, through experimental training, we use the improved model to train on the power dispatching voice dataset, and then evaluate the trained model.