Voice recognition network, method and equipment based on cross-layer connection attention and medium

The invention is suitable for the technical field of speech recognition, and provides a speech recognition network, method and device based on cross-layer connection attention and a storage medium, the speech recognition network is constructed based on a Transform encoder-decoder structure, a Transform encoder of the speech recognition network comprises a plurality of encoding layers, a cross-layer connection module is connected between the adjacent encoding layers, and the cross-layer connection module is connected between the adjacent encoding layers. The coding layer is used for learning information of a middle attention map of a previous coding layer through a cross-layer connection module when the attention map is generated, so that the attention map generated by each coding layer can more accurately express a dependency relationship of a context; therefore, the speech recognition accuracy of the speech recognition network is remarkably improved under the condition that almost neglectable parameter quant