To　enhance　speech　recognition　performance　with　limited　network　capacity,　an　end-To-end　speech　recognition　algorithm　based　on　multi-layer-enriched　supervision　(MES-ASR)　is　proposed　in　this　paper.　This　approach　allows　the　model　to　learn　more　speaking　content　information　despite　the　capacity　constraints.　Specifically,　the　intermediate　layer　supervision　loss　is　optimized　by　introducing　an　additional　supervision　loss　at　that　layer.　This　forces　the　lower　layers　of　the　network　to　focus　on　acquiring　more　speaking　content　information,　leading　to　improved　recognition　performance.　Notably,　this　method　adds　the　supervisory　loss　solely　in　the　middle　layer,　ensuring　simplicity　during　training　and　no　additional　memory　or　computational　overhead　during　testing.　The　proposed　approach　is　evaluated　on　the　Aishell-1　dataset,　employing　WeNet　as　the　baseline　model.　The　experimental　results　demonstrate　a　reduction　in　character　error　rate　and　an　enhancement　in　speech　recognition　performance.　©　2023　IEEE.