In　recent　years,　deep　neural　network　(DNN)　has　been　widely　used　for　monaural　speech　enhancement　due　to　its　good　performance　for　learning　higher-level　information.　In　this　paper,　an　approach　of　speech　enhancement　with　binaural　cues　derived　from　DNN　is　proposed.　A　deep-learning-based　model　is　investigated　to　learn　a　mapping　function　between　the　pre-enhanced　cue　and　clean　cue,　which　are　extracted　from　the　pre-enhanced　speech　and　clean　speech,　respectively.　The　proposed　method　contains　two　stages:　Offline　training　stage　and　online　enhancing　stage.　At　offline　training　stage,　a　stacked　auto-encoder　(SAE)　model,　a　type　of　deep　neural　network,　is　used　to　learn　the　mapping　function.　At　online　stage,　the　clean　cue　is　estimated　by　the　learned　mapping　function　online　first.　Then,　the　noisy　speech　can　be　enhanced　with　the　estimated　clean　cue.　Compared　to　the　reference　methods,　the　experimental　results　yield　significant　improvements　for　three　objective　measurements.　©　2017　IEEE.